pugixml.cpp

浏览该文件的文档.

/**
 * pugixml parser - version 1.9
 * --------------------------------------------------------
 * Copyright (C) 2006-2018, by Arseny Kapoulkine (arseny.kapoulkine@gmail.com)
 * Report bugs and download new versions at http://pugixml.org/
 *
 * This library is distributed under the MIT License. See notice at the end
 * of this file.
 *
 * This work is based on the pugxml parser, which is:
 * Copyright (C) 2003, by Kristen Wegner (kristen@tima.net)
 */
 
#ifndef SOURCE_PUGIXML_CPP
#define SOURCE_PUGIXML_CPP
 
#include "pugixml.hpp"
 
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <assert.h>
#include <limits.h>
 
#ifdef PUGIXML_WCHAR_MODE
#   include <wchar.h>
#endif
 
#ifndef PUGIXML_NO_XPATH
#   include <math.h>
#   include <float.h>
#endif
 
#ifndef PUGIXML_NO_STL
#   include <istream>
#   include <ostream>
#   include <string>
#endif
 
// For placement new
#include <new>
 
#if defined(__clang__)
# pragma clang diagnostic push
# pragma clang diagnostic ignored "-Wconversion"
# pragma clang diagnostic ignored "-Wdouble-promotion"
#endif
 
// 如果代码是在Microsoft Visual C++编译器下编译的
#ifdef _MSC_VER
    // 保存当前的警告状态，以便稍后恢复
    #pragma warning(push)
    
    // 禁用警告4127：条件表达式是常量。这通常发生在if语句的条件始终为真或始终为假时。
    #pragma warning(disable: 4127)
    
    // 禁用警告4324：结构体由于使用了__declspec(align())进行了填充。这通常是为了满足特定的对齐要求。
    #pragma warning(disable: 4324)
    
    // 禁用警告4702：不可达代码。这通常发生在代码路径在逻辑上不可能被执行到。
    #pragma warning(disable: 4702)
    
    // 禁用警告4996：这个函数或变量可能不安全。这通常用于标记那些被认为不安全的函数，如strcpy()等。
    #pragma warning(disable: 4996)
#endif
 
// 如果代码是在Microsoft Visual C++编译器下编译，并且同时使用了Clang编译器前端（如Clang-Cl）
#if defined(_MSC_VER) && defined(__clang__)
    // 保存当前的Clang诊断状态，以便稍后恢复
    #pragma clang diagnostic push
    
    // 忽略警告-Wdeprecated：这个函数或变量可能不安全。这与4996警告类似，但由Clang发出。
    #pragma clang diagnostic ignored "-Wdeprecated"
#endif
 
// 如果代码是在Intel C++编译器下编译的
#ifdef __INTEL_COMPILER
    // 禁用警告177：函数已被声明，但从未被引用。
    #pragma warning(disable: 177)
    
    // 禁用警告279：控制表达式是常量。这与4127警告类似。
    #pragma warning(disable: 279)
    
    // 禁用警告1478和1786：函数被声明为‘弃用’。这表示函数或变量在未来的版本中可能会被移除。
    #pragma warning(disable: 1478 1786)
    
    // 禁用警告1684：从指针转换为同等大小的整型类型。这通常发生在指针和整型大小相同但类型不同的转换中。
    #pragma warning(disable: 1684)
#endif
 
// 如果代码是在Borland C++编译器下编译，并且定义了PUGIXML_HEADER_ONLY宏
#if defined(__BORLANDC__) && defined(PUGIXML_HEADER_ONLY)
    // 禁用警告8080：符号已被声明但从未被使用。这通常发生在定义了但未使用的变量或函数上。
    #pragma warn -8080
#endif
 
#if defined(__BORLANDC__) && defined(PUGIXML_HEADER_ONLY)
#   pragma warn -8080 //符号已被声明但从未被使用；即便在入栈 / 出栈括号内禁用（相关设置），该警告也不会消失
#endif
 
#ifdef __BORLANDC__
#   pragma option push
#   pragma warn -8008 //条件始终为假
#   pragma warn -8066 //不可达代码
#endif
 
#ifdef __SNC__
// 由于编译器的一个漏洞，使用 diag_push 和 diag_pop 无法禁用模板内部的警告。
#   pragma diag_suppress=178 // 函数已声明但从未被引用
#   pragma diag_suppress=237 // 控制表达式是常量
#endif
 
#ifdef __TI_COMPILER_VERSION__
#   pragma diag_suppress 179 // 函数已被声明，但从未被引用
#endif
 
//内联控制
#if defined(_MSC_VER) && _MSC_VER >= 1300
#   define PUGI__NO_INLINE __declspec(noinline)
#elif defined(__GNUC__)
#   define PUGI__NO_INLINE __attribute__((noinline))
#else
#   define PUGI__NO_INLINE
#endif
 
//分支权重控制
#if defined(__GNUC__) && !defined(__c2__)
#   define PUGI__UNLIKELY(cond) __builtin_expect(cond, 0)
#else
#   define PUGI__UNLIKELY(cond) (cond)
#endif
 
//简单静态断言
#define PUGI__STATIC_ASSERT(cond) { static const char condition_failed[(cond) ? 1 : -1] = {0}; (void)condition_failed[0]; }
 
// Digital Mars C++ bug workaround for passing char loaded from memory via stack
#ifdef __DMC__
#   define PUGI__DMC_VOLATILE volatile
#else
#   define PUGI__DMC_VOLATILE
#endif
 
// Integer sanitizer workaround; we only apply this for clang since gcc8 has no_sanitize but not unsigned-integer-overflow and produces "attribute directive ignored" warnings
#if defined(__clang__) && defined(__has_attribute)
#   if __has_attribute(no_sanitize)
#       define PUGI__UNSIGNED_OVERFLOW __attribute__((no_sanitize("unsigned-integer-overflow")))
#   else
#       define PUGI__UNSIGNED_OVERFLOW
#   endif
#else
#   define PUGI__UNSIGNED_OVERFLOW
#endif
 
// Borland C++ bug workaround for not defining ::memcpy depending on header include order (can't always use std::memcpy because some compilers don't have it at all)
#if defined(__BORLANDC__) && !defined(__MEM_H_USING_LIST)
using std::memcpy;
using std::memmove;
using std::memset;
#endif
 
// Some MinGW/GCC versions have headers that erroneously omit LLONG_MIN/LLONG_MAX/ULLONG_MAX definitions from limits.h in some configurations
#if defined(PUGIXML_HAS_LONG_LONG) && defined(__GNUC__) && !defined(LLONG_MAX) && !defined(LLONG_MIN) && !defined(ULLONG_MAX)
#   define LLONG_MIN (-LLONG_MAX - 1LL)
#   define LLONG_MAX __LONG_LONG_MAX__
#   define ULLONG_MAX (LLONG_MAX * 2ULL + 1ULL)
#endif
 
// In some environments MSVC is a compiler but the CRT lacks certain MSVC-specific features
#if defined(_MSC_VER) && !defined(__S3E__)
#   define PUGI__MSVC_CRT_VERSION _MSC_VER
#endif
 
// Not all platforms have snprintf; we define a wrapper that uses snprintf if possible. This only works with buffers with a known size.
#if __cplusplus >= 201103
#   define PUGI__SNPRINTF(buf, ...) snprintf(buf, sizeof(buf), __VA_ARGS__)
#elif defined(PUGI__MSVC_CRT_VERSION) && PUGI__MSVC_CRT_VERSION >= 1400
#   define PUGI__SNPRINTF(buf, ...) _snprintf_s(buf, _countof(buf), _TRUNCATE, __VA_ARGS__)
#else
#   define PUGI__SNPRINTF sprintf
#endif
 
// We put implementation details into an anonymous namespace in source mode, but have to keep it in non-anonymous namespace in header-only mode to prevent binary bloat.
#ifdef PUGIXML_HEADER_ONLY
#   define PUGI__NS_BEGIN namespace pugi { namespace impl {
#   define PUGI__NS_END } }
#   define PUGI__FN inline
#   define PUGI__FN_NO_INLINE inline
#else
#   if defined(_MSC_VER) && _MSC_VER < 1300 // MSVC6 seems to have an amusing bug with anonymous namespaces inside namespaces
#       define PUGI__NS_BEGIN namespace pugi { namespace impl {
#       define PUGI__NS_END } }
#   else
#       define PUGI__NS_BEGIN namespace pugi { namespace impl { namespace {
#       define PUGI__NS_END } } }
#   endif
#   define PUGI__FN
#   define PUGI__FN_NO_INLINE PUGI__NO_INLINE
#endif
 
// uintptr_t
#if (defined(_MSC_VER) && _MSC_VER < 1600) || (defined(__BORLANDC__) && __BORLANDC__ < 0x561)
namespace pugi
{
#   ifndef _UINTPTR_T_DEFINED
    typedef size_t uintptr_t;
#   endif
 
    typedef unsigned __int8 uint8_t;
    typedef unsigned __int16 uint16_t;
    typedef unsigned __int32 uint32_t;
}
#else
#   include <stdint.h>
#endif
 
// Memory allocation
// PUGI__NS_BEGIN 和 PUGI__NS_END 宏通常用于定义命名空间的开始和结束
PUGI__NS_BEGIN
// 定义一个函数，用于分配指定大小的内存。这个函数简单地调用了标准库的 malloc 函数
PUGI__FN void* default_allocate(size_t size)
{
    return malloc(size);
}
// 定义一个函数，用于释放之前分配的内存。这个函数简单地调用了标准库的 free 函数
PUGI__FN void default_deallocate(void* ptr)
{
    free(ptr);
}
// 定义一个模板结构体，用于存储内存管理函数（分配和释放）。
template <typename T>
struct xml_memory_management_function_storage
{
    static allocation_function allocate;
    static deallocation_function deallocate;
};
// 为模板结构体的静态成员变量分配默认值。  
// 这里使用了模板特化的语法来指定每个类型T的 allocate 和 deallocate 成员变量的值
// Global allocation functions are stored in class statics so that in header mode linker deduplicates them
// Without a template<> we'll get multiple definitions of the same static
template <typename T> allocation_function xml_memory_management_function_storage<T>::allocate = default_allocate;
template <typename T> deallocation_function xml_memory_management_function_storage<T>::deallocate = default_deallocate;
// 使用typedef为 xml_memory_management_function_storage<int> 创建一个别名 xml_memory
// 专门用于int类型的内存管理函数存储
typedef xml_memory_management_function_storage<int> xml_memory;
PUGI__NS_END
 
// String utilities
PUGI__NS_BEGIN
// Get string length
// 获取字符串长度  
// char_t 是一个类型别名，它根据是否定义了 PUGIXML_WCHAR_MODE 宏来决定是 char 还是 wchar_t 类型
PUGI__FN size_t strlength(const char_t* s)
{
    // 断言 s 不为 nullptr，确保传入的字符串指针是有效的
    assert(s);
    // 根据是否定义了 PUGIXML_WCHAR_MODE 宏来选择使用 wcslen 还是 strlen 函数来获取字符串长度
#ifdef PUGIXML_WCHAR_MODE
    return wcslen(s);// 宽字符字符串长度
#else
    return strlen(s);// 单字节字符字符串长度
#endif
}
 
// Compare two strings
PUGI__FN bool strequal(const char_t* src, const char_t* dst)
{
    // 断言 src 和 dst 都不为 nullptr，确保传入的字符串指针都是有效的
    assert(src && dst);
 
#ifdef PUGIXML_WCHAR_MODE
    return wcscmp(src, dst) == 0;// 宽字符字符串比较
#else
    return strcmp(src, dst) == 0;// 单字节字符字符串比较
#endif
    }
 
// Compare lhs with [rhs_begin, rhs_end)
PUGI__FN bool strequalrange(const char_t* lhs, const char_t* rhs, size_t count)
{
    // 逐个字符比较 lhs 和 rhs，如果发现有不相等的字符，则返回 false
    for (size_t i = 0; i < count; ++i)
        if (lhs[i] != rhs[i])
            return false;
    // 如果循环结束后没有发现不相等的字符，则检查 lhs 的第 count 个字符是否为空终止符，  
    // 如果是，则返回 true，表示字符串相等（在指定范围内）；否则返回 false
    return lhs[count] == 0;
}
 
// Get length of wide string, even if CRT lacks wide character support
PUGI__FN size_t strlength_wide(const wchar_t* s)
{
    // 断言 s 不为 nullptr，确保传入的宽字符串指针是有效的
    assert(s);
    // 如果定义了 PUGIXML_WCHAR_MODE 宏，则直接使用 wcslen 函数获取长度
#ifdef PUGIXML_WCHAR_MODE
    return wcslen(s);
#else
    // 如果没有定义 PUGIXML_WCHAR_MODE 宏，则手动计算长度。  
    // 通过遍历字符串直到遇到空终止符来计算长度
    const wchar_t* end = s;
    while (*end) end++;
    return static_cast<size_t>(end - s);// 返回字符串的长度
#endif
}
PUGI__NS_END
 
// auto_ptr-like object for exception recovery
PUGI__NS_BEGIN
// 定义一个模板结构体 auto_deleter，它可以用于任何类型 T
template <typename T> struct auto_deleter
{
    // 定义一个类型别名 D，它是一个指向函数的指针，该函数接受一个指向 T 类型的指针作为参数，并返回 void
    typedef void (*D)(T*);
    // 成员变量：
    // data：指向需要被自动删除的对象的指针。
    // deleter：一个函数指针，指向用于删除 data 指向的对象的函数
    T* data;
    D deleter;
    // 构造函数，接受一个指向对象的指针和一个删除该对象的函数指针
    auto_deleter(T* data_, D deleter_) : data(data_), deleter(deleter_)
    {
    }
    // 析构函数，当 auto_deleter 对象被销毁时调用。
    // 如果 data 非空，则调用 deleter 函数来删除 data 指向的对象
    ~auto_deleter()
    {
        if (data) deleter(data);
    }
    // release 函数，用于手动释放对对象的所有权。
    // 它返回 data 指向的对象的指针，并将 data 设置为 0（nullptr）。
    // 这意味着之后 auto_deleter 对象不会再尝试删除该对象
    T* release()
    {
        T* result = data;
        data = 0;
        return result;
    }
};
PUGI__NS_END
 
#ifdef PUGIXML_COMPACT
PUGI__NS_BEGIN
    // 定义了一个名为 compact_hash_table 的类，用于实现紧凑的哈希表
    class compact_hash_table
    {
    public:
        // 类的构造函数，初始化哈希表的项指针为0，容量和计数也为0
        compact_hash_table(): _items(0), _capacity(0), _count(0)
        {
        }
        // 清除哈希表的方法。释放已分配的内存，并将容量和计数重置为0
        void clear()
        {
            // 如果已分配了内存给哈希表的项
            if (_items)
            {
                // 使用 xml_memory::deallocate 方法释放内存
                xml_memory::deallocate(_items);
                // 将项指针、容量和计数重置为0
                _items = 0;
                _capacity = 0;
                _count = 0;
            }
        }
        // 根据给定的键查找值的方法
        void* find(const void* key)
        {
            // 如果哈希表的容量为0，表示哈希表为空，直接返回0（表示未找到
            if (_capacity == 0) return 0;
            // 调用 get_item 方法根据键获取对应的项（这个方法在给出的代码中没有定义，可能是类的私有方法）
            item_t* item = get_item(key);
            // 使用 assert 断言确保获取的项不为空，且项的键要么与给定的键相等，要么是一个空项（键和值都为0）
            assert(item);
            assert(item->key == key || (item->key == 0 && item->value == 0));
            // 返回找到的项的值
            return item->value;
        }
        // 插入一个键值对到哈希表中
        void insert(const void* key, void* value)
        {
            // 断言哈希表的容量不为0，并且当前存储的项的数量小于容量的3/4（为了保持哈希表的负载因子在一个合理的范围内）
            assert(_capacity != 0 && _count < _capacity - _capacity / 4);
            // 调用get_item方法根据键获取哈希表中的项（这个方法在给出的代码中没有定义，可能是类的私有方法）
            item_t* item = get_item(key);
            // 断言确保获取的项不为空
            assert(item);
            // 如果找到的项的键为空，表示这是一个空槽，可以插入新的键值对
            if (item->key == 0)
            {
                // 增加哈希表中当前存储的项的数量
                _count++;
                // 将键设置为给定的键
                item->key = key;
            }
            // 无论是否是新插入的项，都将值设置为给定的值（如果键已经存在，则更新值）
            item->value = value;
        }
        // 为哈希表预留额外的空间（默认是16个单位）
        bool reserve(size_t extra = 16)
        {
            // 如果加上额外空间后的项的数量大于等于当前容量的3/4，则需要重新哈希以扩展容量
            if (_count + extra >= _capacity - _capacity / 4)
                // 调用rehash方法重新分配内存并重新插入所有项（这个方法在给出的代码中没有定义，可能是类的私有方法）
                // 如果rehash方法返回true，表示重新哈希成功；否则，表示失败（尽管在这个实现中它总是返回true或false，但具体取决于rehash的实现）
                return rehash(_count + extra);
                return rehash(_count + extra);
            // 如果不需要重新哈希，则直接返回true
            return true;
        }
    // 私有成员定义开始
    private:
        // 定义一个结构体item_t，用于存储哈希表中的项
        struct item_t
        {
            const void* key;// 指向键的指针，使用void*表示可以是任意类型的键
            void* value; // 指向值的指针，使用void*表示可以是任意类型的值
        };
        // 指向item_t数组的指针，用于存储哈希表中的所有项
        item_t* _items;
        // 哈希表的容量，即_items数组可以存储的项的最大数量
        size_t _capacity;
        // 当前哈希表中存储的项的数量
        size_t _count;
        // 一个成员函数，用于在需要时重新调整哈希表的大小
        bool rehash(size_t count);
        // 一个成员函数，用于根据给定的键查找对应的项
        item_t* get_item(const void* key)
        {
            // 确保传入的键不为空
            assert(key);
            // 确保哈希表的容量大于0
            assert(_capacity > 0);
            // 计算哈希表容量的掩码，用于将哈希值限制在哈希表的范围内
            size_t hashmod = _capacity - 1;
            // 计算给定键的初始桶（bucket）位置
            size_t bucket = hash(key) & hashmod;
            // 使用二次探测法解决哈希冲突
            for (size_t probe = 0; probe <= hashmod; ++probe)
            {
                // 引用当前桶中的项
                item_t& probe_item = _items[bucket];
                // 如果找到了键匹配的项，或者该位置为空（表示哈希表中没有更多的项），则返回该项的指针
                if (probe_item.key == key || probe_item.key == 0)
                    return &probe_item;
 
                // hash collision, quadratic probing
                // 计算下一个探测的位置，使用二次探测法
                bucket = (bucket + probe + 1) & hashmod;
            }
            // 断言失败，表示哈希表已满（理论上应该不可达，因为哈希表在满之前会进行扩容）
            assert(false && "Hash table is full"); // unreachable
            // 如果上面的断言失败，则返回0（实际上由于断言的存在，这行代码不会被执行）
            return 0;
        }
        // 定义一个静态函数，用于计算给定键的哈希值，并处理可能的无符号整数溢出
        static PUGI__UNSIGNED_OVERFLOW unsigned int hash(const void* key)
        {
            // 将指针转换为无符号整数类型（uintptr_t），然后转换为unsigned int。
            // 这里的转换是为了获取一个可以用于哈希计算的数值。
            unsigned int h = static_cast<unsigned int>(reinterpret_cast<uintptr_t>(key));
 
            // MurmurHash3 32-bit finalizer
            h ^= h >> 16;       // 将h右移16位后与自身异或
            h *= 0x85ebca6bu;   // 乘以一个魔术数字（常数）
            h ^= h >> 13;       // 再次右移13位后与自身异或
            h *= 0xc2b2ae35u;   // 再次乘以另一个魔术数字
            h ^= h >> 16;       // 最后，再次右移16位后与自身异或
 
            return h;
        }
    };
    // 定义一个函数，用于根据当前存储的元素数量重新调整哈希表的容量
    PUGI__FN_NO_INLINE bool compact_hash_table::rehash(size_t count)
    {
        // 初始化容量为32
        size_t capacity = 32;
        // 当当前元素数量大于等于当前容量的3/4时，将容量翻倍，直到满足条件
        while (count >= capacity - capacity / 4)
            capacity *= 2;
        // 创建一个新的哈希表实例
        compact_hash_table rt;
        // 设置新哈希表的容量为计算得到的新容量
        rt._capacity = capacity;
        // 为新哈希表分配内存空间
        rt._items = static_cast<item_t*>(xml_memory::allocate(sizeof(item_t) * capacity));
        // 如果内存分配失败，则返回false
        if (!rt._items)
            return false;
        // 将新分配的内存空间初始化为0
        memset(rt._items, 0, sizeof(item_t) * capacity);
        // 遍历旧哈希表的每一项，如果键不为空，则将其插入到新哈希表中
        for (size_t i = 0; i < _capacity; ++i)
            if (_items[i].key)
                rt.insert(_items[i].key, _items[i].value);
        // 如果旧哈希表有已分配的内存，则释放它
        if (_items)
            xml_memory::deallocate(_items);
        // 更新旧哈希表的容量为新的容量，并设置其项指针为新分配的内存
        _capacity = capacity;
        _items = rt._items;
        // 断言确保重新哈希后元素数量未变。
        assert(_count == rt._count);
        // 返回true，表示重新哈希成功
        return true;
    }
 
// PUGI__NS_END 和 PUGI__NS_BEGIN 是命名空间结束和开始的宏，用于创建一个命名空间。
// 这样可以避免全局命名冲突，并组织代码结构。
PUGI__NS_END
#endif
 
PUGI__NS_BEGIN
#ifdef PUGIXML_COMPACT
    // 在紧凑模式下，内存块对齐的字节数为4。
    static const uintptr_t xml_memory_block_alignment = 4;
#else
    // 在非紧凑模式下，内存块对齐的字节数为指针大小，通常是4字节或8字节，取决于平台。
    static const uintptr_t xml_memory_block_alignment = sizeof(void*);
#endif
 
    // 以下是一些额外的元数据位掩码，用于标记XML页面的特定属性。
    static const uintptr_t xml_memory_page_contents_shared_mask = 64;  // 内容是否共享
    static const uintptr_t xml_memory_page_name_allocated_mask = 32;   // 名称是否已分配
    static const uintptr_t xml_memory_page_value_allocated_mask = 16;   // 值是否已分配
    static const uintptr_t xml_memory_page_type_mask = 15;             // 节点类型掩码
 
    // 组合掩码，用于字符串的唯一性检查。
    static const uintptr_t xml_memory_page_name_allocated_or_shared_mask = xml_memory_page_name_allocated_mask | xml_memory_page_contents_shared_mask;
    static const uintptr_t xml_memory_page_value_allocated_or_shared_mask = xml_memory_page_value_allocated_mask | xml_memory_page_contents_shared_mask;
 
#ifdef PUGIXML_COMPACT
    #define PUGI__GETHEADER_IMPL(object, page, flags)// 在紧凑模式下，以下宏定义为空，因为它们不被使用。 
    #define PUGI__GETPAGE_IMPL(header) (header).get_page()
#else
    // 在非紧凑模式下，定义了如何从对象获取其头部信息和页面信息的宏。
    // PUGI__GETHEADER_IMPL 宏用于计算对象头部信息，包括对象相对于页面的位置和标志。
    #define PUGI__GETHEADER_IMPL(object, page, flags) (((reinterpret_cast<char*>(object) - reinterpret_cast<char*>(page)) << 8) | (flags))
    // PUGI__GETPAGE_IMPL 宏用于从头部信息中获取页面指针。
    // 这个宏通过void*类型转换指针，以避免增加目标类型的对齐要求的警告。
    #define PUGI__GETPAGE_IMPL(header) static_cast<impl::xml_memory_page*>(const_cast<void*>(static_cast<const void*>(reinterpret_cast<const char*>(&header) - (header >> 8))))
#endif
 
    // PUGI__GETPAGE 宏用于从节点中获取页面。
    #define PUGI__GETPAGE(n) PUGI__GETPAGE_IMPL((n)->header)
    // PUGI__NODETYPE 宏用于从节点中获取节点类型。
    #define PUGI__NODETYPE(n) static_cast<xml_node_type>((n)->header & impl::xml_memory_page_type_mask)
    struct xml_allocator;
    // 定义xml_memory_page结构体，用于管理内存页
    struct xml_memory_page
    {
        // 静态成员函数，用于在给定内存块上构造xml_memory_page对象
        // 这个函数不分配新的内存，而是将传入的void*内存块转换为xml_memory_page*并初始化它
        static xml_memory_page* construct(void* memory)
        {
            // 将传入的void*内存块转换为xml_memory_page*类型
            xml_memory_page* result = static_cast<xml_memory_page*>(memory);
            // 初始化成员变量
            result->allocator = 0;  // 指向xml_allocator的指针，初始化为0（空指针）
            result->prev = 0;       // 指向前一个内存页的指针，初始化为0（空指针）
            result->next = 0;       // 指向下一个内存页的指针，初始化为0（空指针）
            result->busy_size = 0;  // 当前页中已分配（忙碌）的内存大小，初始化为0
            result->freed_size = 0; // 当前页中已释放（空闲）的内存大小，初始化为0
            // 如果定义了PUGIXML_COMPACT宏，则初始化与紧凑模式相关的成员变量
        #ifdef PUGIXML_COMPACT
            result->compact_string_base = 0;
            result->compact_shared_parent = 0;
            result->compact_page_marker = 0;
        #endif
            // 返回构造并初始化后的xml_memory_page对象的指针
            return result;
        }
        // 成员变量
        xml_allocator* allocator;
        // 内存页之间的双向链表链接
        xml_memory_page* prev;
        xml_memory_page* next;
        // 内存使用情况统计
        size_t busy_size;
        size_t freed_size;
        // 注意：如果定义了PUGIXML_COMPACT宏，则还会存在以下成员变量
        // 这些变量在紧凑模式下用于优化内存使用和访问速度
    #ifdef PUGIXML_COMPACT
        char_t* compact_string_base;
        void* compact_shared_parent;
        uint32_t* compact_page_marker;
    #endif
    };
 
    static const size_t xml_memory_page_size =
    #ifdef PUGIXML_MEMORY_PAGE_SIZE
        (PUGIXML_MEMORY_PAGE_SIZE)
    #else
        32768
    #endif
        - sizeof(xml_memory_page);
    // 定义xml_memory_string_header结构体，用于描述字符串在内存页中的位置和大小
    struct xml_memory_string_header
    {
        // 从内存页数据起始位置到字符串起始位置的偏移量
        uint16_t page_offset; // offset from page->data
        // 如果字符串占据整个内存页，则为0；否则为字符串的实际大小
        uint16_t full_size; // 0 if string occupies whole page
    };
    // 定义xml_allocator结构体，用于管理内存页的分配
    struct xml_allocator
    {
        // 构造函数，接受一个指向根内存页的指针，并初始化成员变量
        xml_allocator(xml_memory_page* root): _root(root), _busy_size(root->busy_size)
        {
        #ifdef PUGIXML_COMPACT
            _hash = 0;// 如果定义了PUGIXML_COMPACT宏，则初始化_hash为0
        #endif
        }
        // 成员函数：分配一个新的内存页
        // 接受一个数据大小参数，用于确定除了结构体本身外还需要多少额外空间
        xml_memory_page* allocate_page(size_t data_size)
        {
            // 计算总大小：内存页结构体的大小加上额外数据的大小
            size_t size = sizeof(xml_memory_page) + data_size;
 
            // allocate block with some alignment, leaving memory for worst-case padding
            void* memory = xml_memory::allocate(size);
            if (!memory) return 0;// 如果分配失败，则返回0（空指针）
 
            // prepare page structure
            // 在分配的内存块上构造xml_memory_page对象
            xml_memory_page* page = xml_memory_page::construct(memory);
            assert(page);// 确保构造成功
            // 设置新页面的分配器为根页面的分配器（可能是为了保持一致性或实现某种内存管理策略）
            page->allocator = _root->allocator;
            // 返回新分配的页面
            return page;
        }
        // 定义deallocate_page函数，用于释放一个内存页
        // 接受一个指向xml_memory_page对象的指针作为参数
        static void deallocate_page(xml_memory_page* page)
        {
            // 调用xml_memory类的静态成员函数deallocate来释放内存页
            // 假设xml_memory是一个管理内存分配的类，类似于标准库中的allocator
            xml_memory::deallocate(page);
        }
        // 它用于在内存不足时分配内存，可能是通过分配一个新的内存页或其他机制
        void* allocate_memory_oob(size_t size, xml_memory_page*& out_page);
        // 定义allocate_memory函数，尝试在当前根内存页中分配指定大小的内存
        // 如果内存不足，则调用allocate_memory_oob函数
        void* allocate_memory(size_t size, xml_memory_page*& out_page)
        {
            // 使用PUGI__UNLIKELY宏来提示编译器这个条件可能不常发生
            // 这有助于优化生成的代码（尽管这取决于编译器的实现）
            if (PUGI__UNLIKELY(_busy_size + size > xml_memory_page_size))
                // 如果当前根内存页中的已分配大小加上请求的大小超过了内存页的总大小
                // 则调用allocate_memory_oob函数来分配内存，并传递请求的大小和out_page引用
                return allocate_memory_oob(size, out_page);
            // 计算分配内存的起始地址
            // _root是指向当前根内存页的指针，sizeof(xml_memory_page)是内存页结构体的大小
            // _busy_size是当前已分配（忙碌）的内存大小
            void* buf = reinterpret_cast<char*>(_root) + sizeof(xml_memory_page) + _busy_size;
            // 更新已分配（忙碌）的内存大小
            _busy_size += size;
            // 将out_page引用设置为指向当前根内存页的指针
            out_page = _root;
            // 返回分配的内存的起始地址
            return buf;
        }
 
    #ifdef PUGIXML_COMPACT
    // allocate_object 函数用于在紧凑模式下分配指定大小的对象，并返回指向分配的内存的指针。
    // 同时，它还更新 out_page 参数以指向包含该内存的对象页。
    void* allocate_object(size_t size, xml_memory_page*& out_page)
    {
        // 首先，调用 allocate_memory 函数分配比请求大小多出 uint32_t 大小的内存块。
        // 额外的空间用于存储一个标记，用于跟踪内存页的元数据。
        void* result = allocate_memory(size + sizeof(uint32_t), out_page);
        if (!result) return 0; // 如果内存分配失败，则返回 nullptr。
 
        // 计算 result 指针与页标记之间的偏移量。
        ptrdiff_t offset = static_cast<char*>(result) - reinterpret_cast<char*>(out_page->compact_page_marker);
 
        // 如果偏移量超过了一个阈值（意味着 result 距离页标记足够远），则需要插入一个新的标记。
        if (PUGI__UNLIKELY(static_cast<uintptr_t>(offset) >= 256 * xml_memory_block_alignment))
        {
            // 将 result 指针转换为 uint32_t 指针，并设置标记。
            uint32_t* marker = static_cast<uint32_t*>(result);
            *marker = static_cast<uint32_t>(reinterpret_cast<char*>(marker) - reinterpret_cast<char*>(out_page));
            out_page->compact_page_marker = marker; // 更新页的标记指针。
 
            // 因为我们不会重用页空间直到重新分配它，所以我们可以将标记块视为已释放。
            // 这将确保 deallocate_memory 正确地跟踪大小。
            out_page->freed_size += sizeof(uint32_t); // 更新已释放的大小。
 
            return marker + 1; // 返回指向标记后的对象内存的指针。
        }
        else
        {
            // 如果不需要插入新标记，则回滚 uint32_t 的部分，并更新 busy_size。
            _busy_size -= sizeof(uint32_t);
 
            return result; // 返回原始的内存分配指针。
        }
    }
    #else
        // 定义一个函数，用于分配指定大小的内存对象，并返回指向该对象的指针。
// 同时，通过out_page参数返回该对象所在的内存页。
        void* allocate_object(size_t size, xml_memory_page*& out_page)
        {
            // 调用另一个函数allocate_memory来实际进行内存分配，并返回分配的指针。
            return allocate_memory(size, out_page);
        }
#endif
        // 定义一个函数，用于释放之前分配的内存
        void deallocate_memory(void* ptr, size_t size, xml_memory_page* page)
        {
            // 如果当前释放的内存页是根页（_root），则更新根页的忙碌大小（_busy_size）。
            if (page == _root) page->busy_size = _busy_size;
            // 断言检查，确保指针ptr指向的内存确实位于page所管理的内存范围内
            assert(ptr >= reinterpret_cast<char*>(page) + sizeof(xml_memory_page) && ptr < reinterpret_cast<char*>(page) + sizeof(xml_memory_page) + page->busy_size);
            (void)!ptr;
            // 更新页面已释放内存的大小
            page->freed_size += size;
            // 断言检查，确保已释放的内存大小不会超过该页面的忙碌大小
            assert(page->freed_size <= page->busy_size);
            // 如果整个页面的内存都已释放，则进行清理操作
            if (page->freed_size == page->busy_size)
            {
                // 如果这是最后一个页面（即没有下一个页面），
                if (page->next == 0)
                {
                    // 断言检查，确保这是根页面。
                    assert(_root == page);
 
                    // 如果是顶部页面被释放，则重置其大小。
                    // top page freed, just reset sizes
                    page->busy_size = 0;
                    page->freed_size = 0;
 
#ifdef PUGIXML_COMPACT
                    // reset compact state to maximize efficiency
                    page->compact_string_base = 0;
                    page->compact_shared_parent = 0;
                    page->compact_page_marker = 0;
#endif
                    // 重置全局忙碌大小。
                    _busy_size = 0;
                }
                else
                {
                    // 断言检查，确保这不是根页面，并且它有前一个页面。
                    assert(_root != page);
                    assert(page->prev);
                    // 从页面中移除该页面（假设页面存储在一个双向链表中）
                    // remove from the list
                    page->prev->next = page->next;
                    page->next->prev = page->prev;
 
                    // 释放该页面。
                    // deallocate
                    deallocate_page(page);
                }
            }
        }
                // allocate_string函数用于分配一段内存空间
        char_t* allocate_string(size_t length)
        {        // 定义一个静态常量，表示最大编码偏移量。
            static const size_t max_encoded_offset = (1 << 16) * xml_memory_block_alignment;
                         // 这是一种在编译期间进行条件检查的方式，
            PUGI__STATIC_ASSERT(xml_memory_page_size <= max_encoded_offset);
                         // 计算为存储字符串以及相关头部信息（可能用于管理字符串内存块的一些元数据）所需分配的内存大小。
            // allocate memory for string and header block
            size_t size = sizeof(xml_memory_string_header) + length * sizeof(char_t);
 
            // round size up to block alignment boundary
            size_t full_size = (size + (xml_memory_block_alignment - 1)) & ~(xml_memory_block_alignment - 1);
 
            xml_memory_page* page;
            xml_memory_string_header* header = static_cast<xml_memory_string_header*>(allocate_memory(full_size, page));
 
            if (!header) return 0;
 
            // setup header
            ptrdiff_t page_offset = reinterpret_cast<char*>(header) - reinterpret_cast<char*>(page) - sizeof(xml_memory_page);
 
            assert(page_offset % xml_memory_block_alignment == 0);
            assert(page_offset >= 0 && static_cast<size_t>(page_offset) < max_encoded_offset);
            header->page_offset = static_cast<uint16_t>(static_cast<size_t>(page_offset) / xml_memory_block_alignment);
 
            // full_size == 0 for large strings that occupy the whole page
            assert(full_size % xml_memory_block_alignment == 0);
            assert(full_size < max_encoded_offset || (page->busy_size == full_size && page_offset == 0));
            header->full_size = static_cast<uint16_t>(full_size < max_encoded_offset ? full_size / xml_memory_block_alignment : 0);
 
            // round-trip through void* to avoid 'cast increases required alignment of target type' warning
            // header is guaranteed a pointer-sized alignment, which should be enough for char_t
            return static_cast<char_t*>(static_cast<void*>(header + 1));
        }
 
        void deallocate_string(char_t* string)
        {
            // this function casts pointers through void* to avoid 'cast increases required alignment of target type' warnings
             // 此函数通过将指针转换为 void* 类型来避免出现“cast increases required alignment of target type”这样的警告
            // we're guaranteed the proper (pointer-sized) alignment on the input string if it was allocated via allocate_string
                        // 前提是如果输入的字符串是通过 allocate_string 函数分配的，那么我们能保证输入字符串具有合适的（指针大小的）对齐方式
            // get header
            // 将传入的字符串指针先转换为 void* 类型，再将其转换回 xml_memory_string_header* 类型，并向前偏移一个单位（减1操作），目的是获取指向该字符串头部信息的指针。
            // 这里假设内存布局中字符串头部信息就在字符串实际数据的前面，通过这样的指针运算来获取相关管理信息。
            xml_memory_string_header* header = static_cast<xml_memory_string_header*>(static_cast<void*>(string)) - 1;
            assert(header);
 
            // deallocate
            // 计算此字符串所在内存页内的偏移量。先加上 xml_memory_page 结构体的大小，再根据字符串头部记录的页内偏移量（乘以每个内存块的对齐大小）来确定其在内存页中的准确偏移位置。
                        // 这样后续就能通过这个偏移量找到对应的内存页。
            size_t page_offset = sizeof(xml_memory_page) + header->page_offset * xml_memory_block_alignment;
            // 通过先将 header 指针转换为 char* 类型（方便进行字节级别的偏移计算），减去前面计算得到的页偏移量，再转换回 xml_memory_page* 类型，从而得到该字符串所在的内存页指针。
            xml_memory_page* page = reinterpret_cast<xml_memory_page*>(static_cast<void*>(reinterpret_cast<char*>(header) - page_offset));
            // 如果 full_size 等于 0，说明这个字符串占据了整个内存页，否则按照头部记录的实际大小（乘以内存块对齐大小，可能涉及到内存对齐相关的换算）来确定字符串实际占用的内存大小。
            // if full_size == 0 then this string occupies the whole page
            size_t full_size = header->full_size == 0 ? page->busy_size : header->full_size * xml_memory_block_alignment;
 
            deallocate_memory(header, full_size, page);
        }
 
        bool reserve()
        {
        #ifdef PUGIXML_COMPACT
            return _hash->reserve();
        #else
            return true;
        #endif
        }
 
        xml_memory_page* _root;
        size_t _busy_size;
 
    #ifdef PUGIXML_COMPACT
        compact_hash_table* _hash;
    #endif
    };
    // 该函数用于在内存不足（out of bounds，简称OOB）的情况下分配内存。
    // 参数size指定了要分配的内存大小，out_page通过引用返回分配的内存所在的页面。
    PUGI__FN_NO_INLINE void* xml_allocator::allocate_memory_oob(size_t size, xml_memory_page*& out_page)
    {
        // 定义一个阈值，用于区分“大”分配和“小”分配。
        // 这里，大分配是指超过页面大小四分之一的分配。
        const size_t large_allocation_threshold = xml_memory_page_size / 4;
        // 根据分配大小分配一个页面。如果分配大小超过阈值，则分配一个足够大的页面
        xml_memory_page* page = allocate_page(size <= large_allocation_threshold ? xml_memory_page_size : size);
        // 通过引用参数返回分配的页面
        out_page = page;
        // 如果页面分配失败（即page为nullptr），则返回nullptr表示分配失败。
        if (!page) return 0;
        // 如果分配大小小于或等于阈值，则执行以下操作
        if (size <= large_allocation_threshold)
        {
            // 更新根页面的忙碌大小为全局忙碌大小
            _root->busy_size = _busy_size;
 
            // insert page at the end of linked list
            page->prev = _root;
            _root->next = page;
            // 更新_root指针，使其指向新插入的页面。
            _root = page;
            // 更新全局忙碌大小为当前分配的大小。
            _busy_size = size;
        }
        else
        {
            // insert page before the end of linked list, so that it is deleted as soon as possible
            // the last page is not deleted even if it's empty (see deallocate_memory)
            // 对于大分配，将页面插入到链表末尾之前的位置。
            assert(_root->prev);
            // 插入新页面到链表中。
            page->prev = _root->prev;
            page->next = _root;
 
            _root->prev->next = page;
            _root->prev = page;
            // 设置新页面的忙碌大小为当前分配的大小
            page->busy_size = size;
        }
        // 返回指向页面内部数据的指针（跳过页面头部）。
        return reinterpret_cast<char*>(page) + sizeof(xml_memory_page);
    }
PUGI__NS_END
 
#ifdef PUGIXML_COMPACT
// 开始命名空间（假设PUGI__NS_BEGIN是一个宏，用于定义或进入特定的命名空间）
PUGI__NS_BEGIN
    // 定义紧凑对齐的log2值，这里为2，意味着对齐是2的2次方，即4字节对齐
    static const uintptr_t compact_alignment_log2 = 2;
    // 根据log2值计算紧凑对齐的实际值，这里是1 << 2 = 4
    static const uintptr_t compact_alignment = 1 << compact_alignment_log2;
    // 定义一个紧凑头部类，用于管理紧凑内存页的头部信息
    class compact_header
    {
    public:
        // 构造函数，接收一个指向xml_memory_page的指针和一个标志位
        compact_header(xml_memory_page* page, unsigned int flags)
        {
            // 静态断言，确保xml_memory_block_alignment与compact_alignment相等
            PUGI__STATIC_ASSERT(xml_memory_block_alignment == compact_alignment);
            // 计算当前对象相对于page->compact_page_marker的偏移量
            ptrdiff_t offset = (reinterpret_cast<char*>(this) - reinterpret_cast<char*>(page->compact_page_marker));
            // 断言偏移量是对齐的，并且小于256个对齐单位
            assert(offset % compact_alignment == 0 && static_cast<uintptr_t>(offset) < 256 * compact_alignment);
            // 计算并存储页面索引
            _page = static_cast<unsigned char>(offset >> compact_alignment_log2);
            // 存储标志位
            _flags = static_cast<unsigned char>(flags);
        }
        // 位与赋值操作符，用于修改标志位
        void operator&=(uintptr_t mod)
        {
            _flags &= static_cast<unsigned char>(mod);
        }
        // 位或赋值操作符，用于修改标志位
        void operator|=(uintptr_t mod)
        {
            _flags |= static_cast<unsigned char>(mod);
        }
        // 位与操作符，用于获取标志位与给定值的交集
        uintptr_t operator&(uintptr_t mod) const
        {
            return _flags & mod;
        }
        // 获取关联的xml_memory_page对象
        xml_memory_page* get_page() const
        {
            // round-trip through void* to silence 'cast increases required alignment of target type' warnings
            const char* page_marker = reinterpret_cast<const char*>(this) - (_page << compact_alignment_log2);
            const char* page = page_marker - *reinterpret_cast<const uint32_t*>(static_cast<const void*>(page_marker));
            // 返回指向xml_memory_page的指针
            return const_cast<xml_memory_page*>(reinterpret_cast<const xml_memory_page*>(static_cast<const void*>(page)));
        }
    // 紧凑头部类的私有成员变量定义
    private:
        // 存储页面索引的变量，用于标识当前对象所在的内存页
        unsigned char _page;
        // 存储标志位的变量，用于记录一些额外的信息或状态
        unsigned char _flags;
    };
    // 一个函数，用于根据给定的对象和头部偏移量获取对应的xml_memory_page
    //   指向xml_memory_page的指针，该页面包含了object
    PUGI__FN xml_memory_page* compact_get_page(const void* object, int header_offset)
    {
        // 将object指针向前移动header_offset，得到compact_header的指针
        const compact_header* header = reinterpret_cast<const compact_header*>(static_cast<const char*>(object) - header_offset);
        // 调用compact_header的get_page方法获取xml_memory_page指针
        return header->get_page();
    }
    // 一个模板函数，用于根据给定的对象和头部偏移量获取对应的值
    //   指向T类型的指针，该值通过哈希表查找得到
    template <int header_offset, typename T> PUGI__FN_NO_INLINE T* compact_get_value(const void* object)
    {// 首先获取包含object的xml_memory_page,然后通过页面的allocator和_hash成员查找object对应的值
        return static_cast<T*>(compact_get_page(object, header_offset)->allocator->_hash->find(object));
    }
    // 一个模板函数，用于根据给定的对象和头部偏移量设置对应的值
    template <int header_offset, typename T> PUGI__FN_NO_INLINE void compact_set_value(const void* object, T* value)
    {
        // 首先获取包含object的xml_memory_page,然后通过页面的allocator和_hash成员插入object和value的对应关系
        compact_get_page(object, header_offset)->allocator->_hash->insert(object, value);
    }
    // 一个模板类，用于管理紧凑指针，这些指针通过哈希表进行存储和查找
    template <typename T, int header_offset, int start = -126> class compact_pointer
    {
    public:
        // compact_pointer的默认构造函数，初始化_data成员为0
        compact_pointer(): _data(0)
        {
        }
        // 重载赋值运算符，用于将一个compact_pointer对象赋值给另一个
        void operator=(const compact_pointer& rhs)
        {
            *this = rhs + 0;
        }
        // 重载赋值运算符，用于将一个原生指针赋值给compact_pointer对象
        void operator=(T* value)
        {
            if (value)
            {
                // value is guaranteed to be compact-aligned; 'this' is not
                // value变量是被确保按照紧凑对齐（compact-aligned）方式进行对齐的；而“this”指针所指向的对象却并非如此。
                // our decoding is based on 'this' aligned to compact alignment downwards (see operator T*)
                // 这里所说的紧凑对齐应该是一种特定的内存对齐要求，意味着value的内存地址满足相应的对齐规则，便于后续某些操作（比如数据的读取、处理等）更高效地进行，而“this”指向的对象不符合该对齐规则。
                // so for negative offsets (e.g. -3) we need to adjust the diff by compact_alignment - 1 to
                // 我们的解码操作是基于将“this”指针按照紧凑对齐方式向下对齐来进行的（可以参考 operator T* 相关部分，推测那里可能定义了具体的对齐操作或者转换逻辑）。
                // compensate for arithmetic shift rounding for negative values
                // 所以，当遇到负的偏移量（例如 -3）时，我们需要通过将偏移量的差值（diff）调整为 compact_alignment - 1。
                ptrdiff_t diff = reinterpret_cast<char*>(value) - reinterpret_cast<char*>(this);
                ptrdiff_t offset = ((diff + int(compact_alignment - 1)) >> compact_alignment_log2) - start;
                // 如果计算出的偏移量（经过调整并转换为无符号后）小于等于253，
                // 则将其加1后存储到_data中（因为0被保留用于空指针）
                if (static_cast<uintptr_t>(offset) <= 253)
                    _data = static_cast<unsigned char>(offset + 1);
                else
                {
                    // compact_set_value是一个模板函数，header_offset是一个占位符，表示可能的头部偏移量
                    compact_set_value<header_offset>(this, value);
 
                    _data = 255;// 设置_data为255，表示指针值存储在外部
                }
            }
            else
                // 如果赋值为nullptr，则将_data设置为0
                _data = 0;
        }
        // 定义一个类型转换运算符，将compact_pointer转换为T*
        operator T*() const
        {
            if (_data)// 如果_data不为0
            {
                if (_data < 255)// 如果_data的值小于255，表示指针值直接存储在_data中（经过编码）
                {
                    // 计算基地址：将this指针向下对齐到compact_alignment的倍数
                    uintptr_t base = reinterpret_cast<uintptr_t>(this) & ~(compact_alignment - 1);
                    // 根据_data计算偏移量，并加上start（可能是一个调整值），然后乘以compact_alignment得到实际地址
                    // 最后，将基地址与计算出的偏移量相加，得到目标指针的地址，并转换为T*类型
                    return reinterpret_cast<T*>(base + (_data - 1 + start) * compact_alignment);
                }
                else  // 如果_data的值等于255，表示指针值存储在外部
                     // 调用compact_get_value函数来获取存储在外部的指针值
                    return compact_get_value<header_offset, T>(this);
            }
            else// 如果_data为0，表示空指针
                return 0;// 返回nullptr
        }
        // 定义一个成员访问运算符，允许通过compact_pointer对象直接访问目标对象的成员
        T* operator->() const
        {
            return *this;
        }
    // compact_pointer类的私有成员定义结束
    private:
        // 用于存储编码后的指针值或标记指针值是否存储在外部
        unsigned char _data;
    };
    // 定义一个模板类compact_pointer_parent，它可能是compact_pointer的基类或用于提供某些共享功能
    template <typename T, int header_offset> class compact_pointer_parent
    {
    // 类的构造函数，初始化成员变量_data为0
    public:
        compact_pointer_parent(): _data(0)
        {
        }
        // 拷贝赋值操作符
        void operator=(const compact_pointer_parent& rhs)
        {
            *this = rhs + 0;
        }
        // 赋值操作符重载，接受一个T*类型的参数
        void operator=(T* value)
        {
            if (value)
            {
                // value is guaranteed to be compact-aligned; 'this' is not
                // our decoding is based on 'this' aligned to compact alignment downwards (see operator T*)
                // so for negative offsets (e.g. -3) we need to adjust the diff by compact_alignment - 1 to
                // compensate for arithmetic shift behavior for negative values
                // 将value和this指针转换为char*，然后计算它们之间的差值diff
                ptrdiff_t diff = reinterpret_cast<char*>(value) - reinterpret_cast<char*>(this);
                ptrdiff_t offset = ((diff + int(compact_alignment - 1)) >> compact_alignment_log2) + 65533;
                // 如果计算出的offset在可存储范围内（即小于或等于65533），则直接存储为_data的值（加1是为了留出0表示null的特殊情况）
                if (static_cast<uintptr_t>(offset) <= 65533)
                {
                    _data = static_cast<unsigned short>(offset + 1);
                }
                else
                {
                    // 如果offset超出范围，则需要使用更复杂的内存管理机制
                    // 首先，获取与this相关的内存页
                    xml_memory_page* page = compact_get_page(this, header_offset);
                    // 如果该页还没有共享父指针，则设置它
                    if (PUGI__UNLIKELY(page->compact_shared_parent == 0))
                        page->compact_shared_parent = value;
                    // 如果共享父指针已经是value，则使用特殊值65534表示
                    if (page->compact_shared_parent == value)
                    {
                        _data = 65534;
                    }
                    else
                    {\
                        // 否则，使用某种机制将value存储到内存页中，并将_data设置为特殊值65535
                        compact_set_value<header_offset>(this, value);
 
                        _data = 65535;
                    }
                }
            }
            else
            {
                // 如果value为nullptr，则将_data设置为0，表示null指针
                _data = 0;
            }
        }
 
        operator T*() const
        {
            if (_data)// 如果_data非零
            {
                if (_data < 65534)// 如果_data的值小于65534
                {
                    // 调整当前对象的地址，使其符合compact_alignment对齐要求
                    uintptr_t base = reinterpret_cast<uintptr_t>(this) & ~(compact_alignment - 1);
                    // 根据_data的值计算实际存储T类型数据的地址，并返回该地址
                    // 这里假设了一个特殊的存储策略，其中65533作为偏移量的基准点
                    return reinterpret_cast<T*>(base + (_data - 1 - 65533) * compact_alignment);
                }
                else if (_data == 65534)// 如果_data的值等于65534
                    // 调用compact_get_page函数获取一个特定的页面，并返回该页面中compact_shared_parent指向的T类型数据的地址
                    return static_cast<T*>(compact_get_page(this, header_offset)->compact_shared_parent);
                else// 如果_data的值大于65534
                    // 调用compact_get_value函数，根据_data的值和header_offset获取并返回T类型数据的地址
                    return compact_get_value<header_offset, T>(this);
            }
            else// 如果_data为零
                // 返回空指针，表示没有有效的T类型数据
                return 0;
        }
 
        T* operator->() const
        {
            // 返回当前对象转换为T*的结果，允许通过->操作符访问T类型对象的成员
            return *this;
        }
 
    private:
        uint16_t _data;
    };
 
    template <int header_offset, int base_offset> class compact_string
    {
    public:
        compact_string(): _data(0)
        {
        }
 
        void operator=(const compact_string& rhs)
        {
            *this = rhs + 0;
        }
 
        void operator=(char_t* value)
        {
            if (value)
            {
                // 获取与当前对象关联的内存页面
                xml_memory_page* page = compact_get_page(this, header_offset);
                // 如果compact_string_base为0（即尚未初始化或没有存储任何字符串），则将其设置为value
                if (PUGI__UNLIKELY(page->compact_string_base == 0))
                    page->compact_string_base = value;
                // 计算value相对于compact_string_base的偏移量
                ptrdiff_t offset = value - page->compact_string_base;
                // 如果偏移量小于65535 * 128（即小于16位带符号整数能表示的最大正偏移量，但这里实际上只使用了15位加上一个标志位）
                if (static_cast<uintptr_t>(offset) < (65535 << 7))
                {
                    // round-trip through void* to silence 'cast increases required alignment of target type' warnings
                    uint16_t* base = reinterpret_cast<uint16_t*>(static_cast<void*>(reinterpret_cast<char*>(this) - base_offset));
                    // 如果base指向的值为0（即尚未存储任何偏移量）
                    if (*base == 0)
                    {
                        // 存储偏移量（右移7位以适应15位的存储，并加1作为标志）到base指向的位置
                        // 同时，将偏移量的低7位（加1作为标志）存储到_data中
                        *base = static_cast<uint16_t>((offset >> 7) + 1);
                        _data = static_cast<unsigned char>((offset & 127) + 1);
                    }
                    else
                    {
                        // 如果已经存储了一个偏移量，则计算新的偏移量是否仍然可以用当前的方式存储
                        ptrdiff_t remainder = offset - ((*base - 1) << 7);
                        // 如果新的偏移量的低7位（加1后）小于等于253（即实际偏移量小于等于252，因为加1作为标志）
                        if (static_cast<uintptr_t>(remainder) <= 253)
                        {
                            // 更新_data以存储新的低7位偏移量（加1作为标志）
                            _data = static_cast<unsigned char>(remainder + 1);
                        }
                        else
                        {
                            // 如果新的偏移量太大，无法用当前的方式存储，则使用另一种存储方法（可能是动态分配）
                            compact_set_value<header_offset>(this, value);
                            // 设置_data为255，作为使用特殊存储方法的标志
                            _data = 255;
                        }
                    }
                }
                else
                {
                    // 如果偏移量太大，无法用当前的方式存储，则同样使用另一种存储方法
                    compact_set_value<header_offset>(this, value);
                    // 设置_data为255，作为使用特殊存储方法的标志
                    _data = 255;
                }
            }
            else
            {
                // 如果提供的值为空，则将_data设置为0，表示没有存储任何值
                _data = 0;
            }
        }
 
        operator char_t*() const
        {
            if (_data)
            {
                if (_data < 255)
                {
                    xml_memory_page* page = compact_get_page(this, header_offset);
 
                    // round-trip through void* to silence 'cast increases required alignment of target type' warnings
                    const uint16_t* base = reinterpret_cast<const uint16_t*>(static_cast<const void*>(reinterpret_cast<const char*>(this) - base_offset));
                    assert(*base);
 
                    ptrdiff_t offset = ((*base - 1) << 7) + (_data - 1);
 
                    return page->compact_string_base + offset;
                }
                else
                {
                    return compact_get_value<header_offset, char_t>(this);
                }
            }
            else
                return 0;
        }
 
    private:
        unsigned char _data;
    };
// PUGI__NS_END 和 PUGI__NS_BEGIN 是命名空间结束和开始的宏，用于创建一个命名空间。
// 这样可以避免全局命名冲突，并组织代码结构。
PUGI__NS_END
#endif
 
#ifdef PUGIXML_COMPACT
namespace pugi
{
    // xml_attribute_struct 结构体表示一个 XML 属性。
    struct xml_attribute_struct
    {
        // 构造函数初始化属性结构。
        xml_attribute_struct(impl::xml_memory_page* page): header(page, 0), namevalue_base(0)
        {
            PUGI__STATIC_ASSERT(sizeof(xml_attribute_struct) == 8);  // 确保结构体大小为 8 字节。
        }
 
        impl::compact_header header;  // 紧凑模式下的头部信息。
 
        uint16_t namevalue_base;  // 名称和值的基础偏移量。
 
        impl::compact_string<4, 2> name;  // 紧凑字符串，用于存储属性名。
        impl::compact_string<5, 3> value;  // 紧凑字符串，用于存储属性值。
 
        impl::compact_pointer<xml_attribute_struct, 6> prev_attribute_c;  // 前一个属性的指针。
        impl::compact_pointer<xml_attribute_struct, 7, 0> next_attribute;  // 下一个属性的指针。
    };
 
    // xml_node_struct 结构体表示一个 XML 节点。
    struct xml_node_struct
    {
        // 构造函数初始化节点结构。
        xml_node_struct(impl::xml_memory_page* page, xml_node_type type): header(page, type), namevalue_base(0)
        {
            PUGI__STATIC_ASSERT(sizeof(xml_node_struct) == 12);  // 确保结构体大小为 12 字节。
        }
 
        impl::compact_header header;  // 紧凑模式下的头部信息。
 
        uint16_t namevalue_base;  // 名称和值的基础偏移量。
 
        impl::compact_string<4, 2> name;  // 紧凑字符串，用于存储节点名。
        impl::compact_string<5, 3> value;  // 紧凑字符串，用于存储节点值。
 
        impl::compact_pointer_parent<xml_node_struct, 6> parent;  // 父节点的指针。
 
        impl::compact_pointer<xml_node_struct, 8, 0> first_child;  // 第一个子节点的指针。
 
        impl::compact_pointer<xml_node_struct, 9> prev_sibling_c;  // 前一个兄弟节点的指针。
        impl::compact_pointer<xml_node_struct, 10, 0> next_sibling;  // 下一个兄弟节点的指针。
 
        impl::compact_pointer<xml_attribute_struct, 11, 0> first_attribute;  // 第一个属性的指针。
    };
}
#else
namespace pugi
{
    // 在非紧凑模式下，xml_attribute_struct 和 xml_node_struct 结构体的定义与紧凑模式略有不同。
    // 它们使用常规指针和简单的成员变量，而不是紧凑模式下的紧凑指针和字符串。
    struct xml_attribute_struct
    {
        xml_attribute_struct(impl::xml_memory_page* page): name(0), value(0), prev_attribute_c(0), next_attribute(0)
        {
            header = PUGI__GETHEADER_IMPL(this, page, 0);  // 初始化头部信息。
        }
 
        uintptr_t header;  // 头部信息。
 
        char_t* name;  // 属性名。
        char_t* value;  // 属性值。
 
        xml_attribute_struct* prev_attribute_c;  // 前一个属性的指针。
        xml_attribute_struct* next_attribute;  // 下一个属性的指针。
    };
 
    struct xml_node_struct
    {
        xml_node_struct(impl::xml_memory_page* page, xml_node_type type): name(0), value(0), parent(0), first_child(0), prev_sibling_c(0), next_sibling(0), first_attribute(0)
        {
            header = PUGI__GETHEADER_IMPL(this, page, type);  // 初始化头部信息。
        }
 
        uintptr_t header;  // 头部信息。
 
        char_t* name;  // 节点名。
        char_t* value;  // 节点值。
 
        xml_node_struct* parent;  // 父节点的指针。
 
        xml_node_struct* first_child;  // 第一个子节点的指针。
 
        xml_node_struct* prev_sibling_c;  // 前一个兄弟节点的指针。
        xml_node_struct* next_sibling;  // 下一个兄弟节点的指针。
 
        xml_attribute_struct* first_attribute;  // 第一个属性的指针。
    };
}
#endif
 
PUGI__NS_BEGIN
    // xml_extra_buffer 结构体表示一个额外的缓冲区，用于存储 XML 文档中的字符数据。
    struct xml_extra_buffer
    {
        char_t* buffer;  // 缓冲区指针。
        xml_extra_buffer* next;  // 下一个额外缓冲区的指针。
    };
 
    // xml_document_struct 结构体表示一个 XML 文档，继承自 xml_node_struct 和 xml_allocator。
    struct xml_document_struct: public xml_node_struct, public xml_allocator
    {
        xml_document_struct(xml_memory_page* page): xml_node_struct(page, node_document), xml_allocator(page), buffer(0), extra_buffers(0)
        {
        }
 
        const char_t* buffer;  // 文档缓冲区指针。
 
        xml_extra_buffer* extra_buffers;  // 额外缓冲区指针。
 
    #ifdef PUGIXML_COMPACT
        compact_hash_table hash;  // 紧凑模式下的哈希表。
    #endif
    };
 
    // get_allocator 模板函数返回给定对象的分配器。
    template <typename Object> inline xml_allocator& get_allocator(const Object* object)
    {
        assert(object);  // 确保对象非空。
 
        return *PUGI__GETPAGE(object)->allocator;  // 返回对象所在页面的分配器。
    }
 
    // get_document 模板函数返回给定对象所属的文档。
    template <typename Object> inline xml_document_struct& get_document(const Object* object)
    {
        assert(object);  // 确保对象非空。
 
        return *static_cast<xml_document_struct*>(PUGI__GETPAGE(object)->allocator);  // 返回对象所在页面的文档。
    }
PUGI__NS_END
 
// Low-level DOM operations
PUGI__NS_BEGIN
    inline xml_attribute_struct* allocate_attribute(xml_allocator& alloc)// 分配一个属性节点
    {
        xml_memory_page* page; // 声明内存页指针
        void* memory = alloc.allocate_object(sizeof(xml_attribute_struct), page);// 分配属性节点所需的内存
        if (!memory) return 0;// 如果内存分配失败，返回空指针
 
        return new (memory) xml_attribute_struct(page); // 在分配的内存上构造属性节点对象，并返回指针
    }
 
    inline xml_node_struct* allocate_node(xml_allocator& alloc, xml_node_type type)// 分配一个节点
    {
        xml_memory_page* page; // 声明内存页指针
        void* memory = alloc.allocate_object(sizeof(xml_node_struct), page);// 分配节点所需的内存
        if (!memory) return 0;// 如果内存分配失败，返回空指针
 
        return new (memory) xml_node_struct(page, type);// 在分配的内存上构造节点对象，并返回指针
    }
 
    inline void destroy_attribute(xml_attribute_struct* a, xml_allocator& alloc)// 销毁属性节点
    {
        if (a->header & impl::xml_memory_page_name_allocated_mask) // 检查属性名是否需要释放
            alloc.deallocate_string(a->name); // 释放属性名所占的内存
 
        if (a->header & impl::xml_memory_page_value_allocated_mask)// 检查属性值是否需要释放
            alloc.deallocate_string(a->value);// 释放属性值所占的内存
 
        alloc.deallocate_memory(a, sizeof(xml_attribute_struct), PUGI__GETPAGE(a)); // 释放属性节点本身所占的内存
    }
 
    inline void destroy_node(xml_node_struct* n, xml_allocator& alloc) // 销毁节点
    {
        if (n->header & impl::xml_memory_page_name_allocated_mask) // 检查节点名称是否需要释放
            alloc.deallocate_string(n->name); // 释放节点名称所占的内存
 
        if (n->header & impl::xml_memory_page_value_allocated_mask)// 检查节点值是否需要释放
            alloc.deallocate_string(n->value);// 释放节点值所占的内存
 
        for (xml_attribute_struct* attr = n->first_attribute; attr; )// 遍历节点的所有属性
        {
            xml_attribute_struct* next = attr->next_attribute;// 保存下一个属性的指针
 
            destroy_attribute(attr, alloc); // 销毁当前属性
 
            attr = next;// 移动到下一个属性
        }
 
        for (xml_node_struct* child = n->first_child; child; ) // 遍历节点的所有子节点
        {
            xml_node_struct* next = child->next_sibling;// 保存下一个子节点的指针
 
            destroy_node(child, alloc);// 销毁当前子节点
 
            child = next;// 移动到下一个子节点
        }
 
        alloc.deallocate_memory(n, sizeof(xml_node_struct), PUGI__GETPAGE(n));// 释放当前节点本身所占的内存
    }
 
    inline void append_node(xml_node_struct* child, xml_node_struct* node) // 将子节点添加到父节点的子节点列表末尾
    {
        child->parent = node;// 设置子节点的父节点指针为当前节点
 
        xml_node_struct* head = node->first_child; // 获取当前节点的第一个子节点
 
        if (head)// 如果当前节点已经有子节点
        {
            xml_node_struct* tail = head->prev_sibling_c;// 获取当前子节点列表的最后一个节点
 
            tail->next_sibling = child;// 将当前子节点列表的最后一个节点的下一个指针指向新的子节点
            child->prev_sibling_c = tail;// 将新的子节点的前一个指针指向当前子节点列表的最后一个节点
            head->prev_sibling_c = child;// 更新子节点列表的第一个节点的前一个指针为新的子节点
        }
        else// 如果当前节点没有子节点
        {
            node->first_child = child;// 将新的子节点设置为第一个子节点
            child->prev_sibling_c = child;// 子节点的前一个指针指向自身（形成循环链表）
        }
    }
 
    inline void prepend_node(xml_node_struct* child, xml_node_struct* node)// 将子节点添加到父节点的子节点列表开头
    {
        child->parent = node; // 设置子节点的父节点指针为当前节点
 
        xml_node_struct* head = node->first_child;// 获取当前节点的第一个子节点
 
        if (head)// 如果当前节点已经有子节点
        {
            child->prev_sibling_c = head->prev_sibling_c;// 将新子节点的前一个指针设置为当前子节点列表的最后一个节点
            head->prev_sibling_c = child; // 更新子节点列表的第一个节点的前一个指针为新的子节点
        }
        else// 如果当前节点没有子节点
            child->prev_sibling_c = child;// 新子节点的前一个指针指向自身（形成循环链表）
 
        child->next_sibling = head; // 将新子节点的下一个指针设置为当前的第一个子节点
        node->first_child = child;// 更新父节点的第一个子节点为新的子节点
    }
 
    inline void insert_node_after(xml_node_struct* child, xml_node_struct* node)// 在指定节点之后插入一个子节点
    {
        xml_node_struct* parent = node->parent;// 获取指定节点的父节点
 
        child->parent = parent;// 设置新节点的父节点指针为当前父节点
 
        if (node->next_sibling)// 如果指定节点有下一个兄弟节点
            node->next_sibling->prev_sibling_c = child;// 更新下一个兄弟节点的前一个指针为新节点
        else// 如果指定节点是最后一个节点
            parent->first_child->prev_sibling_c = child;// 更新父节点子节点列表的最后一个节点的前一个指针为新节点
 
        child->next_sibling = node->next_sibling;// 设置新节点的下一个指针为指定节点的下一个节点
        child->prev_sibling_c = node;// 设置新节点的前一个指针为指定节点
 
        node->next_sibling = child;// 更新指定节点的下一个指针为新节点
    }
 
    inline void insert_node_before(xml_node_struct* child, xml_node_struct* node)// 在指定节点之前插入一个子节点
    {
        xml_node_struct* parent = node->parent;// 获取指定节点的父节点
 
        child->parent = parent;// 设置新节点的父节点指针为当前父节点
 
        if (node->prev_sibling_c->next_sibling) // 如果指定节点的前一个节点存在并有下一个指针
            node->prev_sibling_c->next_sibling = child; // 更新前一个节点的下一个指针为新节点
        else// 如果指定节点是第一个节点
            parent->first_child = child; // 更新父节点的第一个子节点为新节点
 
        child->prev_sibling_c = node->prev_sibling_c;// 设置新节点的前一个指针为指定节点的前一个节点
        child->next_sibling = node;// 设置新节点的下一个指针为指定节点
 
        node->prev_sibling_c = child;// 更新指定节点的前一个指针为新节点
    }
 
    inline void remove_node(xml_node_struct* node)// 从父节点的子节点列表中移除指定节点
    {
        xml_node_struct* parent = node->parent;// 获取当前节点的父节点
 
        if (node->next_sibling) // 如果当前节点有下一个兄弟节点
            node->next_sibling->prev_sibling_c = node->prev_sibling_c;// 更新下一个兄弟节点的前一个指针为当前节点的前一个兄弟节点
        else// 如果当前节点是最后一个节点
            parent->first_child->prev_sibling_c = node->prev_sibling_c;// 更新父节点子节点列表的最后一个节点为当前节点的前一个兄弟节点
 
        if (node->prev_sibling_c->next_sibling) // 如果当前节点有前一个兄弟节点
            node->prev_sibling_c->next_sibling = node->next_sibling;// 更新前一个兄弟节点的下一个指针为当前节点的下一个兄弟节点
        else// 如果当前节点是第一个节点
            parent->first_child = node->next_sibling;// 更新父节点的第一个子节点为当前节点的下一个兄弟节点
 
        node->parent = 0; // 将当前节点的父节点指针清空
        node->prev_sibling_c = 0;// 将当前节点的前一个兄弟节点指针清空
        node->next_sibling = 0;// 将当前节点的下一个兄弟节点指针清空
    }
 
    inline void append_attribute(xml_attribute_struct* attr, xml_node_struct* node)// 将属性添加到节点的属性列表末尾
    {
        xml_attribute_struct* head = node->first_attribute;// 获取节点的第一个属性
 
        if (head)// 如果节点已经有属性
        {
            xml_attribute_struct* tail = head->prev_attribute_c;// 获取属性列表的最后一个属性
 
            tail->next_attribute = attr;// 将最后一个属性的下一个指针指向新属性
            attr->prev_attribute_c = tail;// 将新属性的前一个指针指向最后一个属性
            head->prev_attribute_c = attr;// 更新属性列表的第一个属性的前一个指针为新属性
        }
        else// 如果节点没有属性
        {
            node->first_attribute = attr;// 将新属性设置为第一个属性
            attr->prev_attribute_c = attr;// 新属性的前一个指针指向自身（形成循环链表）
        }
    }
 
    inline void prepend_attribute(xml_attribute_struct* attr, xml_node_struct* node)// 将属性添加到节点的属性列表开头
    {
        xml_attribute_struct* head = node->first_attribute;// 获取节点的第一个属性
 
        if (head)// 如果节点已经有属性
        {
            attr->prev_attribute_c = head->prev_attribute_c;// 将新属性的前一个指针设置为属性列表的最后一个属性
            head->prev_attribute_c = attr;// 更新属性列表第一个属性的前一个指针为新属性
        }
        else// 如果节点没有属性
            attr->prev_attribute_c = attr;// 新属性的前一个指针指向自身（形成循环链表）
 
        attr->next_attribute = head;// 将新属性的下一个指针指向原来的第一个属性
        node->first_attribute = attr;// 更新节点的第一个属性为新属性
    }
 
    inline void insert_attribute_after(xml_attribute_struct* attr, xml_attribute_struct* place, xml_node_struct* node)// 在指定属性之后插入新属性
    {
        if (place->next_attribute)// 如果指定属性有下一个属性
            place->next_attribute->prev_attribute_c = attr; // 更新下一个属性的前一个指针为新属性
        else// 如果指定属性是最后一个属性
            node->first_attribute->prev_attribute_c = attr;// 更新属性列表最后一个属性的前一个指针为新属性
 
        attr->next_attribute = place->next_attribute;// 设置新属性的下一个指针为指定属性的下一个属性
        attr->prev_attribute_c = place; // 设置新属性的前一个指针为指定属性
        place->next_attribute = attr;// 更新指定属性的下一个指针为新属性
    }
 
    inline void insert_attribute_before(xml_attribute_struct* attr, xml_attribute_struct* place, xml_node_struct* node)// 在指定属性之前插入新属性
    {
        if (place->prev_attribute_c->next_attribute)// 如果指定属性的前一个属性有下一个指针
            place->prev_attribute_c->next_attribute = attr;// 更新前一个属性的下一个指针为新属性
        else// 如果指定属性是第一个属性
            node->first_attribute = attr;// 更新节点的第一个属性为新属性
 
        attr->prev_attribute_c = place->prev_attribute_c;// 设置新属性的前一个指针为指定属性的前一个属性
        attr->next_attribute = place;// 设置新属性的下一个指针为指定属性
        place->prev_attribute_c = attr;// 更新指定属性的前一个指针为新属性
    }
 
    inline void remove_attribute(xml_attribute_struct* attr, xml_node_struct* node)// 从节点的属性列表中移除指定属性
    {
        if (attr->next_attribute)// 如果指定属性有下一个属性
            attr->next_attribute->prev_attribute_c = attr->prev_attribute_c;// 更新下一个属性的前一个指针为指定属性的前一个属性
        else// 如果指定属性是最后一个属性
            node->first_attribute->prev_attribute_c = attr->prev_attribute_c;// 更新属性列表最后一个属性为指定属性的前一个属性
 
        if (attr->prev_attribute_c->next_attribute)// 如果指定属性有前一个属性
            attr->prev_attribute_c->next_attribute = attr->next_attribute;// 更新前一个属性的下一个指针为指定属性的下一个属性
        else// 如果指定属性是第一个属性
            node->first_attribute = attr->next_attribute;// 更新节点的第一个属性为指定属性的下一个属性
 
        attr->prev_attribute_c = 0;// 将指定属性的前一个指针清空
        attr->next_attribute = 0;// 将指定属性的下一个指针清空
    }
 
    PUGI__FN_NO_INLINE xml_node_struct* append_new_node(xml_node_struct* node, xml_allocator& alloc, xml_node_type type = node_element)// 在节点末尾添加一个新节点
    {
        if (!alloc.reserve()) return 0;// 检查内存分配器是否有足够的空间，若没有返回空指针
 
        xml_node_struct* child = allocate_node(alloc, type);// 分配一个新的节点
        if (!child) return 0; // 如果分配失败，返回空指针
 
        append_node(child, node);// 将新节点添加到指定节点的子节点列表末尾
 
        return child;// 返回新创建的节点
    }
 
    PUGI__FN_NO_INLINE xml_attribute_struct* append_new_attribute(xml_node_struct* node, xml_allocator& alloc)// 在节点末尾添加一个新属性
    {
        if (!alloc.reserve()) return 0; // 检查内存分配器是否有足够的空间，若没有返回空指针
 
        xml_attribute_struct* attr = allocate_attribute(alloc);// 分配一个新的属性
        if (!attr) return 0;// 如果分配失败，返回空指针
 
        append_attribute(attr, node); // 将新属性添加到指定节点的属性列表末尾
 
        return attr;// 返回新创建的属性
    }
PUGI__NS_END
 
// Helper classes for code generation
PUGI__NS_BEGIN
    struct opt_false // 定义一个表示逻辑假值的结构体
    {
        enum { value = 0 };// 静态成员变量，值为0
    };
 
    struct opt_true// 定义一个表示逻辑真值的结构体
    {
        enum { value = 1 };// 静态成员变量，值为1
    };
PUGI__NS_END
 
// Unicode utilities
PUGI__NS_BEGIN
    inline uint16_t endian_swap(uint16_t value)// 交换16位整数的字节顺序（大小端转换）
    {
        return static_cast<uint16_t>(((value & 0xff) << 8) | (value >> 8));// 将低8位移到高8位，将高8位移到低8位
    }
 
    inline uint32_t endian_swap(uint32_t value)// 交换32位整数的字节顺序（大小端转换）
    {
        return ((value & 0xff) << 24) | ((value & 0xff00) << 8) | ((value & 0xff0000) >> 8) | (value >> 24);
    }
 
    struct utf8_counter
    {
        typedef size_t value_type;
 
        static value_type low(value_type result, uint32_t ch)
        {
            // U+0000..U+007F
            if (ch < 0x80) return result + 1;
            // U+0080..U+07FF
            else if (ch < 0x800) return result + 2;
            // U+0800..U+FFFF
            else return result + 3;
        }
 
        static value_type high(value_type result, uint32_t)
        {
            // U+10000..U+10FFFF
            return result + 4;
        }
    };
 
    struct utf8_writer
    {
        typedef uint8_t* value_type;
 
        static value_type low(value_type result, uint32_t ch)
        {
            // U+0000..U+007F
            if (ch < 0x80)
            {
                *result = static_cast<uint8_t>(ch);
                return result + 1;
            }
            // U+0080..U+07FF
            else if (ch < 0x800)
            {
                result[0] = static_cast<uint8_t>(0xC0 | (ch >> 6));
                result[1] = static_cast<uint8_t>(0x80 | (ch & 0x3F));
                return result + 2;
            }
            // U+0800..U+FFFF
            else
            {
                result[0] = static_cast<uint8_t>(0xE0 | (ch >> 12));
                result[1] = static_cast<uint8_t>(0x80 | ((ch >> 6) & 0x3F));
                result[2] = static_cast<uint8_t>(0x80 | (ch & 0x3F));
                return result + 3;
            }
        }
 
        static value_type high(value_type result, uint32_t ch)
        {
            // U+10000..U+10FFFF
            result[0] = static_cast<uint8_t>(0xF0 | (ch >> 18));
            result[1] = static_cast<uint8_t>(0x80 | ((ch >> 12) & 0x3F));
            result[2] = static_cast<uint8_t>(0x80 | ((ch >> 6) & 0x3F));
            result[3] = static_cast<uint8_t>(0x80 | (ch & 0x3F));
            return result + 4;
        }
 
        static value_type any(value_type result, uint32_t ch)
        {
            return (ch < 0x10000) ? low(result, ch) : high(result, ch);
        }
    };
 
    struct utf16_counter// 用于计数 UTF-16 编码的结构体
    {
        typedef size_t value_type;// 定义 value_type 为 size_t 类型
 
        static value_type low(value_type result, uint32_t)// 处理低位字符时增加计数
        {
            return result + 1;// 返回当前计数加 1
        }
 
        static value_type high(value_type result, uint32_t)// 处理高位字符时增加计数
        {
            return result + 2; // 返回当前计数加 2，因为高位字符占用 2 个字节
        }
    };
 
    struct utf16_writer// 用于将字符写入 UTF-16 编码的结构体
    {
        typedef uint16_t* value_type;// 定义 value_type 为 uint16_t 指针类型
 
        static value_type low(value_type result, uint32_t ch)// 写入低位字符
        {
            *result = static_cast<uint16_t>(ch);// 将字符 ch 转换为 uint16_t 并写入 result
 
            return result + 1;// 返回指向下一个位置的指针
        }
 
        static value_type high(value_type result, uint32_t ch)// 写入高位字符
        {
            uint32_t msh = static_cast<uint32_t>(ch - 0x10000) >> 10;// 获取高 10 位
            uint32_t lsh = static_cast<uint32_t>(ch - 0x10000) & 0x3ff;// 获取低 10 位
 
            result[0] = static_cast<uint16_t>(0xD800 + msh);// 计算并写入高位字符
            result[1] = static_cast<uint16_t>(0xDC00 + lsh);// 计算并写入低位字符
 
            return result + 2;// 返回指向下一个位置的指针，已经写入两个字节
        }
 
        static value_type any(value_type result, uint32_t ch)// 根据字符 ch 判断是写入低位字符还是高位字符
        {
            return (ch < 0x10000) ? low(result, ch) : high(result, ch);// 如果字符小于 0x10000，调用 low，否则调用 high
        }
    };
 
    struct utf32_counter// 用于计数 UTF-32 编码的结构体
    {
        typedef size_t value_type;// 定义 value_type 为 size_t 类型
 
        static value_type low(value_type result, uint32_t)// 处理低位字符时增加计数
        {
            return result + 1;// 返回当前计数加 1
        }
 
        static value_type high(value_type result, uint32_t)// 处理高位字符时增加计数
        {
            return result + 1;// 返回当前计数加 1，UTF-32 中每个字符都是 1 个单位
        }
    };
 
    struct utf32_writer// 用于将字符写入 UTF-32 编码的结构体
    {
        typedef uint32_t* value_type;// 定义 value_type 为 uint32_t 指针类型
 
        static value_type low(value_type result, uint32_t ch)// 写入低位字符
        {
            *result = ch;// 将字符 ch 直接赋值给 result
 
            return result + 1;// 返回指向下一个位置的指针
        }
 
        static value_type high(value_type result, uint32_t ch)// 写入高位字符
        {
            *result = ch;// 将字符 ch 直接赋值给 result
 
            return result + 1;// 返回指向下一个位置的指针
        }
 
        static value_type any(value_type result, uint32_t ch)// 写入任何字符
        {
            *result = ch; // 将字符 ch 直接赋值给 result
 
            return result + 1;// 返回指向下一个位置的指针
        }
    };
 
    struct latin1_writer// 用于将字符写入 Latin-1 编码的结构体
    {
        typedef uint8_t* value_type; // 定义 value_type 为 uint8_t 指针类型
 
        static value_type low(value_type result, uint32_t ch)// 写入低位字符
        {
            *result = static_cast<uint8_t>(ch > 255 ? '?' : ch);// 如果字符值大于 255，则写入 '?'，否则写入字符本身（转换为 uint8_t）
 
            return result + 1;// 返回指向下一个位置的指针
        }
 
        static value_type high(value_type result, uint32_t ch)// 写入高位字符
        {
            (void)ch;// 忽略传入的 ch 参数，因为 Latin-1 只能处理单字节字符
 
            *result = '?';// 始终写入 '?' 作为占位符
 
            return result + 1;// 返回指向下一个位置的指针
        }
    };
 
    struct utf8_decoder// 用于解码 UTF-8 的结构体
    {
        typedef uint8_t type;// 定义 type 为 uint8_t 类型
 
        template <typename Traits> static inline typename Traits::value_type process(const uint8_t* data, size_t size, typename Traits::value_type result, Traits)// 解码过程函数
        {
            const uint8_t utf8_byte_mask = 0x3f;// 定义用于提取有效位的掩码
 
            while (size)// 遍历输入数据
            {
                uint8_t lead = *data;// 获取当前字节作为首字节
 
 
                // 0xxxxxxx -> U+0000..U+007F
                if (lead < 0x80)// 单字节 ASCII 字符处理
                {
                    result = Traits::low(result, lead);// 通过 Traits 处理单字节字符
                    data += 1;// 移动到下一个字节
                    size -= 1;// 减少字节计数
 
                    // process aligned single-byte (ascii) blocks
                    if ((reinterpret_cast<uintptr_t>(data) & 3) == 0)// 检查数据是否对齐到 4 字节边界
                    {
                        // round-trip through void* to silence 'cast increases required alignment of target type' warnings
                        while (size >= 4 && (*static_cast<const uint32_t*>(static_cast<const void*>(data)) & 0x80808080) == 0)// 快速处理连续的 ASCII 字符块
                        {
                            result = Traits::low(result, data[0]);// 处理第一个字节
                            result = Traits::low(result, data[1]);// 处理第二个字节
                            result = Traits::low(result, data[2]);// 处理第三个字节
                            result = Traits::low(result, data[3]);// 处理第四个字节
                            data += 4;// 移动到下一个 4 字节块
                            size -= 4;// 减少字节计数
                        }
                    }
                }
                // 110xxxxx -> U+0080..U+07FF
                else if (static_cast<unsigned int>(lead - 0xC0) < 0x20 && size >= 2 && (data[1] & 0xc0) == 0x80)// 两字节 UTF-8 字符处理
                {
                    result = Traits::low(result, ((lead & ~0xC0) << 6) | (data[1] & utf8_byte_mask));// 解码并处理两字节字符
                    data += 2;// 移动到下一个字节
                    size -= 2;// 减少字节计数
                }
                // 1110xxxx -> U+0800-U+FFFF
                else if (static_cast<unsigned int>(lead - 0xE0) < 0x10 && size >= 3 && (data[1] & 0xc0) == 0x80 && (data[2] & 0xc0) == 0x80)// 三字节 UTF-8 字符处理
                {
                    result = Traits::low(result, ((lead & ~0xE0) << 12) | ((data[1] & utf8_byte_mask) << 6) | (data[2] & utf8_byte_mask));// 解码并处理三字节字符
                    data += 3;// 移动到下一个字节
                    size -= 3;// 减少字节计数
                }
                // 11110xxx -> U+10000..U+10FFFF
                else if (static_cast<unsigned int>(lead - 0xF0) < 0x08 && size >= 4 && (data[1] & 0xc0) == 0x80 && (data[2] & 0xc0) == 0x80 && (data[3] & 0xc0) == 0x80)// 四字节 UTF-8 字符处理
                {
                    result = Traits::high(result, ((lead & ~0xF0) << 18) | ((data[1] & utf8_byte_mask) << 12) | ((data[2] & utf8_byte_mask) << 6) | (data[3] & utf8_byte_mask));// 解码并处理四字节字符
                    data += 4; // 移动到下一个字节
                    size -= 4; // 减少字节计数
                }
                // 10xxxxxx or 11111xxx -> invalid
                else// 非法字节处理
                {
                    data += 1;// 跳过非法字节
                    size -= 1;// 减少字节计数
                }
            }
 
            return result;// 返回处理结果
        }
    };
 
    template <typename opt_swap> struct utf16_decoder
    {
        typedef uint16_t type; // 定义解码器处理的数据类型为uint16_t
// 定义一个模板函数process，用于处理UTF-16编码的数据
        template <typename Traits> static inline typename Traits::value_type process(const uint16_t* data, size_t size, typename Traits::value_type result, Traits)
        {
            while (size)// 当还有数据需要处理时
            {
                uint16_t lead = opt_swap::value ? endian_swap(*data) : *data;// 根据opt_swap决定是否需要字节顺序交换
 
                // U+0000..U+D7FF
                if (lead < 0xD800)
                {
                    result = Traits::low(result, lead); // 处理非代理项高位字符
                    data += 1;// 移动到下一个数据
                    size -= 1;// 数据大小减一
                }
                // U+E000..U+FFFF
                else if (static_cast<unsigned int>(lead - 0xE000) < 0x2000)
                {
                    result = Traits::low(result, lead);// 处理直接映射的单个字符
                    data += 1;// 移动到下一个数据
                    size -= 1;// 数据大小减一
                }
                // surrogate pair lead
                else if (static_cast<unsigned int>(lead - 0xD800) < 0x400 && size >= 2)
                {
                    uint16_t next = opt_swap::value ? endian_swap(data[1]) : data[1];// 根据opt_swap决定是否需要字节顺序交换下一个数据
// 检查是否是一个有效的代理项对低位（U+DC00到U+DFFF）
                    if (static_cast<unsigned int>(next - 0xDC00) < 0x400)
                    {// 合并代理项对高位和低位为一个Unicode码点（U+10000到U+10FFFF）
                        result = Traits::high(result, 0x10000 + ((lead & 0x3ff) << 10) + (next & 0x3ff));
                        data += 2;// 移动到下一个数据对之后
                        size -= 2;// 数据大小减二
                    }
                    else
                    {
                        data += 1;// 如果不是有效的代理项对低位，只移动到一个数据之后
                        size -= 1;// 数据大小减一
                    }
                }
                else
                {
                    data += 1;// 对于不符合上述条件的字符，只移动到一个数据之后
                    size -= 1;// 数据大小减一
                }
            }
 
            return result;// 返回处理后的结果
        }
    };
 
    template <typename opt_swap> struct utf32_decoder
    {// 定义一个类型别名 type，表示 UTF-32 编码的基本单元，即 uint32_t 类型。
        typedef uint32_t type;// type 是 uint32_t 的别名
// 定义一个静态模板方法 process，用于处理 UTF-32 编码的数据。
        template <typename Traits> static inline typename Traits::value_type process(const uint32_t* data, size_t size, typename Traits::value_type result, Traits)
        {
            while (size)
            { 
                // 根据 opt_swap::value 的值决定是否对当前数据进行字节序交换
                // 如果 opt_swap::value 为 true，则调用 endian_swap 函数交换字节序
                // 否则，直接使用原始数据
                uint32_t lead = opt_swap::value ? endian_swap(*data) : *data;// 获取当前 UTF-32 字符，可能已进行字节序交换
 
                // U+0000..U+FFFF
                if (lead < 0x10000)
                {
                    // 使用 Traits 的 low 方法处理低代理对或单独的 BMP 字符
                // low 方法应返回更新后的结果值
                    result = Traits::low(result, lead); // 处理低代理对或 BMP 字符，并更新结果
                    data += 1;// 指针后移，指向下一个 UTF-32 字符
                    size -= 1;// 数据大小减一
                }
                // U+10000..U+10FFFF
                else
                {
                    result = Traits::high(result, lead);// 处理高代理对，并更新结果
                    data += 1;// 指针后移，指向下一个 UTF-32 字符
                    size -= 1;// 数据大小减一
                }
            }
 
            return result;// 返回最终结果
        }
    };
 
    struct latin1_decoder
    {
        typedef uint8_t type;// 定义一个类型别名 `type`，等价于 `uint8_t`。
 
        template <typename Traits> static inline typename Traits::value_type process(const uint8_t* data, size_t size, typename Traits::value_type result, Traits)
        {
            while (size)// 当 `size` 不为 0 时循环。
            {
                result = Traits::low(result, *data);// 调用 `Traits` 的 `low` 函数处理数据并更新 
                data += 1;// 指针向后移动一个字节。
                size -= 1;// 数据大小减 1。
            }
 
            return result;// 返回最终的 `result`。
        }
    };
 
    template <size_t size> struct wchar_selector;// 根据模板参数 `size` 定义 `wchar_selector` 的特化结构。
 
    template <> struct wchar_selector<2>
    {
        typedef uint16_t type; // 定义类型别名 `type`，等价于 `uint16_t`。
        typedef utf16_counter counter; // 定义类型别名 `counter`，等价于 `utf16_counter`。
        typedef utf16_writer writer;// 定义类型别名 `writer`，等价于 `utf16_writer`。
        typedef utf16_decoder<opt_false> decoder; // 定义类型别名 `decoder`，等价于 `utf16_decoder<opt_false>`。
    };
 
    template <> struct wchar_selector<4>
    {
        typedef uint32_t type; // 定义类型别名 `type`，等价于 `uint32_t`。
        typedef utf32_counter counter;// 定义类型别名 `counter`，等价于 `utf32_counter`。
        typedef utf32_writer writer;// 定义类型别名 `writer`，等价于 `utf32_writer`。
        typedef utf32_decoder<opt_false> decoder;// 定义类型别名 `decoder`，等价于 `utf32_decoder<opt_false>`。
    };
 
    typedef wchar_selector<sizeof(wchar_t)>::counter wchar_counter;// 定义类型别名 `wchar_counter`，等价于根据 `wchar_t` 大小选择的 `counter` 类型。
    typedef wchar_selector<sizeof(wchar_t)>::writer wchar_writer; // 定义类型别名 `wchar_writer`，等价于根据 `wchar_t` 大小选择的 `writer` 类型。
 
    struct wchar_decoder
    {
        typedef wchar_t type;// 定义类型别名 `type`，等价于 `wchar_t`。
 
        template <typename Traits> static inline typename Traits::value_type process(const wchar_t* data, size_t size, typename Traits::value_type result, Traits traits)
        {
            typedef wchar_selector<sizeof(wchar_t)>::decoder decoder;// 定义类型别名 `decoder`，根据 `wchar_t` 大小选择合适的解码器。
 
            return decoder::process(reinterpret_cast<const typename decoder::type*>(data), size, result, traits);// 调用 `decoder` 的 `process` 函数，将 `data` 转换为解码器的类型进行处理，并返回处理结果。
        }
    };
 
#ifdef PUGIXML_WCHAR_MODE// 如果定义了 `PUGIXML_WCHAR_MODE` 宏。
    PUGI__FN void convert_wchar_endian_swap(wchar_t* result, const wchar_t* data, size_t length)
    {
        for (size_t i = 0; i < length; ++i)// 遍历输入数据。
            result[i] = static_cast<wchar_t>(endian_swap(static_cast<wchar_selector<sizeof(wchar_t)>::type>(data[i])));// 将输入数据的每个元素转换为选择器类型，进行字节序交换后再转换为 `wchar_t` 存储到结果数组中。
    }
#endif
PUGI__NS_END
 
PUGI__NS_BEGIN
    enum chartype_t
    {
        ct_parse_pcdata = 1,    // \0, &, \r, <
        ct_parse_attr = 2,      // \0, &, \r, ', "
        ct_parse_attr_ws = 4,   // \0, &, \r, ', ", \n, tab
        ct_space = 8,           // \r, \n, space, tab
        ct_parse_cdata = 16,    // \0, ], >, \r
        ct_parse_comment = 32,  // \0, -, >, \r
        ct_symbol = 64,         // Any symbol > 127, a-z, A-Z, 0-9, _, :, -, .
        ct_start_symbol = 128   // Any symbol > 127, a-z, A-Z, _, :
    };
 
    static const unsigned char chartype_table[256] =
    {
        55,  0,   0,   0,   0,   0,   0,   0,      0,   12,  12,  0,   0,   63,  0,   0,   // 0-15
        0,   0,   0,   0,   0,   0,   0,   0,      0,   0,   0,   0,   0,   0,   0,   0,   // 16-31
        8,   0,   6,   0,   0,   0,   7,   6,      0,   0,   0,   0,   0,   96,  64,  0,   // 32-47
        64,  64,  64,  64,  64,  64,  64,  64,     64,  64,  192, 0,   1,   0,   48,  0,   // 48-63
        0,   192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192, // 64-79
        192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 0,   0,   16,  0,   192, // 80-95
        0,   192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192, // 96-111
        192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 0, 0, 0, 0, 0,           // 112-127
 
        192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192, // 128+
        192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192,
        192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192,
        192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192,
        192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192,
        192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192,
        192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192,
        192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192
    };
 
    enum chartypex_t
    {
        ctx_special_pcdata = 1,   // Any symbol >= 0 and < 32 (except \t, \r, \n), &, <, >
        ctx_special_attr = 2,     // Any symbol >= 0 and < 32 (except \t), &, <, >, "
        ctx_start_symbol = 4,     // Any symbol > 127, a-z, A-Z, _
        ctx_digit = 8,            // 0-9
        ctx_symbol = 16           // Any symbol > 127, a-z, A-Z, 0-9, _, -, .
    };
 
    static const unsigned char chartypex_table[256] =
    {
        3,  3,  3,  3,  3,  3,  3,  3,     3,  0,  2,  3,  3,  2,  3,  3,     // 0-15
        3,  3,  3,  3,  3,  3,  3,  3,     3,  3,  3,  3,  3,  3,  3,  3,     // 16-31
        0,  0,  2,  0,  0,  0,  3,  0,     0,  0,  0,  0,  0, 16, 16,  0,     // 32-47
        24, 24, 24, 24, 24, 24, 24, 24,    24, 24, 0,  0,  3,  0,  3,  0,     // 48-63
 
        0,  20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,    // 64-79
        20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 0,  0,  0,  0,  20,    // 80-95
        0,  20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,    // 96-111
        20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 0,  0,  0,  0,  0,     // 112-127
 
        20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,    // 128+
        20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,
        20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,
        20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,
        20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,
        20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,
        20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,
        20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20
    };
 
#ifdef PUGIXML_WCHAR_MODE
    #define PUGI__IS_CHARTYPE_IMPL(c, ct, table) ((static_cast<unsigned int>(c) < 128 ? table[static_cast<unsigned int>(c)] : table[128]) & (ct))
#else
    #define PUGI__IS_CHARTYPE_IMPL(c, ct, table) (table[static_cast<unsigned char>(c)] & (ct))
#endif
 
    #define PUGI__IS_CHARTYPE(c, ct) PUGI__IS_CHARTYPE_IMPL(c, ct, chartype_table)
    #define PUGI__IS_CHARTYPEX(c, ct) PUGI__IS_CHARTYPE_IMPL(c, ct, chartypex_table)
 
    PUGI__FN bool is_little_endian()
    {
        unsigned int ui = 1;// 定义一个无符号整型变量 `ui`，并将其设置为 1。
 
        return *reinterpret_cast<unsigned char*>(&ui) == 1;  // 检查 `ui` 的最低字节是否为 1，如果是，则当前系统是小端序（little-endian）。
    }
 
    PUGI__FN xml_encoding get_wchar_encoding()
    {
        PUGI__STATIC_ASSERT(sizeof(wchar_t) == 2 || sizeof(wchar_t) == 4);// 静态断言，确保 `wchar_t` 的大小为 2 或 4 字节。
 
        if (sizeof(wchar_t) == 2)// 如果 `wchar_t` 的大小为 2 字节。
            return is_little_endian() ? encoding_utf16_le : encoding_utf16_be;// 根据是否为小端序，返回对应的 UTF-16 编码格式。
        else// 如果 `wchar_t` 的大小为 4 字节。
            return is_little_endian() ? encoding_utf32_le : encoding_utf32_be;// 根据是否为小端序，返回对应的 UTF-32 编码格式。
    }
 
    PUGI__FN bool parse_declaration_encoding(const uint8_t* data, size_t size, const uint8_t*& out_encoding, size_t& out_length)
    {
    #define PUGI__SCANCHAR(ch) { if (offset >= size || data[offset] != ch) return false; offset++; }// 定义一个宏，用于检查当前位置的字符是否等于指定字符 `ch`，如果不匹配则返回 false。
    #define PUGI__SCANCHARTYPE(ct) { while (offset < size && PUGI__IS_CHARTYPE(data[offset], ct)) offset++;// 定义一个宏，用于跳过所有匹配特定字符类型 `ct` 的字符。 }
 
        // check if we have a non-empty XML declaration
        if (size < 6 || !((data[0] == '<') & (data[1] == '?') & (data[2] == 'x') & (data[3] == 'm') & (data[4] == 'l') && PUGI__IS_CHARTYPE(data[5], ct_space)))
            return false;// 如果数据长度小于 6 或者数据前 6 个字符不是合法的 XML 声明格式，返回 false。
 
        // scan XML declaration until the encoding field
        for (size_t i = 6; i + 1 < size; ++i)// 声明中不允许在引号内包含 `?` 字符。
        {
            // declaration can not contain ? in quoted values
            if (data[i] == '?')
                return false;
 
            if (data[i] == 'e' && data[i + 1] == 'n')// 找到可能的 `encoding` 字段开头。
            {
                size_t offset = i;
 
                // encoding follows the version field which can't contain 'en' so this has to be the encoding if XML is well formed
                PUGI__SCANCHAR('e'); PUGI__SCANCHAR('n'); PUGI__SCANCHAR('c'); PUGI__SCANCHAR('o');
                PUGI__SCANCHAR('d'); PUGI__SCANCHAR('i'); PUGI__SCANCHAR('n'); PUGI__SCANCHAR('g');
 
                // S? = S?
                PUGI__SCANCHARTYPE(ct_space);
                PUGI__SCANCHAR('=');
                PUGI__SCANCHARTYPE(ct_space);
 
                // the only two valid delimiters are ' and "
                uint8_t delimiter = (offset < size && data[offset] == '"') ? '"' : '\'';
 
                PUGI__SCANCHAR(delimiter);
 
                size_t start = offset;// 记录编码字段的起始位置。
 
                out_encoding = data + offset;// 设置输出编码字段的起始位置。
 
                PUGI__SCANCHARTYPE(ct_symbol);// 跳过所有符号字符
 
                out_length = offset - start;// 计算编码字段的长度。
 
                PUGI__SCANCHAR(delimiter);// 检查结束分隔符。
 
                return true;// 成功找到并解析编码字段，返回 true。
            }
        }
 
        return false;// 如果未找到 `encoding` 字段，返回 false。
 
    #undef PUGI__SCANCHAR
    #undef PUGI__SCANCHARTYPE
    }
 
    PUGI__FN xml_encoding guess_buffer_encoding(const uint8_t* data, size_t size)
    {// 如果输入缓冲区太小，无法进行编码自动检测，默认为 UTF-8。
        // skip encoding autodetection if input buffer is too small
        if (size < 4) return encoding_utf8;
 
        uint8_t d0 = data[0], d1 = data[1], d2 = data[2], d3 = data[3];// 读取缓冲区前四个字节。
 
        // look for BOM in first few bytes
        if (d0 == 0 && d1 == 0 && d2 == 0xfe && d3 == 0xff) return encoding_utf32_be;// UTF-32 大端序。
        if (d0 == 0xff && d1 == 0xfe && d2 == 0 && d3 == 0) return encoding_utf32_le;// UTF-32 小端序。
        if (d0 == 0xfe && d1 == 0xff) return encoding_utf16_be;// UTF-16 大端序。
        if (d0 == 0xff && d1 == 0xfe) return encoding_utf16_le;// UTF-16 小端序。
        if (d0 == 0xef && d1 == 0xbb && d2 == 0xbf) return encoding_utf8;
 
        // look for <, <? or <?xm in various encodings
        if (d0 == 0 && d1 == 0 && d2 == 0 && d3 == 0x3c) return encoding_utf32_be;// UTF-32 大端序中的 `<`。
        if (d0 == 0x3c && d1 == 0 && d2 == 0 && d3 == 0) return encoding_utf32_le;// UTF-32 小端序中的 `<`。
        if (d0 == 0 && d1 == 0x3c && d2 == 0 && d3 == 0x3f) return encoding_utf16_be; // UTF-16 大端序中的 `<?`。
        if (d0 == 0x3c && d1 == 0 && d2 == 0x3f && d3 == 0) return encoding_utf16_le;// UTF-16 小端序中的 `<?`。
 
        // look for utf16 < followed by node name (this may fail, but is better than utf8 since it's zero terminated so early)
        if (d0 == 0 && d1 == 0x3c) return encoding_utf16_be;// UTF-16 大端序中的 `<`。
        if (d0 == 0x3c && d1 == 0) return encoding_utf16_le;// UTF-16 小端序中的 `<`。
 
        // no known BOM detected; parse declaration
        const uint8_t* enc = 0;// 用于存储编码字段的起始位置。
        size_t enc_length = 0;// 用于存储编码字段的长度。
 
        if (d0 == 0x3c && d1 == 0x3f && d2 == 0x78 && d3 == 0x6d && parse_declaration_encoding(data, size, enc, enc_length))
        {
            // iso-8859-1 (case-insensitive)
            if (enc_length == 10
                && (enc[0] | ' ') == 'i' && (enc[1] | ' ') == 's' && (enc[2] | ' ') == 'o'
                && enc[3] == '-' && enc[4] == '8' && enc[5] == '8' && enc[6] == '5' && enc[7] == '9'
                && enc[8] == '-' && enc[9] == '1')
                return encoding_latin1;
 
            // latin1 (case-insensitive)
            if (enc_length == 6
                && (enc[0] | ' ') == 'l' && (enc[1] | ' ') == 'a' && (enc[2] | ' ') == 't'
                && (enc[3] | ' ') == 'i' && (enc[4] | ' ') == 'n'
                && enc[5] == '1')
                return encoding_latin1;
        }// 如果未匹配到任何已知编码，默认为 UTF-8。
 
        return encoding_utf8;
    }
 
    PUGI__FN xml_encoding get_buffer_encoding(xml_encoding encoding, const void* contents, size_t size)
    {
        // replace wchar encoding with utf implementation
        if (encoding == encoding_wchar) return get_wchar_encoding();
 
        // replace utf16 encoding with utf16 with specific endianness
        if (encoding == encoding_utf16) return is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
 
        // replace utf32 encoding with utf32 with specific endianness
        if (encoding == encoding_utf32) return is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
 
        // only do autodetection if no explicit encoding is requested
        if (encoding != encoding_auto) return encoding;
 
        // try to guess encoding (based on XML specification, Appendix F.1)
        const uint8_t* data = static_cast<const uint8_t*>(contents);
 
        return guess_buffer_encoding(data, size);
    }
 
    PUGI__FN bool get_mutable_buffer(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size, bool is_mutable)
    {
        size_t length = size / sizeof(char_t);
// 计算缓冲区内容的字符长度，字节数除以每个字符的大小。
        if (is_mutable)
        {// 如果缓冲区是可变的，直接使用输入的内容。
            out_buffer = static_cast<char_t*>(const_cast<void*>(contents)); // 将输入的内容指针转换为可修改的 `char_t` 类型指针。
            out_length = length;// 设置输出缓冲区的长度。
        }
        else
        {
            char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t))); // 为缓冲区分配内存，多分配一个字符用于添加终止符。
            if (!buffer) return false;// 如果内存分配失败，返回 false。
 
            if (contents)
                memcpy(buffer, contents, length * sizeof(char_t));// 如果内容不为空，将输入的内容复制到新分配的缓冲区中。
            else
                assert(length == 0);
// 如果内容为空，断言长度为 0，确保数据一致性。
            buffer[length] = 0;
// 在缓冲区的末尾添加终止符。
            out_buffer = buffer; // 将新分配的缓冲区地址赋值给输出指针。
            out_length = length + 1;// 设置输出缓冲区的长度，包含终止符。
        }
 
        return true; // 返回 true，表示操作成功。
    }
 
#ifdef PUGIXML_WCHAR_MODE
    PUGI__FN bool need_endian_swap_utf(xml_encoding le, xml_encoding re)
    {
        return (le == encoding_utf16_be && re == encoding_utf16_le) || (le == encoding_utf16_le && re == encoding_utf16_be) ||
               (le == encoding_utf32_be && re == encoding_utf32_le) || (le == encoding_utf32_le && re == encoding_utf32_be);
    // 判断是否需要进行字节序转换。当源编码和目标编码的字节序不一致时（大端序与小端序互换），需要转换。}
 
    PUGI__FN bool convert_buffer_endian_swap(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size, bool is_mutable)
    {
        const char_t* data = static_cast<const char_t*>(contents);  // 将输入内容指针转换为 `const char_t*` 类型。
        size_t length = size / sizeof(char_t);// 计算缓冲区的字符长度。
 
        if (is_mutable)
        {
            char_t* buffer = const_cast<char_t*>(data); // 将只读指针转换为可修改的指针。
 
            convert_wchar_endian_swap(buffer, data, length);// 对缓冲区进行字节序转换。
 
            out_buffer = buffer;  // 设置输出缓冲区指针。
            out_length = length; // 设置输出缓冲区的长度。
        }
        else
        {// 如果缓冲区不可变，需要分配新的缓冲区进行字节序转换。
            char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t))); // 分配内存，多分配一个字符用于添加终止符。
            if (!buffer) return false; // 如果内存分配失败，返回 false。
 
            convert_wchar_endian_swap(buffer, data, length);// 对新分配的缓冲区进行字节序转换。
            buffer[length] = 0;// 在缓冲区末尾添加终止符。
 
            out_buffer = buffer;// 设置输出缓冲区指针。
            out_length = length + 1; // 设置输出缓冲区的长度，包括终止符。
        }
 
        return true; // 返回 true，表示转换成功。
    }
 
    template <typename D> PUGI__FN bool convert_buffer_generic(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size, D)
    {
        const typename D::type* data = static_cast<const typename D::type*>(contents);
        size_t data_length = size / sizeof(typename D::type);
 
        // first pass: get length in wchar_t units
        size_t length = D::process(data, data_length, 0, wchar_counter());
 
        // allocate buffer of suitable length
        char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
        if (!buffer) return false;
 
        // second pass: convert utf16 input to wchar_t
        wchar_writer::value_type obegin = reinterpret_cast<wchar_writer::value_type>(buffer);
        wchar_writer::value_type oend = D::process(data, data_length, obegin, wchar_writer());
 
        assert(oend == obegin + length);
        *oend = 0;
 
        out_buffer = buffer;
        out_length = length + 1;
 
        return true;
    }
 
    PUGI__FN bool convert_buffer(char_t*& out_buffer, size_t& out_length, xml_encoding encoding, const void* contents, size_t size, bool is_mutable)
    {
        // get native encoding
        xml_encoding wchar_encoding = get_wchar_encoding();
 
        // fast path: no conversion required
        if (encoding == wchar_encoding)
            return get_mutable_buffer(out_buffer, out_length, contents, size, is_mutable);
 
        // only endian-swapping is required
        if (need_endian_swap_utf(encoding, wchar_encoding))
            return convert_buffer_endian_swap(out_buffer, out_length, contents, size, is_mutable);
 
        // source encoding is utf8
        if (encoding == encoding_utf8)
            return convert_buffer_generic(out_buffer, out_length, contents, size, utf8_decoder());
 
        // source encoding is utf16
        if (encoding == encoding_utf16_be || encoding == encoding_utf16_le)
        {
            xml_encoding native_encoding = is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
 
            return (native_encoding == encoding) ?
                convert_buffer_generic(out_buffer, out_length, contents, size, utf16_decoder<opt_false>()) :
                convert_buffer_generic(out_buffer, out_length, contents, size, utf16_decoder<opt_true>());
        }
 
        // source encoding is utf32
        if (encoding == encoding_utf32_be || encoding == encoding_utf32_le)
        {
            xml_encoding native_encoding = is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
 
            return (native_encoding == encoding) ?
                convert_buffer_generic(out_buffer, out_length, contents, size, utf32_decoder<opt_false>()) :
                convert_buffer_generic(out_buffer, out_length, contents, size, utf32_decoder<opt_true>());
        }
 
        // source encoding is latin1
        if (encoding == encoding_latin1)
            return convert_buffer_generic(out_buffer, out_length, contents, size, latin1_decoder());
 
        assert(false && "Invalid encoding"); // unreachable
        return false;
    }
#else
    template <typename D> PUGI__FN bool convert_buffer_generic(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size, D)
    {
        const typename D::type* data = static_cast<const typename D::type*>(contents);
        size_t data_length = size / sizeof(typename D::type);
 
        // first pass: get length in utf8 units
        size_t length = D::process(data, data_length, 0, utf8_counter());
 
        // allocate buffer of suitable length
        char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
        if (!buffer) return false;
 
        // second pass: convert utf16 input to utf8
        uint8_t* obegin = reinterpret_cast<uint8_t*>(buffer);
        uint8_t* oend = D::process(data, data_length, obegin, utf8_writer());
 
        assert(oend == obegin + length);
        *oend = 0;
 
        out_buffer = buffer;
        out_length = length + 1;
 
        return true;
    }
 
    PUGI__FN size_t get_latin1_7bit_prefix_length(const uint8_t* data, size_t size)
    {
        for (size_t i = 0; i < size; ++i)
            if (data[i] > 127)
                return i;
 
        return size;
    }
 
    PUGI__FN bool convert_buffer_latin1(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size, bool is_mutable)
    {
        const uint8_t* data = static_cast<const uint8_t*>(contents);
        size_t data_length = size;
 
        // get size of prefix that does not need utf8 conversion
        size_t prefix_length = get_latin1_7bit_prefix_length(data, data_length);
        assert(prefix_length <= data_length);
 
        const uint8_t* postfix = data + prefix_length;
        size_t postfix_length = data_length - prefix_length;
 
        // if no conversion is needed, just return the original buffer
        if (postfix_length == 0) return get_mutable_buffer(out_buffer, out_length, contents, size, is_mutable);
 
        // first pass: get length in utf8 units
        size_t length = prefix_length + latin1_decoder::process(postfix, postfix_length, 0, utf8_counter());
 
        // allocate buffer of suitable length
        char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
        if (!buffer) return false;
 
        // second pass: convert latin1 input to utf8
        memcpy(buffer, data, prefix_length);
 
        uint8_t* obegin = reinterpret_cast<uint8_t*>(buffer);
        uint8_t* oend = latin1_decoder::process(postfix, postfix_length, obegin + prefix_length, utf8_writer());
 
        assert(oend == obegin + length);
        *oend = 0;
 
        out_buffer = buffer;
        out_length = length + 1;
 
        return true;
    }
 
    PUGI__FN bool convert_buffer(char_t*& out_buffer, size_t& out_length, xml_encoding encoding, const void* contents, size_t size, bool is_mutable)
    {
        // fast path: no conversion required
        if (encoding == encoding_utf8)
            return get_mutable_buffer(out_buffer, out_length, contents, size, is_mutable);
 
        // source encoding is utf16
        if (encoding == encoding_utf16_be || encoding == encoding_utf16_le)
        {
            xml_encoding native_encoding = is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
 
            return (native_encoding == encoding) ?
                convert_buffer_generic(out_buffer, out_length, contents, size, utf16_decoder<opt_false>()) :
                convert_buffer_generic(out_buffer, out_length, contents, size, utf16_decoder<opt_true>());
        }
 
        // source encoding is utf32
        if (encoding == encoding_utf32_be || encoding == encoding_utf32_le)
        {
            xml_encoding native_encoding = is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
 
            return (native_encoding == encoding) ?
                convert_buffer_generic(out_buffer, out_length, contents, size, utf32_decoder<opt_false>()) :
                convert_buffer_generic(out_buffer, out_length, contents, size, utf32_decoder<opt_true>());
        }
 
        // source encoding is latin1
        if (encoding == encoding_latin1)
            return convert_buffer_latin1(out_buffer, out_length, contents, size, is_mutable);
 
        assert(false && "Invalid encoding"); // unreachable
        return false;
    }
#endif
 
    PUGI__FN size_t as_utf8_begin(const wchar_t* str, size_t length)
    {
        // get length in utf8 characters
        return wchar_decoder::process(str, length, 0, utf8_counter());
    }
 
    PUGI__FN void as_utf8_end(char* buffer, size_t size, const wchar_t* str, size_t length)
    {
        // convert to utf8
        uint8_t* begin = reinterpret_cast<uint8_t*>(buffer);
        uint8_t* end = wchar_decoder::process(str, length, begin, utf8_writer());
 
        assert(begin + size == end);
        (void)!end;
        (void)!size;
    }
 
#ifndef PUGIXML_NO_STL
    PUGI__FN std::string as_utf8_impl(const wchar_t* str, size_t length)
    {
        // first pass: get length in utf8 characters
        size_t size = as_utf8_begin(str, length);
 
        // allocate resulting string
        std::string result;
        result.resize(size);
 
        // second pass: convert to utf8
        if (size > 0) as_utf8_end(&result[0], size, str, length);
 
        return result;
    }
 
    PUGI__FN std::basic_string<wchar_t> as_wide_impl(const char* str, size_t size)
    {
        const uint8_t* data = reinterpret_cast<const uint8_t*>(str);
 
        // first pass: get length in wchar_t units
        size_t length = utf8_decoder::process(data, size, 0, wchar_counter());
 
        // allocate resulting string
        std::basic_string<wchar_t> result;
        result.resize(length);
 
        // second pass: convert to wchar_t
        if (length > 0)
        {
            wchar_writer::value_type begin = reinterpret_cast<wchar_writer::value_type>(&result[0]);
            wchar_writer::value_type end = utf8_decoder::process(data, size, begin, wchar_writer());
 
            assert(begin + length == end);
            (void)!end;
        }
 
        return result;
    }
#endif
 
    template <typename Header>
    inline bool strcpy_insitu_allow(size_t length, const Header& header, uintptr_t header_mask, char_t* target)
    {
        // never reuse shared memory
        if (header & xml_memory_page_contents_shared_mask) return false;
 
        size_t target_length = strlength(target);
 
        // always reuse document buffer memory if possible
        if ((header & header_mask) == 0) return target_length >= length;
 
        // reuse heap memory if waste is not too great
        const size_t reuse_threshold = 32;
 
        return target_length >= length && (target_length < reuse_threshold || target_length - length < target_length / 2);
    }
 
    template <typename String, typename Header>
    PUGI__FN bool strcpy_insitu(String& dest, Header& header, uintptr_t header_mask, const char_t* source, size_t source_length)
    {
        if (source_length == 0)
        {
            // empty string and null pointer are equivalent, so just deallocate old memory
            xml_allocator* alloc = PUGI__GETPAGE_IMPL(header)->allocator;
 
            if (header & header_mask) alloc->deallocate_string(dest);
 
            // mark the string as not allocated
            dest = 0;
            header &= ~header_mask;
 
            return true;
        }
        else if (dest && strcpy_insitu_allow(source_length, header, header_mask, dest))
        {
            // we can reuse old buffer, so just copy the new data (including zero terminator)
            memcpy(dest, source, source_length * sizeof(char_t));
            dest[source_length] = 0;
 
            return true;
        }
        else
        {
            xml_allocator* alloc = PUGI__GETPAGE_IMPL(header)->allocator;
 
            if (!alloc->reserve()) return false;
 
            // allocate new buffer
            char_t* buf = alloc->allocate_string(source_length + 1);
            if (!buf) return false;
 
            // copy the string (including zero terminator)
            memcpy(buf, source, source_length * sizeof(char_t));
            buf[source_length] = 0;
 
            // deallocate old buffer (*after* the above to protect against overlapping memory and/or allocation failures)
            if (header & header_mask) alloc->deallocate_string(dest);
 
            // the string is now allocated, so set the flag
            dest = buf;
            header |= header_mask;
 
            return true;
        }
    }
 
    struct gap
    {
        char_t* end;
        size_t size;
 
        gap(): end(0), size(0)
        {
        }
 
        // Push new gap, move s count bytes further (skipping the gap).
        // Collapse previous gap.
        void push(char_t*& s, size_t count)
        {
            if (end) // there was a gap already; collapse it
            {
                // Move [old_gap_end, new_gap_start) to [old_gap_start, ...)
                assert(s >= end);
                memmove(end - size, end, reinterpret_cast<char*>(s) - reinterpret_cast<char*>(end));
            }
 
            s += count; // end of current gap
 
            // "merge" two gaps
            end = s;
            size += count;
        }
 
        // Collapse all gaps, return past-the-end pointer
        char_t* flush(char_t* s)
        {
            if (end)
            {
                // Move [old_gap_end, current_pos) to [old_gap_start, ...)
                assert(s >= end);
                memmove(end - size, end, reinterpret_cast<char*>(s) - reinterpret_cast<char*>(end));
 
                return s - size;
            }
            else return s;
        }
    };
 
    PUGI__FN char_t* strconv_escape(char_t* s, gap& g)
    {
        char_t* stre = s + 1;
 
        switch (*stre)
        {
            case '#':   // &#...
            {
                unsigned int ucsc = 0;
 
                if (stre[1] == 'x') // &#x... (hex code)
                {
                    stre += 2;
 
                    char_t ch = *stre;
 
                    if (ch == ';') return stre;
 
                    for (;;)
                    {
                        if (static_cast<unsigned int>(ch - '0') <= 9)
                            ucsc = 16 * ucsc + (ch - '0');
                        else if (static_cast<unsigned int>((ch | ' ') - 'a') <= 5)
                            ucsc = 16 * ucsc + ((ch | ' ') - 'a' + 10);
                        else if (ch == ';')
                            break;
                        else // cancel
                            return stre;
 
                        ch = *++stre;
                    }
 
                    ++stre;
                }
                else    // &#... (dec code)
                {
                    char_t ch = *++stre;
 
                    if (ch == ';') return stre;
 
                    for (;;)
                    {
                        if (static_cast<unsigned int>(ch - '0') <= 9)
                            ucsc = 10 * ucsc + (ch - '0');
                        else if (ch == ';')
                            break;
                        else // cancel
                            return stre;
 
                        ch = *++stre;
                    }
 
                    ++stre;
                }
 
            #ifdef PUGIXML_WCHAR_MODE
                s = reinterpret_cast<char_t*>(wchar_writer::any(reinterpret_cast<wchar_writer::value_type>(s), ucsc));
            #else
                s = reinterpret_cast<char_t*>(utf8_writer::any(reinterpret_cast<uint8_t*>(s), ucsc));
            #endif
 
                g.push(s, stre - s);
                return stre;
            }
 
            case 'a':   // &a
            {
                ++stre;
 
                if (*stre == 'm') // &am
                {
                    if (*++stre == 'p' && *++stre == ';') // &amp;
                    {
                        *s++ = '&';
                        ++stre;
 
                        g.push(s, stre - s);
                        return stre;
                    }
                }
                else if (*stre == 'p') // &ap
                {
                    if (*++stre == 'o' && *++stre == 's' && *++stre == ';') // &apos;
                    {
                        *s++ = '\'';
                        ++stre;
 
                        g.push(s, stre - s);
                        return stre;
                    }
                }
                break;
            }
 
            case 'g': // &g
            {
                if (*++stre == 't' && *++stre == ';') // &gt;
                {
                    *s++ = '>';
                    ++stre;
 
                    g.push(s, stre - s);
                    return stre;
                }
                break;
            }
 
            case 'l': // &l
            {
                if (*++stre == 't' && *++stre == ';') // &lt;
                {
                    *s++ = '<';
                    ++stre;
 
                    g.push(s, stre - s);
                    return stre;
                }
                break;
            }
 
            case 'q': // &q
            {
                if (*++stre == 'u' && *++stre == 'o' && *++stre == 't' && *++stre == ';') // &quot;
                {
                    *s++ = '"';
                    ++stre;
 
                    g.push(s, stre - s);
                    return stre;
                }
                break;
            }
 
            default:
                break;
        }
 
        return stre;
    }
 
    // Parser utilities
    #define PUGI__ENDSWITH(c, e)        ((c) == (e) || ((c) == 0 && endch == (e)))
    #define PUGI__SKIPWS()              { while (PUGI__IS_CHARTYPE(*s, ct_space)) ++s; }
    #define PUGI__OPTSET(OPT)           ( optmsk & (OPT) )
    #define PUGI__PUSHNODE(TYPE)        { cursor = append_new_node(cursor, *alloc, TYPE); if (!cursor) PUGI__THROW_ERROR(status_out_of_memory, s); }
    #define PUGI__POPNODE()             { cursor = cursor->parent; }
    #define PUGI__SCANFOR(X)            { while (*s != 0 && !(X)) ++s; }
    #define PUGI__SCANWHILE(X)          { while (X) ++s; }
    #define PUGI__SCANWHILE_UNROLL(X)   { for (;;) { char_t ss = s[0]; if (PUGI__UNLIKELY(!(X))) { break; } ss = s[1]; if (PUGI__UNLIKELY(!(X))) { s += 1; break; } ss = s[2]; if (PUGI__UNLIKELY(!(X))) { s += 2; break; } ss = s[3]; if (PUGI__UNLIKELY(!(X))) { s += 3; break; } s += 4; } }
    #define PUGI__ENDSEG()              { ch = *s; *s = 0; ++s; }
    #define PUGI__THROW_ERROR(err, m)   return error_offset = m, error_status = err, static_cast<char_t*>(0)
    #define PUGI__CHECK_ERROR(err, m)   { if (*s == 0) PUGI__THROW_ERROR(err, m); }
 
    PUGI__FN char_t* strconv_comment(char_t* s, char_t endch)
    {
        gap g;
 
        while (true)
        {
            PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_comment));
 
            if (*s == '\r') // Either a single 0x0d or 0x0d 0x0a pair
            {
                *s++ = '\n'; // replace first one with 0x0a
 
                if (*s == '\n') g.push(s, 1);
            }
            else if (s[0] == '-' && s[1] == '-' && PUGI__ENDSWITH(s[2], '>')) // comment ends here
            {
                *g.flush(s) = 0;
 
                return s + (s[2] == '>' ? 3 : 2);
            }
            else if (*s == 0)
            {
                return 0;
            }
            else ++s;
        }
    }
 
    PUGI__FN char_t* strconv_cdata(char_t* s, char_t endch)
    {
        gap g;
 
        while (true)
        {
            PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_cdata));
 
            if (*s == '\r') // Either a single 0x0d or 0x0d 0x0a pair
            {
                *s++ = '\n'; // replace first one with 0x0a
 
                if (*s == '\n') g.push(s, 1);
            }
            else if (s[0] == ']' && s[1] == ']' && PUGI__ENDSWITH(s[2], '>')) // CDATA ends here
            {
                *g.flush(s) = 0;
 
                return s + 1;
            }
            else if (*s == 0)
            {
                return 0;
            }
            else ++s;
        }
    }
 
    typedef char_t* (*strconv_pcdata_t)(char_t*);
 
    template <typename opt_trim, typename opt_eol, typename opt_escape> struct strconv_pcdata_impl
    {
        static char_t* parse(char_t* s)
        {
            gap g;
 
            char_t* begin = s;
 
            while (true)
            {
                PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_pcdata));
 
                if (*s == '<') // PCDATA ends here
                {
                    char_t* end = g.flush(s);
 
                    if (opt_trim::value)
                        while (end > begin && PUGI__IS_CHARTYPE(end[-1], ct_space))
                            --end;
 
                    *end = 0;
 
                    return s + 1;
                }
                else if (opt_eol::value && *s == '\r') // Either a single 0x0d or 0x0d 0x0a pair
                {
                    *s++ = '\n'; // replace first one with 0x0a
 
                    if (*s == '\n') g.push(s, 1);
                }
                else if (opt_escape::value && *s == '&')
                {
                    s = strconv_escape(s, g);
                }
                else if (*s == 0)
                {
                    char_t* end = g.flush(s);
 
                    if (opt_trim::value)
                        while (end > begin && PUGI__IS_CHARTYPE(end[-1], ct_space))
                            --end;
 
                    *end = 0;
 
                    return s;
                }
                else ++s;
            }
        }
    };
 
    PUGI__FN strconv_pcdata_t get_strconv_pcdata(unsigned int optmask)
    {
        PUGI__STATIC_ASSERT(parse_escapes == 0x10 && parse_eol == 0x20 && parse_trim_pcdata == 0x0800);
 
        switch (((optmask >> 4) & 3) | ((optmask >> 9) & 4)) // get bitmask for flags (eol escapes trim)
        {
        case 0: return strconv_pcdata_impl<opt_false, opt_false, opt_false>::parse;
        case 1: return strconv_pcdata_impl<opt_false, opt_false, opt_true>::parse;
        case 2: return strconv_pcdata_impl<opt_false, opt_true, opt_false>::parse;
        case 3: return strconv_pcdata_impl<opt_false, opt_true, opt_true>::parse;
        case 4: return strconv_pcdata_impl<opt_true, opt_false, opt_false>::parse;
        case 5: return strconv_pcdata_impl<opt_true, opt_false, opt_true>::parse;
        case 6: return strconv_pcdata_impl<opt_true, opt_true, opt_false>::parse;
        case 7: return strconv_pcdata_impl<opt_true, opt_true, opt_true>::parse;
        default: assert(false); return 0; // unreachable
        }
    }
 
    typedef char_t* (*strconv_attribute_t)(char_t*, char_t);
 
    template <typename opt_escape> struct strconv_attribute_impl
    {
        static char_t* parse_wnorm(char_t* s, char_t end_quote)
        {
            gap g;
 
            // trim leading whitespaces
            if (PUGI__IS_CHARTYPE(*s, ct_space))
            {
                char_t* str = s;
 
                do ++str;
                while (PUGI__IS_CHARTYPE(*str, ct_space));
 
                g.push(s, str - s);
            }
 
            while (true)
            {
                PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_attr_ws | ct_space));
 
                if (*s == end_quote)
                {
                    char_t* str = g.flush(s);
 
                    do *str-- = 0;
                    while (PUGI__IS_CHARTYPE(*str, ct_space));
 
                    return s + 1;
                }
                else if (PUGI__IS_CHARTYPE(*s, ct_space))
                {
                    *s++ = ' ';
 
                    if (PUGI__IS_CHARTYPE(*s, ct_space))
                    {
                        char_t* str = s + 1;
                        while (PUGI__IS_CHARTYPE(*str, ct_space)) ++str;
 
                        g.push(s, str - s);
                    }
                }
                else if (opt_escape::value && *s == '&')
                {
                    s = strconv_escape(s, g);
                }
                else if (!*s)
                {
                    return 0;
                }
                else ++s;
            }
        }
 
        static char_t* parse_wconv(char_t* s, char_t end_quote)
        {
            gap g;
 
            while (true)
            {
                PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_attr_ws));
 
                if (*s == end_quote)
                {
                    *g.flush(s) = 0;
 
                    return s + 1;
                }
                else if (PUGI__IS_CHARTYPE(*s, ct_space))
                {
                    if (*s == '\r')
                    {
                        *s++ = ' ';
 
                        if (*s == '\n') g.push(s, 1);
                    }
                    else *s++ = ' ';
                }
                else if (opt_escape::value && *s == '&')
                {
                    s = strconv_escape(s, g);
                }
                else if (!*s)
                {
                    return 0;
                }
                else ++s;
            }
        }
 
        static char_t* parse_eol(char_t* s, char_t end_quote)
        {
            gap g;
 
            while (true)
            {
                PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_attr));
 
                if (*s == end_quote)
                {
                    *g.flush(s) = 0;
 
                    return s + 1;
                }
                else if (*s == '\r')
                {
                    *s++ = '\n';
 
                    if (*s == '\n') g.push(s, 1);
                }
                else if (opt_escape::value && *s == '&')
                {
                    s = strconv_escape(s, g);
                }
                else if (!*s)
                {
                    return 0;
                }
                else ++s;
            }
        }
 
        static char_t* parse_simple(char_t* s, char_t end_quote)
        {
            gap g;
 
            while (true)
            {
                PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_attr));
 
                if (*s == end_quote)
                {
                    *g.flush(s) = 0;
 
                    return s + 1;
                }
                else if (opt_escape::value && *s == '&')
                {
                    s = strconv_escape(s, g);
                }
                else if (!*s)
                {
                    return 0;
                }
                else ++s;
            }
        }
    };
 
    PUGI__FN strconv_attribute_t get_strconv_attribute(unsigned int optmask)
    {
        PUGI__STATIC_ASSERT(parse_escapes == 0x10 && parse_eol == 0x20 && parse_wconv_attribute == 0x40 && parse_wnorm_attribute == 0x80);
 
        switch ((optmask >> 4) & 15) // get bitmask for flags (wconv wnorm eol escapes)
        {
        case 0:  return strconv_attribute_impl<opt_false>::parse_simple;
        case 1:  return strconv_attribute_impl<opt_true>::parse_simple;
        case 2:  return strconv_attribute_impl<opt_false>::parse_eol;
        case 3:  return strconv_attribute_impl<opt_true>::parse_eol;
        case 4:  return strconv_attribute_impl<opt_false>::parse_wconv;
        case 5:  return strconv_attribute_impl<opt_true>::parse_wconv;
        case 6:  return strconv_attribute_impl<opt_false>::parse_wconv;
        case 7:  return strconv_attribute_impl<opt_true>::parse_wconv;
        case 8:  return strconv_attribute_impl<opt_false>::parse_wnorm;
        case 9:  return strconv_attribute_impl<opt_true>::parse_wnorm;
        case 10: return strconv_attribute_impl<opt_false>::parse_wnorm;
        case 11: return strconv_attribute_impl<opt_true>::parse_wnorm;
        case 12: return strconv_attribute_impl<opt_false>::parse_wnorm;
        case 13: return strconv_attribute_impl<opt_true>::parse_wnorm;
        case 14: return strconv_attribute_impl<opt_false>::parse_wnorm;
        case 15: return strconv_attribute_impl<opt_true>::parse_wnorm;
        default: assert(false); return 0; // unreachable
        }
    }
 
    inline xml_parse_result make_parse_result(xml_parse_status status, ptrdiff_t offset = 0)
    {
        xml_parse_result result;
        result.status = status;
        result.offset = offset;
 
        return result;
    }
 
    struct xml_parser
    {
        xml_allocator* alloc;
        char_t* error_offset;
        xml_parse_status error_status;
 
        xml_parser(xml_allocator* alloc_): alloc(alloc_), error_offset(0), error_status(status_ok)
        {
        }
 
        // DOCTYPE consists of nested sections of the following possible types:
        // <!-- ... -->, <? ... ?>, "...", '...'
        // <![...]]>
        // <!...>
        // First group can not contain nested groups
        // Second group can contain nested groups of the same type
        // Third group can contain all other groups
        char_t* parse_doctype_primitive(char_t* s)
        {
            if (*s == '"' || *s == '\'')
            {
                // quoted string
                char_t ch = *s++;
                PUGI__SCANFOR(*s == ch);
                if (!*s) PUGI__THROW_ERROR(status_bad_doctype, s);
 
                s++;
            }
            else if (s[0] == '<' && s[1] == '?')
            {
                // <? ... ?>
                s += 2;
                PUGI__SCANFOR(s[0] == '?' && s[1] == '>'); // no need for ENDSWITH because ?> can't terminate proper doctype
                if (!*s) PUGI__THROW_ERROR(status_bad_doctype, s);
 
                s += 2;
            }
            else if (s[0] == '<' && s[1] == '!' && s[2] == '-' && s[3] == '-')
            {
                s += 4;
                PUGI__SCANFOR(s[0] == '-' && s[1] == '-' && s[2] == '>'); // no need for ENDSWITH because --> can't terminate proper doctype
                if (!*s) PUGI__THROW_ERROR(status_bad_doctype, s);
 
                s += 3;
            }
            else PUGI__THROW_ERROR(status_bad_doctype, s);
 
            return s;
        }
 
        char_t* parse_doctype_ignore(char_t* s)
        {
            size_t depth = 0;
 
            assert(s[0] == '<' && s[1] == '!' && s[2] == '[');
            s += 3;
 
            while (*s)
            {
                if (s[0] == '<' && s[1] == '!' && s[2] == '[')
                {
                    // nested ignore section
                    s += 3;
                    depth++;
                }
                else if (s[0] == ']' && s[1] == ']' && s[2] == '>')
                {
                    // ignore section end
                    s += 3;
 
                    if (depth == 0)
                        return s;
 
                    depth--;
                }
                else s++;
            }
 
            PUGI__THROW_ERROR(status_bad_doctype, s);
        }
 
        char_t* parse_doctype_group(char_t* s, char_t endch)
        {
            size_t depth = 0;
 
            assert((s[0] == '<' || s[0] == 0) && s[1] == '!');
            s += 2;
 
            while (*s)
            {
                if (s[0] == '<' && s[1] == '!' && s[2] != '-')
                {
                    if (s[2] == '[')
                    {
                        // ignore
                        s = parse_doctype_ignore(s);
                        if (!s) return s;
                    }
                    else
                    {
                        // some control group
                        s += 2;
                        depth++;
                    }
                }
                else if (s[0] == '<' || s[0] == '"' || s[0] == '\'')
                {
                    // unknown tag (forbidden), or some primitive group
                    s = parse_doctype_primitive(s);
                    if (!s) return s;
                }
                else if (*s == '>')
                {
                    if (depth == 0)
                        return s;
 
                    depth--;
                    s++;
                }
                else s++;
            }
 
            if (depth != 0 || endch != '>') PUGI__THROW_ERROR(status_bad_doctype, s);
 
            return s;
        }
 
        char_t* parse_exclamation(char_t* s, xml_node_struct* cursor, unsigned int optmsk, char_t endch)
        {
            // parse node contents, starting with exclamation mark
            ++s;
 
            if (*s == '-') // '<!-...'
            {
                ++s;
 
                if (*s == '-') // '<!--...'
                {
                    ++s;
 
                    if (PUGI__OPTSET(parse_comments))
                    {
                        PUGI__PUSHNODE(node_comment); // Append a new node on the tree.
                        cursor->value = s; // Save the offset.
                    }
 
                    if (PUGI__OPTSET(parse_eol) && PUGI__OPTSET(parse_comments))
                    {
                        s = strconv_comment(s, endch);
 
                        if (!s) PUGI__THROW_ERROR(status_bad_comment, cursor->value);
                    }
                    else
                    {
                        // Scan for terminating '-->'.
                        PUGI__SCANFOR(s[0] == '-' && s[1] == '-' && PUGI__ENDSWITH(s[2], '>'));
                        PUGI__CHECK_ERROR(status_bad_comment, s);
 
                        if (PUGI__OPTSET(parse_comments))
                            *s = 0; // Zero-terminate this segment at the first terminating '-'.
 
                        s += (s[2] == '>' ? 3 : 2); // Step over the '\0->'.
                    }
                }
                else PUGI__THROW_ERROR(status_bad_comment, s);
            }
            else if (*s == '[')
            {
                // '<![CDATA[...'
                if (*++s=='C' && *++s=='D' && *++s=='A' && *++s=='T' && *++s=='A' && *++s == '[')
                {
                    ++s;
 
                    if (PUGI__OPTSET(parse_cdata))
                    {
                        PUGI__PUSHNODE(node_cdata); // Append a new node on the tree.
                        cursor->value = s; // Save the offset.
 
                        if (PUGI__OPTSET(parse_eol))
                        {
                            s = strconv_cdata(s, endch);
 
                            if (!s) PUGI__THROW_ERROR(status_bad_cdata, cursor->value);
                        }
                        else
                        {
                            // Scan for terminating ']]>'.
                            PUGI__SCANFOR(s[0] == ']' && s[1] == ']' && PUGI__ENDSWITH(s[2], '>'));
                            PUGI__CHECK_ERROR(status_bad_cdata, s);
 
                            *s++ = 0; // Zero-terminate this segment.
                        }
                    }
                    else // Flagged for discard, but we still have to scan for the terminator.
                    {
                        // Scan for terminating ']]>'.
                        PUGI__SCANFOR(s[0] == ']' && s[1] == ']' && PUGI__ENDSWITH(s[2], '>'));
                        PUGI__CHECK_ERROR(status_bad_cdata, s);
 
                        ++s;
                    }
 
                    s += (s[1] == '>' ? 2 : 1); // Step over the last ']>'.
                }
                else PUGI__THROW_ERROR(status_bad_cdata, s);
            }
            else if (s[0] == 'D' && s[1] == 'O' && s[2] == 'C' && s[3] == 'T' && s[4] == 'Y' && s[5] == 'P' && PUGI__ENDSWITH(s[6], 'E'))
            {
                s -= 2;
 
                if (cursor->parent) PUGI__THROW_ERROR(status_bad_doctype, s);
 
                char_t* mark = s + 9;
 
                s = parse_doctype_group(s, endch);
                if (!s) return s;
 
                assert((*s == 0 && endch == '>') || *s == '>');
                if (*s) *s++ = 0;
 
                if (PUGI__OPTSET(parse_doctype))
                {
                    while (PUGI__IS_CHARTYPE(*mark, ct_space)) ++mark;
 
                    PUGI__PUSHNODE(node_doctype);
 
                    cursor->value = mark;
                }
            }
            else if (*s == 0 && endch == '-') PUGI__THROW_ERROR(status_bad_comment, s);
            else if (*s == 0 && endch == '[') PUGI__THROW_ERROR(status_bad_cdata, s);
            else PUGI__THROW_ERROR(status_unrecognized_tag, s);
 
            return s;
        }
 
        char_t* parse_question(char_t* s, xml_node_struct*& ref_cursor, unsigned int optmsk, char_t endch)
        {
            // load into registers
            xml_node_struct* cursor = ref_cursor;
            char_t ch = 0;
 
            // parse node contents, starting with question mark
            ++s;
 
            // read PI target
            char_t* target = s;
 
            if (!PUGI__IS_CHARTYPE(*s, ct_start_symbol)) PUGI__THROW_ERROR(status_bad_pi, s);
 
            PUGI__SCANWHILE(PUGI__IS_CHARTYPE(*s, ct_symbol));
            PUGI__CHECK_ERROR(status_bad_pi, s);
 
            // determine node type; stricmp / strcasecmp is not portable
            bool declaration = (target[0] | ' ') == 'x' && (target[1] | ' ') == 'm' && (target[2] | ' ') == 'l' && target + 3 == s;
 
            if (declaration ? PUGI__OPTSET(parse_declaration) : PUGI__OPTSET(parse_pi))
            {
                if (declaration)
                {
                    // disallow non top-level declarations
                    if (cursor->parent) PUGI__THROW_ERROR(status_bad_pi, s);
 
                    PUGI__PUSHNODE(node_declaration);
                }
                else
                {
                    PUGI__PUSHNODE(node_pi);
                }
 
                cursor->name = target;
 
                PUGI__ENDSEG();
 
                // parse value/attributes
                if (ch == '?')
                {
                    // empty node
                    if (!PUGI__ENDSWITH(*s, '>')) PUGI__THROW_ERROR(status_bad_pi, s);
                    s += (*s == '>');
 
                    PUGI__POPNODE();
                }
                else if (PUGI__IS_CHARTYPE(ch, ct_space))
                {
                    PUGI__SKIPWS();
 
                    // scan for tag end
                    char_t* value = s;
 
                    PUGI__SCANFOR(s[0] == '?' && PUGI__ENDSWITH(s[1], '>'));
                    PUGI__CHECK_ERROR(status_bad_pi, s);
 
                    if (declaration)
                    {
                        // replace ending ? with / so that 'element' terminates properly
                        *s = '/';
 
                        // we exit from this function with cursor at node_declaration, which is a signal to parse() to go to LOC_ATTRIBUTES
                        s = value;
                    }
                    else
                    {
                        // store value and step over >
                        cursor->value = value;
 
                        PUGI__POPNODE();
 
                        PUGI__ENDSEG();
 
                        s += (*s == '>');
                    }
                }
                else PUGI__THROW_ERROR(status_bad_pi, s);
            }
            else
            {
                // scan for tag end
                PUGI__SCANFOR(s[0] == '?' && PUGI__ENDSWITH(s[1], '>'));
                PUGI__CHECK_ERROR(status_bad_pi, s);
 
                s += (s[1] == '>' ? 2 : 1);
            }
 
            // store from registers
            ref_cursor = cursor;
 
            return s;
        }
 
        char_t* parse_tree(char_t* s, xml_node_struct* root, unsigned int optmsk, char_t endch)
        {
            strconv_attribute_t strconv_attribute = get_strconv_attribute(optmsk);
            strconv_pcdata_t strconv_pcdata = get_strconv_pcdata(optmsk);
 
            char_t ch = 0;
            xml_node_struct* cursor = root;
            char_t* mark = s;
 
            while (*s != 0)
            {
                if (*s == '<')
                {
                    ++s;
 
                LOC_TAG:
                    if (PUGI__IS_CHARTYPE(*s, ct_start_symbol)) // '<#...'
                    {
                        PUGI__PUSHNODE(node_element); // Append a new node to the tree.
 
                        cursor->name = s;
 
                        PUGI__SCANWHILE_UNROLL(PUGI__IS_CHARTYPE(ss, ct_symbol)); // Scan for a terminator.
                        PUGI__ENDSEG(); // Save char in 'ch', terminate & step over.
 
                        if (ch == '>')
                        {
                            // end of tag
                        }
                        else if (PUGI__IS_CHARTYPE(ch, ct_space))
                        {
                        LOC_ATTRIBUTES:
                            while (true)
                            {
                                PUGI__SKIPWS(); // Eat any whitespace.
 
                                if (PUGI__IS_CHARTYPE(*s, ct_start_symbol)) // <... #...
                                {
                                    xml_attribute_struct* a = append_new_attribute(cursor, *alloc); // Make space for this attribute.
                                    if (!a) PUGI__THROW_ERROR(status_out_of_memory, s);
 
                                    a->name = s; // Save the offset.
 
                                    PUGI__SCANWHILE_UNROLL(PUGI__IS_CHARTYPE(ss, ct_symbol)); // Scan for a terminator.
                                    PUGI__ENDSEG(); // Save char in 'ch', terminate & step over.
 
                                    if (PUGI__IS_CHARTYPE(ch, ct_space))
                                    {
                                        PUGI__SKIPWS(); // Eat any whitespace.
 
                                        ch = *s;
                                        ++s;
                                    }
 
                                    if (ch == '=') // '<... #=...'
                                    {
                                        PUGI__SKIPWS(); // Eat any whitespace.
 
                                        if (*s == '"' || *s == '\'') // '<... #="...'
                                        {
                                            ch = *s; // Save quote char to avoid breaking on "''" -or- '""'.
                                            ++s; // Step over the quote.
                                            a->value = s; // Save the offset.
 
                                            s = strconv_attribute(s, ch);
 
                                            if (!s) PUGI__THROW_ERROR(status_bad_attribute, a->value);
 
                                            // After this line the loop continues from the start;
                                            // Whitespaces, / and > are ok, symbols and EOF are wrong,
                                            // everything else will be detected
                                            if (PUGI__IS_CHARTYPE(*s, ct_start_symbol)) PUGI__THROW_ERROR(status_bad_attribute, s);
                                        }
                                        else PUGI__THROW_ERROR(status_bad_attribute, s);
                                    }
                                    else PUGI__THROW_ERROR(status_bad_attribute, s);
                                }
                                else if (*s == '/')
                                {
                                    ++s;
 
                                    if (*s == '>')
                                    {
                                        PUGI__POPNODE();
                                        s++;
                                        break;
                                    }
                                    else if (*s == 0 && endch == '>')
                                    {
                                        PUGI__POPNODE();
                                        break;
                                    }
                                    else PUGI__THROW_ERROR(status_bad_start_element, s);
                                }
                                else if (*s == '>')
                                {
                                    ++s;
 
                                    break;
                                }
                                else if (*s == 0 && endch == '>')
                                {
                                    break;
                                }
                                else PUGI__THROW_ERROR(status_bad_start_element, s);
                            }
 
                            // !!!
                        }
                        else if (ch == '/') // '<#.../'
                        {
                            if (!PUGI__ENDSWITH(*s, '>')) PUGI__THROW_ERROR(status_bad_start_element, s);
 
                            PUGI__POPNODE(); // Pop.
 
                            s += (*s == '>');
                        }
                        else if (ch == 0)
                        {
                            // we stepped over null terminator, backtrack & handle closing tag
                            --s;
 
                            if (endch != '>') PUGI__THROW_ERROR(status_bad_start_element, s);
                        }
                        else PUGI__THROW_ERROR(status_bad_start_element, s);
                    }
                    else if (*s == '/')
                    {
                        ++s;
 
                        mark = s;
 
                        char_t* name = cursor->name;
                        if (!name) PUGI__THROW_ERROR(status_end_element_mismatch, mark);
 
                        while (PUGI__IS_CHARTYPE(*s, ct_symbol))
                        {
                            if (*s++ != *name++) PUGI__THROW_ERROR(status_end_element_mismatch, mark);
                        }
 
                        if (*name)
                        {
                            if (*s == 0 && name[0] == endch && name[1] == 0) PUGI__THROW_ERROR(status_bad_end_element, s);
                            else PUGI__THROW_ERROR(status_end_element_mismatch, mark);
                        }
 
                        PUGI__POPNODE(); // Pop.
 
                        PUGI__SKIPWS();
 
                        if (*s == 0)
                        {
                            if (endch != '>') PUGI__THROW_ERROR(status_bad_end_element, s);
                        }
                        else
                        {
                            if (*s != '>') PUGI__THROW_ERROR(status_bad_end_element, s);
                            ++s;
                        }
                    }
                    else if (*s == '?') // '<?...'
                    {
                        s = parse_question(s, cursor, optmsk, endch);
                        if (!s) return s;
 
                        assert(cursor);
                        if (PUGI__NODETYPE(cursor) == node_declaration) goto LOC_ATTRIBUTES;
                    }
                    else if (*s == '!') // '<!...'
                    {
                        s = parse_exclamation(s, cursor, optmsk, endch);
                        if (!s) return s;
                    }
                    else if (*s == 0 && endch == '?') PUGI__THROW_ERROR(status_bad_pi, s);
                    else PUGI__THROW_ERROR(status_unrecognized_tag, s);
                }
                else
                {
                    mark = s; // Save this offset while searching for a terminator.
 
                    PUGI__SKIPWS(); // Eat whitespace if no genuine PCDATA here.
 
                    if (*s == '<' || !*s)
                    {
                        // We skipped some whitespace characters because otherwise we would take the tag branch instead of PCDATA one
                        assert(mark != s);
 
                        if (!PUGI__OPTSET(parse_ws_pcdata | parse_ws_pcdata_single) || PUGI__OPTSET(parse_trim_pcdata))
                        {
                            continue;
                        }
                        else if (PUGI__OPTSET(parse_ws_pcdata_single))
                        {
                            if (s[0] != '<' || s[1] != '/' || cursor->first_child) continue;
                        }
                    }
 
                    if (!PUGI__OPTSET(parse_trim_pcdata))
                        s = mark;
 
                    if (cursor->parent || PUGI__OPTSET(parse_fragment))
                    {
                        if (PUGI__OPTSET(parse_embed_pcdata) && cursor->parent && !cursor->first_child && !cursor->value)
                        {
                            cursor->value = s; // Save the offset.
                        }
                        else
                        {
                            PUGI__PUSHNODE(node_pcdata); // Append a new node on the tree.
 
                            cursor->value = s; // Save the offset.
 
                            PUGI__POPNODE(); // Pop since this is a standalone.
                        }
 
                        s = strconv_pcdata(s);
 
                        if (!*s) break;
                    }
                    else
                    {
                        PUGI__SCANFOR(*s == '<'); // '...<'
                        if (!*s) break;
 
                        ++s;
                    }
 
                    // We're after '<'
                    goto LOC_TAG;
                }
            }
 
            // check that last tag is closed
            if (cursor != root) PUGI__THROW_ERROR(status_end_element_mismatch, s);
 
            return s;
        }
 
    #ifdef PUGIXML_WCHAR_MODE
        static char_t* parse_skip_bom(char_t* s)
        {
            unsigned int bom = 0xfeff;
            return (s[0] == static_cast<wchar_t>(bom)) ? s + 1 : s;
        }
    #else
        static char_t* parse_skip_bom(char_t* s)
        {
            return (s[0] == '\xef' && s[1] == '\xbb' && s[2] == '\xbf') ? s + 3 : s;
        }
    #endif
 
        static bool has_element_node_siblings(xml_node_struct* node)
        {
            while (node)
            {
                if (PUGI__NODETYPE(node) == node_element) return true;
 
                node = node->next_sibling;
            }
 
            return false;
        }
 
        static xml_parse_result parse(char_t* buffer, size_t length, xml_document_struct* xmldoc, xml_node_struct* root, unsigned int optmsk)
        {
            // early-out for empty documents
            if (length == 0)
                return make_parse_result(PUGI__OPTSET(parse_fragment) ? status_ok : status_no_document_element);
 
            // get last child of the root before parsing
            xml_node_struct* last_root_child = root->first_child ? root->first_child->prev_sibling_c + 0 : 0;
 
            // create parser on stack
            xml_parser parser(static_cast<xml_allocator*>(xmldoc));
 
            // save last character and make buffer zero-terminated (speeds up parsing)
            char_t endch = buffer[length - 1];
            buffer[length - 1] = 0;
 
            // skip BOM to make sure it does not end up as part of parse output
            char_t* buffer_data = parse_skip_bom(buffer);
 
            // perform actual parsing
            parser.parse_tree(buffer_data, root, optmsk, endch);
 
            xml_parse_result result = make_parse_result(parser.error_status, parser.error_offset ? parser.error_offset - buffer : 0);
            assert(result.offset >= 0 && static_cast<size_t>(result.offset) <= length);
 
            if (result)
            {
                // since we removed last character, we have to handle the only possible false positive (stray <)
                if (endch == '<')
                    return make_parse_result(status_unrecognized_tag, length - 1);
 
                // check if there are any element nodes parsed
                xml_node_struct* first_root_child_parsed = last_root_child ? last_root_child->next_sibling + 0 : root->first_child+ 0;
 
                if (!PUGI__OPTSET(parse_fragment) && !has_element_node_siblings(first_root_child_parsed))
                    return make_parse_result(status_no_document_element, length - 1);
            }
            else
            {
                // roll back offset if it occurs on a null terminator in the source buffer
                if (result.offset > 0 && static_cast<size_t>(result.offset) == length - 1 && endch == 0)
                    result.offset--;
            }
 
            return result;
        }
    };
 
    // Output facilities
    PUGI__FN xml_encoding get_write_native_encoding()
    {
    #ifdef PUGIXML_WCHAR_MODE
        return get_wchar_encoding();
    #else
        return encoding_utf8;
    #endif
    }
 
    PUGI__FN xml_encoding get_write_encoding(xml_encoding encoding)
    {
        // replace wchar encoding with utf implementation
        if (encoding == encoding_wchar) return get_wchar_encoding();
 
        // replace utf16 encoding with utf16 with specific endianness
        if (encoding == encoding_utf16) return is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
 
        // replace utf32 encoding with utf32 with specific endianness
        if (encoding == encoding_utf32) return is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
 
        // only do autodetection if no explicit encoding is requested
        if (encoding != encoding_auto) return encoding;
 
        // assume utf8 encoding
        return encoding_utf8;
    }
 
    template <typename D, typename T> PUGI__FN size_t convert_buffer_output_generic(typename T::value_type dest, const char_t* data, size_t length, D, T)
    {
        PUGI__STATIC_ASSERT(sizeof(char_t) == sizeof(typename D::type));
 
        typename T::value_type end = D::process(reinterpret_cast<const typename D::type*>(data), length, dest, T());
 
        return static_cast<size_t>(end - dest) * sizeof(*dest);
    }
 
    template <typename D, typename T> PUGI__FN size_t convert_buffer_output_generic(typename T::value_type dest, const char_t* data, size_t length, D, T, bool opt_swap)
    {
        PUGI__STATIC_ASSERT(sizeof(char_t) == sizeof(typename D::type));
 
        typename T::value_type end = D::process(reinterpret_cast<const typename D::type*>(data), length, dest, T());
 
        if (opt_swap)
        {
            for (typename T::value_type i = dest; i != end; ++i)
                *i = endian_swap(*i);
        }
 
        return static_cast<size_t>(end - dest) * sizeof(*dest);
    }
 
#ifdef PUGIXML_WCHAR_MODE
    PUGI__FN size_t get_valid_length(const char_t* data, size_t length)
    {
        if (length < 1) return 0;
 
        // discard last character if it's the lead of a surrogate pair
        return (sizeof(wchar_t) == 2 && static_cast<unsigned int>(static_cast<uint16_t>(data[length - 1]) - 0xD800) < 0x400) ? length - 1 : length;
    }
 
    PUGI__FN size_t convert_buffer_output(char_t* r_char, uint8_t* r_u8, uint16_t* r_u16, uint32_t* r_u32, const char_t* data, size_t length, xml_encoding encoding)
    {
        // only endian-swapping is required
        if (need_endian_swap_utf(encoding, get_wchar_encoding()))
        {
            convert_wchar_endian_swap(r_char, data, length);
 
            return length * sizeof(char_t);
        }
 
        // convert to utf8
        if (encoding == encoding_utf8)
            return convert_buffer_output_generic(r_u8, data, length, wchar_decoder(), utf8_writer());
 
        // convert to utf16
        if (encoding == encoding_utf16_be || encoding == encoding_utf16_le)
        {
            xml_encoding native_encoding = is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
 
            return convert_buffer_output_generic(r_u16, data, length, wchar_decoder(), utf16_writer(), native_encoding != encoding);
        }
 
        // convert to utf32
        if (encoding == encoding_utf32_be || encoding == encoding_utf32_le)
        {
            xml_encoding native_encoding = is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
 
            return convert_buffer_output_generic(r_u32, data, length, wchar_decoder(), utf32_writer(), native_encoding != encoding);
        }
 
        // convert to latin1
        if (encoding == encoding_latin1)
            return convert_buffer_output_generic(r_u8, data, length, wchar_decoder(), latin1_writer());
 
        assert(false && "Invalid encoding"); // unreachable
        return 0;
    }
#else
    PUGI__FN size_t get_valid_length(const char_t* data, size_t length)
    {
        if (length < 5) return 0;
 
        for (size_t i = 1; i <= 4; ++i)
        {
            uint8_t ch = static_cast<uint8_t>(data[length - i]);
 
            // either a standalone character or a leading one
            if ((ch & 0xc0) != 0x80) return length - i;
        }
 
        // there are four non-leading characters at the end, sequence tail is broken so might as well process the whole chunk
        return length;
    }
 
    PUGI__FN size_t convert_buffer_output(char_t* /* r_char */, uint8_t* r_u8, uint16_t* r_u16, uint32_t* r_u32, const char_t* data, size_t length, xml_encoding encoding)
    {
        if (encoding == encoding_utf16_be || encoding == encoding_utf16_le)
        {
            xml_encoding native_encoding = is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
 
            return convert_buffer_output_generic(r_u16, data, length, utf8_decoder(), utf16_writer(), native_encoding != encoding);
        }
 
        if (encoding == encoding_utf32_be || encoding == encoding_utf32_le)
        {
            xml_encoding native_encoding = is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
 
            return convert_buffer_output_generic(r_u32, data, length, utf8_decoder(), utf32_writer(), native_encoding != encoding);
        }
 
        if (encoding == encoding_latin1)
            return convert_buffer_output_generic(r_u8, data, length, utf8_decoder(), latin1_writer());
 
        assert(false && "Invalid encoding"); // unreachable
        return 0;
    }
#endif
 
    class xml_buffered_writer
    {
        xml_buffered_writer(const xml_buffered_writer&);
        xml_buffered_writer& operator=(const xml_buffered_writer&);
 
    public:
        xml_buffered_writer(xml_writer& writer_, xml_encoding user_encoding): writer(writer_), bufsize(0), encoding(get_write_encoding(user_encoding))
        {
            PUGI__STATIC_ASSERT(bufcapacity >= 8);
        }
 
        size_t flush()
        {
            flush(buffer, bufsize);
            bufsize = 0;
            return 0;
        }
 
        void flush(const char_t* data, size_t size)
        {
            if (size == 0) return;
 
            // fast path, just write data
            if (encoding == get_write_native_encoding())
                writer.write(data, size * sizeof(char_t));
            else
            {
                // convert chunk
                size_t result = convert_buffer_output(scratch.data_char, scratch.data_u8, scratch.data_u16, scratch.data_u32, data, size, encoding);
                assert(result <= sizeof(scratch));
 
                // write data
                writer.write(scratch.data_u8, result);
            }
        }
 
        void write_direct(const char_t* data, size_t length)
        {
            // flush the remaining buffer contents
            flush();
 
            // handle large chunks
            if (length > bufcapacity)
            {
                if (encoding == get_write_native_encoding())
                {
                    // fast path, can just write data chunk
                    writer.write(data, length * sizeof(char_t));
                    return;
                }
 
                // need to convert in suitable chunks
                while (length > bufcapacity)
                {
                    // get chunk size by selecting such number of characters that are guaranteed to fit into scratch buffer
                    // and form a complete codepoint sequence (i.e. discard start of last codepoint if necessary)
                    size_t chunk_size = get_valid_length(data, bufcapacity);
                    assert(chunk_size);
 
                    // convert chunk and write
                    flush(data, chunk_size);
 
                    // iterate
                    data += chunk_size;
                    length -= chunk_size;
                }
 
                // small tail is copied below
                bufsize = 0;
            }
 
            memcpy(buffer + bufsize, data, length * sizeof(char_t));
            bufsize += length;
        }
 
        void write_buffer(const char_t* data, size_t length)
        {
            size_t offset = bufsize;
 
            if (offset + length <= bufcapacity)
            {
                memcpy(buffer + offset, data, length * sizeof(char_t));
                bufsize = offset + length;
            }
            else
            {
                write_direct(data, length);
            }
        }
 
        void write_string(const char_t* data)
        {
            // write the part of the string that fits in the buffer
            size_t offset = bufsize;
 
            while (*data && offset < bufcapacity)
                buffer[offset++] = *data++;
 
            // write the rest
            if (offset < bufcapacity)
            {
                bufsize = offset;
            }
            else
            {
                // backtrack a bit if we have split the codepoint
                size_t length = offset - bufsize;
                size_t extra = length - get_valid_length(data - length, length);
 
                bufsize = offset - extra;
 
                write_direct(data - extra, strlength(data) + extra);
            }
        }
 
        void write(char_t d0)
        {
            size_t offset = bufsize;
            if (offset > bufcapacity - 1) offset = flush();
 
            buffer[offset + 0] = d0;
            bufsize = offset + 1;
        }
 
        void write(char_t d0, char_t d1)
        {
            size_t offset = bufsize;
            if (offset > bufcapacity - 2) offset = flush();
 
            buffer[offset + 0] = d0;
            buffer[offset + 1] = d1;
            bufsize = offset + 2;
        }
 
        void write(char_t d0, char_t d1, char_t d2)
        {
            size_t offset = bufsize;
            if (offset > bufcapacity - 3) offset = flush();
 
            buffer[offset + 0] = d0;
            buffer[offset + 1] = d1;
            buffer[offset + 2] = d2;
            bufsize = offset + 3;
        }
 
        void write(char_t d0, char_t d1, char_t d2, char_t d3)
        {
            size_t offset = bufsize;
            if (offset > bufcapacity - 4) offset = flush();
 
            buffer[offset + 0] = d0;
            buffer[offset + 1] = d1;
            buffer[offset + 2] = d2;
            buffer[offset + 3] = d3;
            bufsize = offset + 4;
        }
 
        void write(char_t d0, char_t d1, char_t d2, char_t d3, char_t d4)
        {
            size_t offset = bufsize;
            if (offset > bufcapacity - 5) offset = flush();
 
            buffer[offset + 0] = d0;
            buffer[offset + 1] = d1;
            buffer[offset + 2] = d2;
            buffer[offset + 3] = d3;
            buffer[offset + 4] = d4;
            bufsize = offset + 5;
        }
 
        void write(char_t d0, char_t d1, char_t d2, char_t d3, char_t d4, char_t d5)
        {
            size_t offset = bufsize;
            if (offset > bufcapacity - 6) offset = flush();
 
            buffer[offset + 0] = d0;
            buffer[offset + 1] = d1;
            buffer[offset + 2] = d2;
            buffer[offset + 3] = d3;
            buffer[offset + 4] = d4;
            buffer[offset + 5] = d5;
            bufsize = offset + 6;
        }
 
        // utf8 maximum expansion: x4 (-> utf32)
        // utf16 maximum expansion: x2 (-> utf32)
        // utf32 maximum expansion: x1
        enum
        {
            bufcapacitybytes =
            #ifdef PUGIXML_MEMORY_OUTPUT_STACK
                PUGIXML_MEMORY_OUTPUT_STACK
            #else
                10240
            #endif
            ,
            bufcapacity = bufcapacitybytes / (sizeof(char_t) + 4)
        };
 
        char_t buffer[bufcapacity];
 
        union
        {
            uint8_t data_u8[4 * bufcapacity];
            uint16_t data_u16[2 * bufcapacity];
            uint32_t data_u32[bufcapacity];
            char_t data_char[bufcapacity];
        } scratch;
 
        xml_writer& writer;
        size_t bufsize;
        xml_encoding encoding;
    };
 
    PUGI__FN void text_output_escaped(xml_buffered_writer& writer, const char_t* s, chartypex_t type)
    {
        while (*s)
        {
            const char_t* prev = s;
 
            // While *s is a usual symbol
            PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPEX(ss, type));
 
            writer.write_buffer(prev, static_cast<size_t>(s - prev));
 
            switch (*s)
            {
                case 0: break;
                case '&':
                    writer.write('&', 'a', 'm', 'p', ';');
                    ++s;
                    break;
                case '<':
                    writer.write('&', 'l', 't', ';');
                    ++s;
                    break;
                case '>':
                    writer.write('&', 'g', 't', ';');
                    ++s;
                    break;
                case '"':
                    writer.write('&', 'q', 'u', 'o', 't', ';');
                    ++s;
                    break;
                default: // s is not a usual symbol
                {
                    unsigned int ch = static_cast<unsigned int>(*s++);
                    assert(ch < 32);
 
                    writer.write('&', '#', static_cast<char_t>((ch / 10) + '0'), static_cast<char_t>((ch % 10) + '0'), ';');
                }
            }
        }
    }
 
    PUGI__FN void text_output(xml_buffered_writer& writer, const char_t* s, chartypex_t type, unsigned int flags)
    {
        if (flags & format_no_escapes)
            writer.write_string(s);
        else
            text_output_escaped(writer, s, type);
    }
 
    PUGI__FN void text_output_cdata(xml_buffered_writer& writer, const char_t* s)
    {
        do
        {
            writer.write('<', '!', '[', 'C', 'D');
            writer.write('A', 'T', 'A', '[');
 
            const char_t* prev = s;
 
            // look for ]]> sequence - we can't output it as is since it terminates CDATA
            while (*s && !(s[0] == ']' && s[1] == ']' && s[2] == '>')) ++s;
 
            // skip ]] if we stopped at ]]>, > will go to the next CDATA section
            if (*s) s += 2;
 
            writer.write_buffer(prev, static_cast<size_t>(s - prev));
 
            writer.write(']', ']', '>');
        }
        while (*s);
    }
 
    PUGI__FN void text_output_indent(xml_buffered_writer& writer, const char_t* indent, size_t indent_length, unsigned int depth)
    {
        switch (indent_length)
        {
        case 1:
        {
            for (unsigned int i = 0; i < depth; ++i)
                writer.write(indent[0]);
            break;
        }
 
        case 2:
        {
            for (unsigned int i = 0; i < depth; ++i)
                writer.write(indent[0], indent[1]);
            break;
        }
 
        case 3:
        {
            for (unsigned int i = 0; i < depth; ++i)
                writer.write(indent[0], indent[1], indent[2]);
            break;
        }
 
        case 4:
        {
            for (unsigned int i = 0; i < depth; ++i)
                writer.write(indent[0], indent[1], indent[2], indent[3]);
            break;
        }
 
        default:
        {
            for (unsigned int i = 0; i < depth; ++i)
                writer.write_buffer(indent, indent_length);
        }
        }
    }
 
    PUGI__FN void node_output_comment(xml_buffered_writer& writer, const char_t* s)
    {
        writer.write('<', '!', '-', '-');
 
        while (*s)
        {
            const char_t* prev = s;
 
            // look for -\0 or -- sequence - we can't output it since -- is illegal in comment body
            while (*s && !(s[0] == '-' && (s[1] == '-' || s[1] == 0))) ++s;
 
            writer.write_buffer(prev, static_cast<size_t>(s - prev));
 
            if (*s)
            {
                assert(*s == '-');
 
                writer.write('-', ' ');
                ++s;
            }
        }
 
        writer.write('-', '-', '>');
    }
 
    PUGI__FN void node_output_pi_value(xml_buffered_writer& writer, const char_t* s)
    {
        while (*s)
        {
            const char_t* prev = s;
 
            // look for ?> sequence - we can't output it since ?> terminates PI
            while (*s && !(s[0] == '?' && s[1] == '>')) ++s;
 
            writer.write_buffer(prev, static_cast<size_t>(s - prev));
 
            if (*s)
            {
                assert(s[0] == '?' && s[1] == '>');
 
                writer.write('?', ' ', '>');
                s += 2;
            }
        }
    }
 
    PUGI__FN void node_output_attributes(xml_buffered_writer& writer, xml_node_struct* node, const char_t* indent, size_t indent_length, unsigned int flags, unsigned int depth)
    {
        const char_t* default_name = PUGIXML_TEXT(":anonymous");
 
        for (xml_attribute_struct* a = node->first_attribute; a; a = a->next_attribute)
        {
            if ((flags & (format_indent_attributes | format_raw)) == format_indent_attributes)
            {
                writer.write('\n');
 
                text_output_indent(writer, indent, indent_length, depth + 1);
            }
            else
            {
                writer.write(' ');
            }
 
            writer.write_string(a->name ? a->name + 0 : default_name);
            writer.write('=', '"');
 
            if (a->value)
                text_output(writer, a->value, ctx_special_attr, flags);
 
            writer.write('"');
        }
    }
 
    PUGI__FN bool node_output_start(xml_buffered_writer& writer, xml_node_struct* node, const char_t* indent, size_t indent_length, unsigned int flags, unsigned int depth)
    {
        const char_t* default_name = PUGIXML_TEXT(":anonymous");
        const char_t* name = node->name ? node->name + 0 : default_name;
 
        writer.write('<');
        writer.write_string(name);
 
        if (node->first_attribute)
            node_output_attributes(writer, node, indent, indent_length, flags, depth);
 
        // element nodes can have value if parse_embed_pcdata was used
        if (!node->value)
        {
            if (!node->first_child)
            {
                if (flags & format_no_empty_element_tags)
                {
                    writer.write('>', '<', '/');
                    writer.write_string(name);
                    writer.write('>');
 
                    return false;
                }
                else
                {
                    if ((flags & format_raw) == 0)
                        writer.write(' ');
 
                    writer.write('/', '>');
 
                    return false;
                }
            }
            else
            {
                writer.write('>');
 
                return true;
            }
        }
        else
        {
            writer.write('>');
 
            text_output(writer, node->value, ctx_special_pcdata, flags);
 
            if (!node->first_child)
            {
                writer.write('<', '/');
                writer.write_string(name);
                writer.write('>');
 
                return false;
            }
            else
            {
                return true;
            }
        }
    }
 
    PUGI__FN void node_output_end(xml_buffered_writer& writer, xml_node_struct* node)
    {
        const char_t* default_name = PUGIXML_TEXT(":anonymous");
        const char_t* name = node->name ? node->name + 0 : default_name;
 
        writer.write('<', '/');
        writer.write_string(name);
        writer.write('>');
    }
 
    PUGI__FN void node_output_simple(xml_buffered_writer& writer, xml_node_struct* node, unsigned int flags)
    {
        const char_t* default_name = PUGIXML_TEXT(":anonymous");
 
        switch (PUGI__NODETYPE(node))
        {
            case node_pcdata:
                text_output(writer, node->value ? node->value + 0 : PUGIXML_TEXT(""), ctx_special_pcdata, flags);
                break;
 
            case node_cdata:
                text_output_cdata(writer, node->value ? node->value + 0 : PUGIXML_TEXT(""));
                break;
 
            case node_comment:
                node_output_comment(writer, node->value ? node->value + 0 : PUGIXML_TEXT(""));
                break;
 
            case node_pi:
                writer.write('<', '?');
                writer.write_string(node->name ? node->name + 0 : default_name);
 
                if (node->value)
                {
                    writer.write(' ');
                    node_output_pi_value(writer, node->value);
                }
 
                writer.write('?', '>');
                break;
 
            case node_declaration:
                writer.write('<', '?');
                writer.write_string(node->name ? node->name + 0 : default_name);
                node_output_attributes(writer, node, PUGIXML_TEXT(""), 0, flags | format_raw, 0);
                writer.write('?', '>');
                break;
 
            case node_doctype:
                writer.write('<', '!', 'D', 'O', 'C');
                writer.write('T', 'Y', 'P', 'E');
 
                if (node->value)
                {
                    writer.write(' ');
                    writer.write_string(node->value);
                }
 
                writer.write('>');
                break;
 
            default:
                assert(false && "Invalid node type"); // unreachable
        }
    }
 
    enum indent_flags_t
    {
        indent_newline = 1,
        indent_indent = 2
    };
 
    PUGI__FN void node_output(xml_buffered_writer& writer, xml_node_struct* root, const char_t* indent, unsigned int flags, unsigned int depth)
    {
        size_t indent_length = ((flags & (format_indent | format_indent_attributes)) && (flags & format_raw) == 0) ? strlength(indent) : 0;
        unsigned int indent_flags = indent_indent;
 
        xml_node_struct* node = root;
 
        do
        {
            assert(node);
 
            // begin writing current node
            if (PUGI__NODETYPE(node) == node_pcdata || PUGI__NODETYPE(node) == node_cdata)
            {
                node_output_simple(writer, node, flags);
 
                indent_flags = 0;
            }
            else
            {
                if ((indent_flags & indent_newline) && (flags & format_raw) == 0)
                    writer.write('\n');
 
                if ((indent_flags & indent_indent) && indent_length)
                    text_output_indent(writer, indent, indent_length, depth);
 
                if (PUGI__NODETYPE(node) == node_element)
                {
                    indent_flags = indent_newline | indent_indent;
 
                    if (node_output_start(writer, node, indent, indent_length, flags, depth))
                    {
                        // element nodes can have value if parse_embed_pcdata was used
                        if (node->value)
                            indent_flags = 0;
 
                        node = node->first_child;
                        depth++;
                        continue;
                    }
                }
                else if (PUGI__NODETYPE(node) == node_document)
                {
                    indent_flags = indent_indent;
 
                    if (node->first_child)
                    {
                        node = node->first_child;
                        continue;
                    }
                }
                else
                {
                    node_output_simple(writer, node, flags);
 
                    indent_flags = indent_newline | indent_indent;
                }
            }
 
            // continue to the next node
            while (node != root)
            {
                if (node->next_sibling)
                {
                    node = node->next_sibling;
                    break;
                }
 
                node = node->parent;
 
                // write closing node
                if (PUGI__NODETYPE(node) == node_element)
                {
                    depth--;
 
                    if ((indent_flags & indent_newline) && (flags & format_raw) == 0)
                        writer.write('\n');
 
                    if ((indent_flags & indent_indent) && indent_length)
                        text_output_indent(writer, indent, indent_length, depth);
 
                    node_output_end(writer, node);
 
                    indent_flags = indent_newline | indent_indent;
                }
            }
        }
        while (node != root);
 
        if ((indent_flags & indent_newline) && (flags & format_raw) == 0)
            writer.write('\n');
    }
 
    PUGI__FN bool has_declaration(xml_node_struct* node)
    {
        for (xml_node_struct* child = node->first_child; child; child = child->next_sibling)
        {
            xml_node_type type = PUGI__NODETYPE(child);
 
            if (type == node_declaration) return true;
            if (type == node_element) return false;
        }
 
        return false;
    }
 
    PUGI__FN bool is_attribute_of(xml_attribute_struct* attr, xml_node_struct* node)
    {
        for (xml_attribute_struct* a = node->first_attribute; a; a = a->next_attribute)
            if (a == attr)
                return true;
 
        return false;
    }
 
    PUGI__FN bool allow_insert_attribute(xml_node_type parent)
    {
        return parent == node_element || parent == node_declaration;
    }
 
    PUGI__FN bool allow_insert_child(xml_node_type parent, xml_node_type child)
    {
        if (parent != node_document && parent != node_element) return false;
        if (child == node_document || child == node_null) return false;
        if (parent != node_document && (child == node_declaration || child == node_doctype)) return false;
 
        return true;
    }
 
    PUGI__FN bool allow_move(xml_node parent, xml_node child)
    {
        // check that child can be a child of parent
        if (!allow_insert_child(parent.type(), child.type()))
            return false;
 
        // check that node is not moved between documents
        if (parent.root() != child.root())
            return false;
 
        // check that new parent is not in the child subtree
        xml_node cur = parent;
 
        while (cur)
        {
            if (cur == child)
                return false;
 
            cur = cur.parent();
        }
 
        return true;
    }
 
    template <typename String, typename Header>
    PUGI__FN void node_copy_string(String& dest, Header& header, uintptr_t header_mask, char_t* source, Header& source_header, xml_allocator* alloc)
    {
        assert(!dest && (header & header_mask) == 0);
 
        if (source)
        {
            if (alloc && (source_header & header_mask) == 0)
            {
                dest = source;
 
                // since strcpy_insitu can reuse document buffer memory we need to mark both source and dest as shared
                header |= xml_memory_page_contents_shared_mask;
                source_header |= xml_memory_page_contents_shared_mask;
            }
            else
                strcpy_insitu(dest, header, header_mask, source, strlength(source));
        }
    }
 
    PUGI__FN void node_copy_contents(xml_node_struct* dn, xml_node_struct* sn, xml_allocator* shared_alloc)
    {
        node_copy_string(dn->name, dn->header, xml_memory_page_name_allocated_mask, sn->name, sn->header, shared_alloc);
        node_copy_string(dn->value, dn->header, xml_memory_page_value_allocated_mask, sn->value, sn->header, shared_alloc);
 
        for (xml_attribute_struct* sa = sn->first_attribute; sa; sa = sa->next_attribute)
        {
            xml_attribute_struct* da = append_new_attribute(dn, get_allocator(dn));
 
            if (da)
            {
                node_copy_string(da->name, da->header, xml_memory_page_name_allocated_mask, sa->name, sa->header, shared_alloc);
                node_copy_string(da->value, da->header, xml_memory_page_value_allocated_mask, sa->value, sa->header, shared_alloc);
            }
        }
    }
 
    PUGI__FN void node_copy_tree(xml_node_struct* dn, xml_node_struct* sn)
    {
        xml_allocator& alloc = get_allocator(dn);
        xml_allocator* shared_alloc = (&alloc == &get_allocator(sn)) ? &alloc : 0;
 
        node_copy_contents(dn, sn, shared_alloc);
 
        xml_node_struct* dit = dn;
        xml_node_struct* sit = sn->first_child;
 
        while (sit && sit != sn)
        {
            // when a tree is copied into one of the descendants, we need to skip that subtree to avoid an infinite loop
            if (sit != dn)
            {
                xml_node_struct* copy = append_new_node(dit, alloc, PUGI__NODETYPE(sit));
 
                if (copy)
                {
                    node_copy_contents(copy, sit, shared_alloc);
 
                    if (sit->first_child)
                    {
                        dit = copy;
                        sit = sit->first_child;
                        continue;
                    }
                }
            }
 
            // continue to the next node
            do
            {
                if (sit->next_sibling)
                {
                    sit = sit->next_sibling;
                    break;
                }
 
                sit = sit->parent;
                dit = dit->parent;
            }
            while (sit != sn);
        }
    }
 
    PUGI__FN void node_copy_attribute(xml_attribute_struct* da, xml_attribute_struct* sa)
    {
        xml_allocator& alloc = get_allocator(da);
        xml_allocator* shared_alloc = (&alloc == &get_allocator(sa)) ? &alloc : 0;
 
        node_copy_string(da->name, da->header, xml_memory_page_name_allocated_mask, sa->name, sa->header, shared_alloc);
        node_copy_string(da->value, da->header, xml_memory_page_value_allocated_mask, sa->value, sa->header, shared_alloc);
    }
 
    inline bool is_text_node(xml_node_struct* node)
    {
        xml_node_type type = PUGI__NODETYPE(node);
 
        return type == node_pcdata || type == node_cdata;
    }
 
    // get value with conversion functions
    template <typename U> PUGI__FN PUGI__UNSIGNED_OVERFLOW U string_to_integer(const char_t* value, U minv, U maxv)
    {
        U result = 0;
        const char_t* s = value;
 
        while (PUGI__IS_CHARTYPE(*s, ct_space))
            s++;
 
        bool negative = (*s == '-');
 
        s += (*s == '+' || *s == '-');
 
        bool overflow = false;
 
        if (s[0] == '0' && (s[1] | ' ') == 'x')
        {
            s += 2;
 
            // since overflow detection relies on length of the sequence skip leading zeros
            while (*s == '0')
                s++;
 
            const char_t* start = s;
 
            for (;;)
            {
                if (static_cast<unsigned>(*s - '0') < 10)
                    result = result * 16 + (*s - '0');
                else if (static_cast<unsigned>((*s | ' ') - 'a') < 6)
                    result = result * 16 + ((*s | ' ') - 'a' + 10);
                else
                    break;
 
                s++;
            }
 
            size_t digits = static_cast<size_t>(s - start);
 
            overflow = digits > sizeof(U) * 2;
        }
        else
        {
            // since overflow detection relies on length of the sequence skip leading zeros
            while (*s == '0')
                s++;
 
            const char_t* start = s;
 
            for (;;)
            {
                if (static_cast<unsigned>(*s - '0') < 10)
                    result = result * 10 + (*s - '0');
                else
                    break;
 
                s++;
            }
 
            size_t digits = static_cast<size_t>(s - start);
 
            PUGI__STATIC_ASSERT(sizeof(U) == 8 || sizeof(U) == 4 || sizeof(U) == 2);
 
            const size_t max_digits10 = sizeof(U) == 8 ? 20 : sizeof(U) == 4 ? 10 : 5;
            const char_t max_lead = sizeof(U) == 8 ? '1' : sizeof(U) == 4 ? '4' : '6';
            const size_t high_bit = sizeof(U) * 8 - 1;
 
            overflow = digits >= max_digits10 && !(digits == max_digits10 && (*start < max_lead || (*start == max_lead && result >> high_bit)));
        }
 
        if (negative)
        {
            // Workaround for crayc++ CC-3059: Expected no overflow in routine.
        #ifdef _CRAYC
            return (overflow || result > ~minv + 1) ? minv : ~result + 1;
        #else
            return (overflow || result > 0 - minv) ? minv : 0 - result;
        #endif
        }
        else
            return (overflow || result > maxv) ? maxv : result;
    }
 
    PUGI__FN int get_value_int(const char_t* value)
    {
        return string_to_integer<unsigned int>(value, static_cast<unsigned int>(INT_MIN), INT_MAX);
    }
 
    PUGI__FN unsigned int get_value_uint(const char_t* value)
    {
        return string_to_integer<unsigned int>(value, 0, UINT_MAX);
    }
 
    PUGI__FN double get_value_double(const char_t* value)
    {
    #ifdef PUGIXML_WCHAR_MODE
        return wcstod(value, 0);
    #else
        return strtod(value, 0);
    #endif
    }
 
    PUGI__FN float get_value_float(const char_t* value)
    {
    #ifdef PUGIXML_WCHAR_MODE
        return static_cast<float>(wcstod(value, 0));
    #else
        return static_cast<float>(strtod(value, 0));
    #endif
    }
 
    PUGI__FN bool get_value_bool(const char_t* value)
    {
        // only look at first char
        char_t first = *value;
 
        // 1*, t* (true), T* (True), y* (yes), Y* (YES)
        return (first == '1' || first == 't' || first == 'T' || first == 'y' || first == 'Y');
    }
 
#ifdef PUGIXML_HAS_LONG_LONG
    PUGI__FN long long get_value_llong(const char_t* value)
    {
        return string_to_integer<unsigned long long>(value, static_cast<unsigned long long>(LLONG_MIN), LLONG_MAX);
    }
 
    PUGI__FN unsigned long long get_value_ullong(const char_t* value)
    {
        return string_to_integer<unsigned long long>(value, 0, ULLONG_MAX);
    }
#endif
 
    template <typename U> PUGI__FN PUGI__UNSIGNED_OVERFLOW char_t* integer_to_string(char_t* begin, char_t* end, U value, bool negative)
    {
        char_t* result = end - 1;
        U rest = negative ? 0 - value : value;
 
        do
        {
            *result-- = static_cast<char_t>('0' + (rest % 10));
            rest /= 10;
        }
        while (rest);
 
        assert(result >= begin);
        (void)begin;
 
        *result = '-';
 
        return result + !negative;
    }
 
    // set value with conversion functions
    template <typename String, typename Header>
    PUGI__FN bool set_value_ascii(String& dest, Header& header, uintptr_t header_mask, char* buf)
    {
    #ifdef PUGIXML_WCHAR_MODE
        char_t wbuf[128];
        assert(strlen(buf) < sizeof(wbuf) / sizeof(wbuf[0]));
 
        size_t offset = 0;
        for (; buf[offset]; ++offset) wbuf[offset] = buf[offset];
 
        return strcpy_insitu(dest, header, header_mask, wbuf, offset);
    #else
        return strcpy_insitu(dest, header, header_mask, buf, strlen(buf));
    #endif
    }
 
    template <typename U, typename String, typename Header>
    PUGI__FN bool set_value_integer(String& dest, Header& header, uintptr_t header_mask, U value, bool negative)
    {
        char_t buf[64];
        char_t* end = buf + sizeof(buf) / sizeof(buf[0]);
        char_t* begin = integer_to_string(buf, end, value, negative);
 
        return strcpy_insitu(dest, header, header_mask, begin, end - begin);
    }
 
    template <typename String, typename Header>
    PUGI__FN bool set_value_convert(String& dest, Header& header, uintptr_t header_mask, float value)
    {
        char buf[128];
        PUGI__SNPRINTF(buf, "%.9g", value);
 
        return set_value_ascii(dest, header, header_mask, buf);
    }
 
    template <typename String, typename Header>
    PUGI__FN bool set_value_convert(String& dest, Header& header, uintptr_t header_mask, double value)
    {
        char buf[128];
        PUGI__SNPRINTF(buf, "%.17g", value);
 
        return set_value_ascii(dest, header, header_mask, buf);
    }
 
    template <typename String, typename Header>
    PUGI__FN bool set_value_bool(String& dest, Header& header, uintptr_t header_mask, bool value)
    {
        return strcpy_insitu(dest, header, header_mask, value ? PUGIXML_TEXT("true") : PUGIXML_TEXT("false"), value ? 4 : 5);
    }
 
    PUGI__FN xml_parse_result load_buffer_impl(xml_document_struct* doc, xml_node_struct* root, void* contents, size_t size, unsigned int options, xml_encoding encoding, bool is_mutable, bool own, char_t** out_buffer)
    {
        // check input buffer
        if (!contents && size) return make_parse_result(status_io_error);
 
        // get actual encoding
        xml_encoding buffer_encoding = impl::get_buffer_encoding(encoding, contents, size);
 
        // get private buffer
        char_t* buffer = 0;
        size_t length = 0;
 
        if (!impl::convert_buffer(buffer, length, buffer_encoding, contents, size, is_mutable)) return impl::make_parse_result(status_out_of_memory);
 
        // delete original buffer if we performed a conversion
        if (own && buffer != contents && contents) impl::xml_memory::deallocate(contents);
 
        // grab onto buffer if it's our buffer, user is responsible for deallocating contents himself
        if (own || buffer != contents) *out_buffer = buffer;
 
        // store buffer for offset_debug
        doc->buffer = buffer;
 
        // parse
        xml_parse_result res = impl::xml_parser::parse(buffer, length, doc, root, options);
 
        // remember encoding
        res.encoding = buffer_encoding;
 
        return res;
    }
 
    // we need to get length of entire file to load it in memory; the only (relatively) sane way to do it is via seek/tell trick
    PUGI__FN xml_parse_status get_file_size(FILE* file, size_t& out_result)
    {
    #if defined(PUGI__MSVC_CRT_VERSION) && PUGI__MSVC_CRT_VERSION >= 1400 && !defined(_WIN32_WCE)
        // there are 64-bit versions of fseek/ftell, let's use them
        typedef __int64 length_type;
 
        _fseeki64(file, 0, SEEK_END);
        length_type length = _ftelli64(file);
        _fseeki64(file, 0, SEEK_SET);
    #elif defined(__MINGW32__) && !defined(__NO_MINGW_LFS) && (!defined(__STRICT_ANSI__) || defined(__MINGW64_VERSION_MAJOR))
        // there are 64-bit versions of fseek/ftell, let's use them
        typedef off64_t length_type;
 
        fseeko64(file, 0, SEEK_END);
        length_type length = ftello64(file);
        fseeko64(file, 0, SEEK_SET);
    #else
        // if this is a 32-bit OS, long is enough; if this is a unix system, long is 64-bit, which is enough; otherwise we can't do anything anyway.
        typedef long length_type;
 
        fseek(file, 0, SEEK_END);
        length_type length = ftell(file);
        fseek(file, 0, SEEK_SET);
    #endif
 
        // check for I/O errors
        if (length < 0) return status_io_error;
 
        // check for overflow
        size_t result = static_cast<size_t>(length);
 
        if (static_cast<length_type>(result) != length) return status_out_of_memory;
 
        // finalize
        out_result = result;
 
        return status_ok;
    }
 
    // This function assumes that buffer has extra sizeof(char_t) writable bytes after size
    PUGI__FN size_t zero_terminate_buffer(void* buffer, size_t size, xml_encoding encoding)
    {
        // We only need to zero-terminate if encoding conversion does not do it for us
    #ifdef PUGIXML_WCHAR_MODE
        xml_encoding wchar_encoding = get_wchar_encoding();
 
        if (encoding == wchar_encoding || need_endian_swap_utf(encoding, wchar_encoding))
        {
            size_t length = size / sizeof(char_t);
 
            static_cast<char_t*>(buffer)[length] = 0;
            return (length + 1) * sizeof(char_t);
        }
    #else
        if (encoding == encoding_utf8)
        {
            static_cast<char*>(buffer)[size] = 0;
            return size + 1;
        }
    #endif
 
        return size;
    }
 
    PUGI__FN xml_parse_result load_file_impl(xml_document_struct* doc, FILE* file, unsigned int options, xml_encoding encoding, char_t** out_buffer)
    {
        if (!file) return make_parse_result(status_file_not_found);
 
        // get file size (can result in I/O errors)
        size_t size = 0;
        xml_parse_status size_status = get_file_size(file, size);
        if (size_status != status_ok) return make_parse_result(size_status);
 
        size_t max_suffix_size = sizeof(char_t);
 
        // allocate buffer for the whole file
        char* contents = static_cast<char*>(xml_memory::allocate(size + max_suffix_size));
        if (!contents) return make_parse_result(status_out_of_memory);
 
        // read file in memory
        size_t read_size = fread(contents, 1, size, file);
 
        if (read_size != size)
        {
            xml_memory::deallocate(contents);
            return make_parse_result(status_io_error);
        }
 
        xml_encoding real_encoding = get_buffer_encoding(encoding, contents, size);
 
        return load_buffer_impl(doc, doc, contents, zero_terminate_buffer(contents, size, real_encoding), options, real_encoding, true, true, out_buffer);
    }
 
    PUGI__FN void close_file(FILE* file)
    {
        fclose(file);
    }
 
#ifndef PUGIXML_NO_STL
    template <typename T> struct xml_stream_chunk
    {
        static xml_stream_chunk* create()
        {
            void* memory = xml_memory::allocate(sizeof(xml_stream_chunk));
            if (!memory) return 0;
 
            return new (memory) xml_stream_chunk();
        }
 
        static void destroy(xml_stream_chunk* chunk)
        {
            // free chunk chain
            while (chunk)
            {
                xml_stream_chunk* next_ = chunk->next;
 
                xml_memory::deallocate(chunk);
 
                chunk = next_;
            }
        }
 
        xml_stream_chunk(): next(0), size(0)
        {
        }
 
        xml_stream_chunk* next;
        size_t size;
 
        T data[xml_memory_page_size / sizeof(T)];
    };
 
    template <typename T> PUGI__FN xml_parse_status load_stream_data_noseek(std::basic_istream<T>& stream, void** out_buffer, size_t* out_size)
    {
        auto_deleter<xml_stream_chunk<T> > chunks(0, xml_stream_chunk<T>::destroy);
 
        // read file to a chunk list
        size_t total = 0;
        xml_stream_chunk<T>* last = 0;
 
        while (!stream.eof())
        {
            // allocate new chunk
            xml_stream_chunk<T>* chunk = xml_stream_chunk<T>::create();
            if (!chunk) return status_out_of_memory;
 
            // append chunk to list
            if (last) last = last->next = chunk;
            else chunks.data = last = chunk;
 
            // read data to chunk
            stream.read(chunk->data, static_cast<std::streamsize>(sizeof(chunk->data) / sizeof(T)));
            chunk->size = static_cast<size_t>(stream.gcount()) * sizeof(T);
 
            // read may set failbit | eofbit in case gcount() is less than read length, so check for other I/O errors
            if (stream.bad() || (!stream.eof() && stream.fail())) return status_io_error;
 
            // guard against huge files (chunk size is small enough to make this overflow check work)
            if (total + chunk->size < total) return status_out_of_memory;
            total += chunk->size;
        }
 
        size_t max_suffix_size = sizeof(char_t);
 
        // copy chunk list to a contiguous buffer
        char* buffer = static_cast<char*>(xml_memory::allocate(total + max_suffix_size));
        if (!buffer) return status_out_of_memory;
 
        char* write = buffer;
 
        for (xml_stream_chunk<T>* chunk = chunks.data; chunk; chunk = chunk->next)
        {
            assert(write + chunk->size <= buffer + total);
            memcpy(write, chunk->data, chunk->size);
            write += chunk->size;
        }
 
        assert(write == buffer + total);
 
        // return buffer
        *out_buffer = buffer;
        *out_size = total;
 
        return status_ok;
    }
 
    template <typename T> PUGI__FN xml_parse_status load_stream_data_seek(std::basic_istream<T>& stream, void** out_buffer, size_t* out_size)
    {
        // get length of remaining data in stream
        typename std::basic_istream<T>::pos_type pos = stream.tellg();
        stream.seekg(0, std::ios::end);
        std::streamoff length = stream.tellg() - pos;
        stream.seekg(pos);
 
        if (stream.fail() || pos < 0) return status_io_error;
 
        // guard against huge files
        size_t read_length = static_cast<size_t>(length);
 
        if (static_cast<std::streamsize>(read_length) != length || length < 0) return status_out_of_memory;
 
        size_t max_suffix_size = sizeof(char_t);
 
        // read stream data into memory (guard against stream exceptions with buffer holder)
        auto_deleter<void> buffer(xml_memory::allocate(read_length * sizeof(T) + max_suffix_size), xml_memory::deallocate);
        if (!buffer.data) return status_out_of_memory;
 
        stream.read(static_cast<T*>(buffer.data), static_cast<std::streamsize>(read_length));
 
        // read may set failbit | eofbit in case gcount() is less than read_length (i.e. line ending conversion), so check for other I/O errors
        if (stream.bad() || (!stream.eof() && stream.fail())) return status_io_error;
 
        // return buffer
        size_t actual_length = static_cast<size_t>(stream.gcount());
        assert(actual_length <= read_length);
 
        *out_buffer = buffer.release();
        *out_size = actual_length * sizeof(T);
 
        return status_ok;
    }
 
    template <typename T> PUGI__FN xml_parse_result load_stream_impl(xml_document_struct* doc, std::basic_istream<T>& stream, unsigned int options, xml_encoding encoding, char_t** out_buffer)
    {
        void* buffer = 0;
        size_t size = 0;
        xml_parse_status status = status_ok;
 
        // if stream has an error bit set, bail out (otherwise tellg() can fail and we'll clear error bits)
        if (stream.fail()) return make_parse_result(status_io_error);
 
        // load stream to memory (using seek-based implementation if possible, since it's faster and takes less memory)
        if (stream.tellg() < 0)
        {
            stream.clear(); // clear error flags that could be set by a failing tellg
            status = load_stream_data_noseek(stream, &buffer, &size);
        }
        else
            status = load_stream_data_seek(stream, &buffer, &size);
 
        if (status != status_ok) return make_parse_result(status);
 
        xml_encoding real_encoding = get_buffer_encoding(encoding, buffer, size);
 
        return load_buffer_impl(doc, doc, buffer, zero_terminate_buffer(buffer, size, real_encoding), options, real_encoding, true, true, out_buffer);
    }
#endif
 
#if defined(PUGI__MSVC_CRT_VERSION) || defined(__BORLANDC__) || (defined(__MINGW32__) && (!defined(__STRICT_ANSI__) || defined(__MINGW64_VERSION_MAJOR)))
    PUGI__FN FILE* open_file_wide(const wchar_t* path, const wchar_t* mode)
    {
        return _wfopen(path, mode);
    }
#else
    PUGI__FN char* convert_path_heap(const wchar_t* str)
    {
        assert(str);
 
        // first pass: get length in utf8 characters
        size_t length = strlength_wide(str);
        size_t size = as_utf8_begin(str, length);
 
        // allocate resulting string
        char* result = static_cast<char*>(xml_memory::allocate(size + 1));
        if (!result) return 0;
 
        // second pass: convert to utf8
        as_utf8_end(result, size, str, length);
 
        // zero-terminate
        result[size] = 0;
 
        return result;
    }
 
    PUGI__FN FILE* open_file_wide(const wchar_t* path, const wchar_t* mode)
    {
        // there is no standard function to open wide paths, so our best bet is to try utf8 path
        char* path_utf8 = convert_path_heap(path);
        if (!path_utf8) return 0;
 
        // convert mode to ASCII (we mirror _wfopen interface)
        char mode_ascii[4] = {0};
        for (size_t i = 0; mode[i]; ++i) mode_ascii[i] = static_cast<char>(mode[i]);
 
        // try to open the utf8 path
        FILE* result = fopen(path_utf8, mode_ascii);
 
        // free dummy buffer
        xml_memory::deallocate(path_utf8);
 
        return result;
    }
#endif
 
    PUGI__FN bool save_file_impl(const xml_document& doc, FILE* file, const char_t* indent, unsigned int flags, xml_encoding encoding)
    {
        if (!file) return false;
 
        xml_writer_file writer(file);
        doc.save(writer, indent, flags, encoding);
 
        return ferror(file) == 0;
    }
 
    struct name_null_sentry
    {
        xml_node_struct* node;
        char_t* name;
 
        name_null_sentry(xml_node_struct* node_): node(node_), name(node_->name)
        {
            node->name = 0;
        }
 
        ~name_null_sentry()
        {
            node->name = name;
        }
    };
PUGI__NS_END
 
namespace pugi
{
    PUGI__FN xml_writer_file::xml_writer_file(void* file_): file(file_)
    {
    }
 
    PUGI__FN void xml_writer_file::write(const void* data, size_t size)
    {
        size_t result = fwrite(data, 1, size, static_cast<FILE*>(file));
        (void)!result; // unfortunately we can't do proper error handling here
    }
 
#ifndef PUGIXML_NO_STL
    PUGI__FN xml_writer_stream::xml_writer_stream(std::basic_ostream<char, std::char_traits<char> >& stream): narrow_stream(&stream), wide_stream(0)
    {
    }
 
    PUGI__FN xml_writer_stream::xml_writer_stream(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >& stream): narrow_stream(0), wide_stream(&stream)
    {
    }
 
    PUGI__FN void xml_writer_stream::write(const void* data, size_t size)
    {
        if (narrow_stream)
        {
            assert(!wide_stream);
            narrow_stream->write(reinterpret_cast<const char*>(data), static_cast<std::streamsize>(size));
        }
        else
        {
            assert(wide_stream);
            assert(size % sizeof(wchar_t) == 0);
 
            wide_stream->write(reinterpret_cast<const wchar_t*>(data), static_cast<std::streamsize>(size / sizeof(wchar_t)));
        }
    }
#endif
 
    PUGI__FN xml_tree_walker::xml_tree_walker(): _depth(0)
    {
    }
 
    PUGI__FN xml_tree_walker::~xml_tree_walker()
    {
    }
 
    PUGI__FN int xml_tree_walker::depth() const
    {
        return _depth;
    }
 
    PUGI__FN bool xml_tree_walker::begin(xml_node&)
    {
        return true;
    }
 
    PUGI__FN bool xml_tree_walker::end(xml_node&)
    {
        return true;
    }
 
    PUGI__FN xml_attribute::xml_attribute(): _attr(0)
    {
    }
 
    PUGI__FN xml_attribute::xml_attribute(xml_attribute_struct* attr): _attr(attr)
    {
    }
 
    PUGI__FN static void unspecified_bool_xml_attribute(xml_attribute***)
    {
    }
 
    PUGI__FN xml_attribute::operator xml_attribute::unspecified_bool_type() const
    {
        return _attr ? unspecified_bool_xml_attribute : 0;
    }
 
    PUGI__FN bool xml_attribute::operator!() const
    {
        return !_attr;
    }
 
    PUGI__FN bool xml_attribute::operator==(const xml_attribute& r) const
    {
        return (_attr == r._attr);
    }
 
    PUGI__FN bool xml_attribute::operator!=(const xml_attribute& r) const
    {
        return (_attr != r._attr);
    }
 
    PUGI__FN bool xml_attribute::operator<(const xml_attribute& r) const
    {
        return (_attr < r._attr);
    }
 
    PUGI__FN bool xml_attribute::operator>(const xml_attribute& r) const
    {
        return (_attr > r._attr);
    }
 
    PUGI__FN bool xml_attribute::operator<=(const xml_attribute& r) const
    {
        return (_attr <= r._attr);
    }
 
    PUGI__FN bool xml_attribute::operator>=(const xml_attribute& r) const
    {
        return (_attr >= r._attr);
    }
 
    PUGI__FN xml_attribute xml_attribute::next_attribute() const
    {
        return _attr ? xml_attribute(_attr->next_attribute) : xml_attribute();
    }
 
    PUGI__FN xml_attribute xml_attribute::previous_attribute() const
    {
        return _attr && _attr->prev_attribute_c->next_attribute ? xml_attribute(_attr->prev_attribute_c) : xml_attribute();
    }
 
    PUGI__FN const char_t* xml_attribute::as_string(const char_t* def) const
    {
        return (_attr && _attr->value) ? _attr->value + 0 : def;
    }
 
    PUGI__FN int xml_attribute::as_int(int def) const
    {
        return (_attr && _attr->value) ? impl::get_value_int(_attr->value) : def;
    }
 
    PUGI__FN unsigned int xml_attribute::as_uint(unsigned int def) const
    {
        return (_attr && _attr->value) ? impl::get_value_uint(_attr->value) : def;
    }
 
    PUGI__FN double xml_attribute::as_double(double def) const
    {
        return (_attr && _attr->value) ? impl::get_value_double(_attr->value) : def;
    }
 
    PUGI__FN float xml_attribute::as_float(float def) const
    {
        return (_attr && _attr->value) ? impl::get_value_float(_attr->value) : def;
    }
 
    PUGI__FN bool xml_attribute::as_bool(bool def) const
    {
        return (_attr && _attr->value) ? impl::get_value_bool(_attr->value) : def;
    }
 
#ifdef PUGIXML_HAS_LONG_LONG
    PUGI__FN long long xml_attribute::as_llong(long long def) const
    {
        return (_attr && _attr->value) ? impl::get_value_llong(_attr->value) : def;
    }
 
    PUGI__FN unsigned long long xml_attribute::as_ullong(unsigned long long def) const
    {
        return (_attr && _attr->value) ? impl::get_value_ullong(_attr->value) : def;
    }
#endif
 
    PUGI__FN bool xml_attribute::empty() const
    {
        return !_attr;
    }
 
    PUGI__FN const char_t* xml_attribute::name() const
    {
        return (_attr && _attr->name) ? _attr->name + 0 : PUGIXML_TEXT("");
    }
 
    PUGI__FN const char_t* xml_attribute::value() const
    {
        return (_attr && _attr->value) ? _attr->value + 0 : PUGIXML_TEXT("");
    }
 
    PUGI__FN size_t xml_attribute::hash_value() const
    {
        return static_cast<size_t>(reinterpret_cast<uintptr_t>(_attr) / sizeof(xml_attribute_struct));
    }
 
    PUGI__FN xml_attribute_struct* xml_attribute::internal_object() const
    {
        return _attr;
    }
 
    PUGI__FN xml_attribute& xml_attribute::operator=(const char_t* rhs)
    {
        set_value(rhs);
        return *this;
    }
 
    PUGI__FN xml_attribute& xml_attribute::operator=(int rhs)
    {
        set_value(rhs);
        return *this;
    }
 
    PUGI__FN xml_attribute& xml_attribute::operator=(unsigned int rhs)
    {
        set_value(rhs);
        return *this;
    }
 
    PUGI__FN xml_attribute& xml_attribute::operator=(long rhs)
    {
        set_value(rhs);
        return *this;
    }
 
    PUGI__FN xml_attribute& xml_attribute::operator=(unsigned long rhs)
    {
        set_value(rhs);
        return *this;
    }
 
    PUGI__FN xml_attribute& xml_attribute::operator=(double rhs)
    {
        set_value(rhs);
        return *this;
    }
 
    PUGI__FN xml_attribute& xml_attribute::operator=(float rhs)
    {
        set_value(rhs);
        return *this;
    }
 
    PUGI__FN xml_attribute& xml_attribute::operator=(bool rhs)
    {
        set_value(rhs);
        return *this;
    }
 
#ifdef PUGIXML_HAS_LONG_LONG
    PUGI__FN xml_attribute& xml_attribute::operator=(long long rhs)
    {
        set_value(rhs);
        return *this;
    }
 
    PUGI__FN xml_attribute& xml_attribute::operator=(unsigned long long rhs)
    {
        set_value(rhs);
        return *this;
    }
#endif
 
    PUGI__FN bool xml_attribute::set_name(const char_t* rhs)
    {
        if (!_attr) return false;
 
        return impl::strcpy_insitu(_attr->name, _attr->header, impl::xml_memory_page_name_allocated_mask, rhs, impl::strlength(rhs));
    }
 
    PUGI__FN bool xml_attribute::set_value(const char_t* rhs)
    {
        if (!_attr) return false;
 
        return impl::strcpy_insitu(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs, impl::strlength(rhs));
    }
 
    PUGI__FN bool xml_attribute::set_value(int rhs)
    {
        if (!_attr) return false;
 
        return impl::set_value_integer<unsigned int>(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs, rhs < 0);
    }
 
    PUGI__FN bool xml_attribute::set_value(unsigned int rhs)
    {
        if (!_attr) return false;
 
        return impl::set_value_integer<unsigned int>(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs, false);
    }
 
    PUGI__FN bool xml_attribute::set_value(long rhs)
    {
        if (!_attr) return false;
 
        return impl::set_value_integer<unsigned long>(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs, rhs < 0);
    }
 
    PUGI__FN bool xml_attribute::set_value(unsigned long rhs)
    {
        if (!_attr) return false;
 
        return impl::set_value_integer<unsigned long>(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs, false);
    }
 
    PUGI__FN bool xml_attribute::set_value(double rhs)
    {
        if (!_attr) return false;
 
        return impl::set_value_convert(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs);
    }
 
    PUGI__FN bool xml_attribute::set_value(float rhs)
    {
        if (!_attr) return false;
 
        return impl::set_value_convert(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs);
    }
 
    PUGI__FN bool xml_attribute::set_value(bool rhs)
    {
        if (!_attr) return false;
 
        return impl::set_value_bool(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs);
    }
 
#ifdef PUGIXML_HAS_LONG_LONG
    PUGI__FN bool xml_attribute::set_value(long long rhs)
    {
        if (!_attr) return false;
 
        return impl::set_value_integer<unsigned long long>(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs, rhs < 0);
    }
 
    PUGI__FN bool xml_attribute::set_value(unsigned long long rhs)
    {
        if (!_attr) return false;
 
        return impl::set_value_integer<unsigned long long>(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs, false);
    }
#endif
 
#ifdef __BORLANDC__
    PUGI__FN bool operator&&(const xml_attribute& lhs, bool rhs)
    {
        return (bool)lhs && rhs;
    }
 
    PUGI__FN bool operator||(const xml_attribute& lhs, bool rhs)
    {
        return (bool)lhs || rhs;
    }
#endif
 
    PUGI__FN xml_node::xml_node(): _root(0)
    {
    }
 
    PUGI__FN xml_node::xml_node(xml_node_struct* p): _root(p)
    {
    }
 
    PUGI__FN static void unspecified_bool_xml_node(xml_node***)
    {
    }
 
    PUGI__FN xml_node::operator xml_node::unspecified_bool_type() const
    {
        return _root ? unspecified_bool_xml_node : 0;
    }
 
    PUGI__FN bool xml_node::operator!() const
    {
        return !_root;
    }
 
    PUGI__FN xml_node::iterator xml_node::begin() const
    {
        return iterator(_root ? _root->first_child + 0 : 0, _root);
    }
 
    PUGI__FN xml_node::iterator xml_node::end() const
    {
        return iterator(0, _root);
    }
 
    PUGI__FN xml_node::attribute_iterator xml_node::attributes_begin() const
    {
        return attribute_iterator(_root ? _root->first_attribute + 0 : 0, _root);
    }
 
    PUGI__FN xml_node::attribute_iterator xml_node::attributes_end() const
    {
        return attribute_iterator(0, _root);
    }
 
    PUGI__FN xml_object_range<xml_node_iterator> xml_node::children() const
    {
        return xml_object_range<xml_node_iterator>(begin(), end());
    }
 
    PUGI__FN xml_object_range<xml_named_node_iterator> xml_node::children(const char_t* name_) const
    {
        return xml_object_range<xml_named_node_iterator>(xml_named_node_iterator(child(name_)._root, _root, name_), xml_named_node_iterator(0, _root, name_));
    }
 
    PUGI__FN xml_object_range<xml_attribute_iterator> xml_node::attributes() const
    {
        return xml_object_range<xml_attribute_iterator>(attributes_begin(), attributes_end());
    }
 
    PUGI__FN bool xml_node::operator==(const xml_node& r) const
    {
        return (_root == r._root);
    }
 
    PUGI__FN bool xml_node::operator!=(const xml_node& r) const
    {
        return (_root != r._root);
    }
 
    PUGI__FN bool xml_node::operator<(const xml_node& r) const
    {
        return (_root < r._root);
    }
 
    PUGI__FN bool xml_node::operator>(const xml_node& r) const
    {
        return (_root > r._root);
    }
 
    PUGI__FN bool xml_node::operator<=(const xml_node& r) const
    {
        return (_root <= r._root);
    }
 
    PUGI__FN bool xml_node::operator>=(const xml_node& r) const
    {
        return (_root >= r._root);
    }
 
    PUGI__FN bool xml_node::empty() const
    {
        return !_root;
    }
 
    PUGI__FN const char_t* xml_node::name() const
    {
        return (_root && _root->name) ? _root->name + 0 : PUGIXML_TEXT("");
    }
 
    PUGI__FN xml_node_type xml_node::type() const
    {
        return _root ? PUGI__NODETYPE(_root) : node_null;
    }
 
    PUGI__FN const char_t* xml_node::value() const
    {
        return (_root && _root->value) ? _root->value + 0 : PUGIXML_TEXT("");
    }
 
    PUGI__FN xml_node xml_node::child(const char_t* name_) const
    {
        if (!_root) return xml_node();
 
        for (xml_node_struct* i = _root->first_child; i; i = i->next_sibling)
            if (i->name && impl::strequal(name_, i->name)) return xml_node(i);
 
        return xml_node();
    }
 
    PUGI__FN xml_attribute xml_node::attribute(const char_t* name_) const
    {
        if (!_root) return xml_attribute();
 
        for (xml_attribute_struct* i = _root->first_attribute; i; i = i->next_attribute)
            if (i->name && impl::strequal(name_, i->name))
                return xml_attribute(i);
 
        return xml_attribute();
    }
 
    PUGI__FN xml_node xml_node::next_sibling(const char_t* name_) const
    {
        if (!_root) return xml_node();
 
        for (xml_node_struct* i = _root->next_sibling; i; i = i->next_sibling)
            if (i->name && impl::strequal(name_, i->name)) return xml_node(i);
 
        return xml_node();
    }
 
    PUGI__FN xml_node xml_node::next_sibling() const
    {
        return _root ? xml_node(_root->next_sibling) : xml_node();
    }
 
    PUGI__FN xml_node xml_node::previous_sibling(const char_t* name_) const
    {
        if (!_root) return xml_node();
 
        for (xml_node_struct* i = _root->prev_sibling_c; i->next_sibling; i = i->prev_sibling_c)
            if (i->name && impl::strequal(name_, i->name)) return xml_node(i);
 
        return xml_node();
    }
 
    PUGI__FN xml_attribute xml_node::attribute(const char_t* name_, xml_attribute& hint_) const
    {
        xml_attribute_struct* hint = hint_._attr;
 
        // if hint is not an attribute of node, behavior is not defined
        assert(!hint || (_root && impl::is_attribute_of(hint, _root)));
 
        if (!_root) return xml_attribute();
 
        // optimistically search from hint up until the end
        for (xml_attribute_struct* i = hint; i; i = i->next_attribute)
            if (i->name && impl::strequal(name_, i->name))
            {
                // update hint to maximize efficiency of searching for consecutive attributes
                hint_._attr = i->next_attribute;
 
                return xml_attribute(i);
            }
 
        // wrap around and search from the first attribute until the hint
        // 'j' null pointer check is technically redundant, but it prevents a crash in case the assertion above fails
        for (xml_attribute_struct* j = _root->first_attribute; j && j != hint; j = j->next_attribute)
            if (j->name && impl::strequal(name_, j->name))
            {
                // update hint to maximize efficiency of searching for consecutive attributes
                hint_._attr = j->next_attribute;
 
                return xml_attribute(j);
            }
 
        return xml_attribute();
    }
 
    PUGI__FN xml_node xml_node::previous_sibling() const
    {
        if (!_root) return xml_node();
 
        if (_root->prev_sibling_c->next_sibling) return xml_node(_root->prev_sibling_c);
        else return xml_node();
    }
 
    PUGI__FN xml_node xml_node::parent() const
    {
        return _root ? xml_node(_root->parent) : xml_node();
    }
 
    PUGI__FN xml_node xml_node::root() const
    {
        return _root ? xml_node(&impl::get_document(_root)) : xml_node();
    }
 
    PUGI__FN xml_text xml_node::text() const
    {
        return xml_text(_root);
    }
 
    PUGI__FN const char_t* xml_node::child_value() const
    {
        if (!_root) return PUGIXML_TEXT("");
 
        // element nodes can have value if parse_embed_pcdata was used
        if (PUGI__NODETYPE(_root) == node_element && _root->value)
            return _root->value;
 
        for (xml_node_struct* i = _root->first_child; i; i = i->next_sibling)
            if (impl::is_text_node(i) && i->value)
                return i->value;
 
        return PUGIXML_TEXT("");
    }
 
    PUGI__FN const char_t* xml_node::child_value(const char_t* name_) const
    {
        return child(name_).child_value();
    }
 
    PUGI__FN xml_attribute xml_node::first_attribute() const
    {
        return _root ? xml_attribute(_root->first_attribute) : xml_attribute();
    }
 
    PUGI__FN xml_attribute xml_node::last_attribute() const
    {
        return _root && _root->first_attribute ? xml_attribute(_root->first_attribute->prev_attribute_c) : xml_attribute();
    }
 
    PUGI__FN xml_node xml_node::first_child() const
    {
        return _root ? xml_node(_root->first_child) : xml_node();
    }
 
    PUGI__FN xml_node xml_node::last_child() const
    {
        return _root && _root->first_child ? xml_node(_root->first_child->prev_sibling_c) : xml_node();
    }
 
    PUGI__FN bool xml_node::set_name(const char_t* rhs)
    {
        xml_node_type type_ = _root ? PUGI__NODETYPE(_root) : node_null;
 
        if (type_ != node_element && type_ != node_pi && type_ != node_declaration)
            return false;
 
        return impl::strcpy_insitu(_root->name, _root->header, impl::xml_memory_page_name_allocated_mask, rhs, impl::strlength(rhs));
    }
 
    PUGI__FN bool xml_node::set_value(const char_t* rhs)
    {
        xml_node_type type_ = _root ? PUGI__NODETYPE(_root) : node_null;
 
        if (type_ != node_pcdata && type_ != node_cdata && type_ != node_comment && type_ != node_pi && type_ != node_doctype)
            return false;
 
        return impl::strcpy_insitu(_root->value, _root->header, impl::xml_memory_page_value_allocated_mask, rhs, impl::strlength(rhs));
    }
 
    PUGI__FN xml_attribute xml_node::append_attribute(const char_t* name_)
    {
        if (!impl::allow_insert_attribute(type())) return xml_attribute();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_attribute();
 
        xml_attribute a(impl::allocate_attribute(alloc));
        if (!a) return xml_attribute();
 
        impl::append_attribute(a._attr, _root);
 
        a.set_name(name_);
 
        return a;
    }
 
    PUGI__FN xml_attribute xml_node::prepend_attribute(const char_t* name_)
    {
        if (!impl::allow_insert_attribute(type())) return xml_attribute();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_attribute();
 
        xml_attribute a(impl::allocate_attribute(alloc));
        if (!a) return xml_attribute();
 
        impl::prepend_attribute(a._attr, _root);
 
        a.set_name(name_);
 
        return a;
    }
 
    PUGI__FN xml_attribute xml_node::insert_attribute_after(const char_t* name_, const xml_attribute& attr)
    {
        if (!impl::allow_insert_attribute(type())) return xml_attribute();
        if (!attr || !impl::is_attribute_of(attr._attr, _root)) return xml_attribute();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_attribute();
 
        xml_attribute a(impl::allocate_attribute(alloc));
        if (!a) return xml_attribute();
 
        impl::insert_attribute_after(a._attr, attr._attr, _root);
 
        a.set_name(name_);
 
        return a;
    }
 
    PUGI__FN xml_attribute xml_node::insert_attribute_before(const char_t* name_, const xml_attribute& attr)
    {
        if (!impl::allow_insert_attribute(type())) return xml_attribute();
        if (!attr || !impl::is_attribute_of(attr._attr, _root)) return xml_attribute();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_attribute();
 
        xml_attribute a(impl::allocate_attribute(alloc));
        if (!a) return xml_attribute();
 
        impl::insert_attribute_before(a._attr, attr._attr, _root);
 
        a.set_name(name_);
 
        return a;
    }
 
    PUGI__FN xml_attribute xml_node::append_copy(const xml_attribute& proto)
    {
        if (!proto) return xml_attribute();
        if (!impl::allow_insert_attribute(type())) return xml_attribute();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_attribute();
 
        xml_attribute a(impl::allocate_attribute(alloc));
        if (!a) return xml_attribute();
 
        impl::append_attribute(a._attr, _root);
        impl::node_copy_attribute(a._attr, proto._attr);
 
        return a;
    }
 
    PUGI__FN xml_attribute xml_node::prepend_copy(const xml_attribute& proto)
    {
        if (!proto) return xml_attribute();
        if (!impl::allow_insert_attribute(type())) return xml_attribute();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_attribute();
 
        xml_attribute a(impl::allocate_attribute(alloc));
        if (!a) return xml_attribute();
 
        impl::prepend_attribute(a._attr, _root);
        impl::node_copy_attribute(a._attr, proto._attr);
 
        return a;
    }
 
    PUGI__FN xml_attribute xml_node::insert_copy_after(const xml_attribute& proto, const xml_attribute& attr)
    {
        if (!proto) return xml_attribute();
        if (!impl::allow_insert_attribute(type())) return xml_attribute();
        if (!attr || !impl::is_attribute_of(attr._attr, _root)) return xml_attribute();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_attribute();
 
        xml_attribute a(impl::allocate_attribute(alloc));
        if (!a) return xml_attribute();
 
        impl::insert_attribute_after(a._attr, attr._attr, _root);
        impl::node_copy_attribute(a._attr, proto._attr);
 
        return a;
    }
 
    PUGI__FN xml_attribute xml_node::insert_copy_before(const xml_attribute& proto, const xml_attribute& attr)
    {
        if (!proto) return xml_attribute();
        if (!impl::allow_insert_attribute(type())) return xml_attribute();
        if (!attr || !impl::is_attribute_of(attr._attr, _root)) return xml_attribute();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_attribute();
 
        xml_attribute a(impl::allocate_attribute(alloc));
        if (!a) return xml_attribute();
 
        impl::insert_attribute_before(a._attr, attr._attr, _root);
        impl::node_copy_attribute(a._attr, proto._attr);
 
        return a;
    }
 
    PUGI__FN xml_node xml_node::append_child(xml_node_type type_)
    {
        if (!impl::allow_insert_child(type(), type_)) return xml_node();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_node();
 
        xml_node n(impl::allocate_node(alloc, type_));
        if (!n) return xml_node();
 
        impl::append_node(n._root, _root);
 
        if (type_ == node_declaration) n.set_name(PUGIXML_TEXT("xml"));
 
        return n;
    }
 
    PUGI__FN xml_node xml_node::prepend_child(xml_node_type type_)
    {
        if (!impl::allow_insert_child(type(), type_)) return xml_node();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_node();
 
        xml_node n(impl::allocate_node(alloc, type_));
        if (!n) return xml_node();
 
        impl::prepend_node(n._root, _root);
 
        if (type_ == node_declaration) n.set_name(PUGIXML_TEXT("xml"));
 
        return n;
    }
 
    PUGI__FN xml_node xml_node::insert_child_before(xml_node_type type_, const xml_node& node)
    {
        if (!impl::allow_insert_child(type(), type_)) return xml_node();
        if (!node._root || node._root->parent != _root) return xml_node();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_node();
 
        xml_node n(impl::allocate_node(alloc, type_));
        if (!n) return xml_node();
 
        impl::insert_node_before(n._root, node._root);
 
        if (type_ == node_declaration) n.set_name(PUGIXML_TEXT("xml"));
 
        return n;
    }
 
    PUGI__FN xml_node xml_node::insert_child_after(xml_node_type type_, const xml_node& node)
    {
        if (!impl::allow_insert_child(type(), type_)) return xml_node();
        if (!node._root || node._root->parent != _root) return xml_node();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_node();
 
        xml_node n(impl::allocate_node(alloc, type_));
        if (!n) return xml_node();
 
        impl::insert_node_after(n._root, node._root);
 
        if (type_ == node_declaration) n.set_name(PUGIXML_TEXT("xml"));
 
        return n;
    }
 
    PUGI__FN xml_node xml_node::append_child(const char_t* name_)
    {
        xml_node result = append_child(node_element);
 
        result.set_name(name_);
 
        return result;
    }
 
    PUGI__FN xml_node xml_node::prepend_child(const char_t* name_)
    {
        xml_node result = prepend_child(node_element);
 
        result.set_name(name_);
 
        return result;
    }
 
    PUGI__FN xml_node xml_node::insert_child_after(const char_t* name_, const xml_node& node)
    {
        xml_node result = insert_child_after(node_element, node);
 
        result.set_name(name_);
 
        return result;
    }
 
    PUGI__FN xml_node xml_node::insert_child_before(const char_t* name_, const xml_node& node)
    {
        xml_node result = insert_child_before(node_element, node);
 
        result.set_name(name_);
 
        return result;
    }
 
    PUGI__FN xml_node xml_node::append_copy(const xml_node& proto)
    {
        xml_node_type type_ = proto.type();
        if (!impl::allow_insert_child(type(), type_)) return xml_node();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_node();
 
        xml_node n(impl::allocate_node(alloc, type_));
        if (!n) return xml_node();
 
        impl::append_node(n._root, _root);
        impl::node_copy_tree(n._root, proto._root);
 
        return n;
    }
 
    PUGI__FN xml_node xml_node::prepend_copy(const xml_node& proto)
    {
        xml_node_type type_ = proto.type();
        if (!impl::allow_insert_child(type(), type_)) return xml_node();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_node();
 
        xml_node n(impl::allocate_node(alloc, type_));
        if (!n) return xml_node();
 
        impl::prepend_node(n._root, _root);
        impl::node_copy_tree(n._root, proto._root);
 
        return n;
    }
 
    PUGI__FN xml_node xml_node::insert_copy_after(const xml_node& proto, const xml_node& node)
    {
        xml_node_type type_ = proto.type();
        if (!impl::allow_insert_child(type(), type_)) return xml_node();
        if (!node._root || node._root->parent != _root) return xml_node();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_node();
 
        xml_node n(impl::allocate_node(alloc, type_));
        if (!n) return xml_node();
 
        impl::insert_node_after(n._root, node._root);
        impl::node_copy_tree(n._root, proto._root);
 
        return n;
    }
 
    PUGI__FN xml_node xml_node::insert_copy_before(const xml_node& proto, const xml_node& node)
    {
        xml_node_type type_ = proto.type();
        if (!impl::allow_insert_child(type(), type_)) return xml_node();
        if (!node._root || node._root->parent != _root) return xml_node();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_node();
 
        xml_node n(impl::allocate_node(alloc, type_));
        if (!n) return xml_node();
 
        impl::insert_node_before(n._root, node._root);
        impl::node_copy_tree(n._root, proto._root);
 
        return n;
    }
 
    PUGI__FN xml_node xml_node::append_move(const xml_node& moved)
    {
        if (!impl::allow_move(*this, moved)) return xml_node();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_node();
 
        // disable document_buffer_order optimization since moving nodes around changes document order without changing buffer pointers
        impl::get_document(_root).header |= impl::xml_memory_page_contents_shared_mask;
 
        impl::remove_node(moved._root);
        impl::append_node(moved._root, _root);
 
        return moved;
    }
 
    PUGI__FN xml_node xml_node::prepend_move(const xml_node& moved)
    {
        if (!impl::allow_move(*this, moved)) return xml_node();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_node();
 
        // disable document_buffer_order optimization since moving nodes around changes document order without changing buffer pointers
        impl::get_document(_root).header |= impl::xml_memory_page_contents_shared_mask;
 
        impl::remove_node(moved._root);
        impl::prepend_node(moved._root, _root);
 
        return moved;
    }
 
    PUGI__FN xml_node xml_node::insert_move_after(const xml_node& moved, const xml_node& node)
    {
        if (!impl::allow_move(*this, moved)) return xml_node();
        if (!node._root || node._root->parent != _root) return xml_node();
        if (moved._root == node._root) return xml_node();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_node();
 
        // disable document_buffer_order optimization since moving nodes around changes document order without changing buffer pointers
        impl::get_document(_root).header |= impl::xml_memory_page_contents_shared_mask;
 
        impl::remove_node(moved._root);
        impl::insert_node_after(moved._root, node._root);
 
        return moved;
    }
 
    PUGI__FN xml_node xml_node::insert_move_before(const xml_node& moved, const xml_node& node)
    {
        if (!impl::allow_move(*this, moved)) return xml_node();
        if (!node._root || node._root->parent != _root) return xml_node();
        if (moved._root == node._root) return xml_node();
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return xml_node();
 
        // disable document_buffer_order optimization since moving nodes around changes document order without changing buffer pointers
        impl::get_document(_root).header |= impl::xml_memory_page_contents_shared_mask;
 
        impl::remove_node(moved._root);
        impl::insert_node_before(moved._root, node._root);
 
        return moved;
    }
 
    PUGI__FN bool xml_node::remove_attribute(const char_t* name_)
    {
        return remove_attribute(attribute(name_));
    }
 
    PUGI__FN bool xml_node::remove_attribute(const xml_attribute& a)
    {
        if (!_root || !a._attr) return false;
        if (!impl::is_attribute_of(a._attr, _root)) return false;
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return false;
 
        impl::remove_attribute(a._attr, _root);
        impl::destroy_attribute(a._attr, alloc);
 
        return true;
    }
 
    PUGI__FN bool xml_node::remove_child(const char_t* name_)
    {
        return remove_child(child(name_));
    }
 
    PUGI__FN bool xml_node::remove_child(const xml_node& n)
    {
        if (!_root || !n._root || n._root->parent != _root) return false;
 
        impl::xml_allocator& alloc = impl::get_allocator(_root);
        if (!alloc.reserve()) return false;
 
        impl::remove_node(n._root);
        impl::destroy_node(n._root, alloc);
 
        return true;
    }
 
    PUGI__FN xml_parse_result xml_node::append_buffer(const void* contents, size_t size, unsigned int options, xml_encoding encoding)
    {
        // append_buffer is only valid for elements/documents
        if (!impl::allow_insert_child(type(), node_element)) return impl::make_parse_result(status_append_invalid_root);
 
        // get document node
        impl::xml_document_struct* doc = &impl::get_document(_root);
 
        // disable document_buffer_order optimization since in a document with multiple buffers comparing buffer pointers does not make sense
        doc->header |= impl::xml_memory_page_contents_shared_mask;
 
        // get extra buffer element (we'll store the document fragment buffer there so that we can deallocate it later)
        impl::xml_memory_page* page = 0;
        impl::xml_extra_buffer* extra = static_cast<impl::xml_extra_buffer*>(doc->allocate_memory(sizeof(impl::xml_extra_buffer) + sizeof(void*), page));
        (void)page;
 
        if (!extra) return impl::make_parse_result(status_out_of_memory);
 
    #ifdef PUGIXML_COMPACT
        // align the memory block to a pointer boundary; this is required for compact mode where memory allocations are only 4b aligned
        // note that this requires up to sizeof(void*)-1 additional memory, which the allocation above takes into account
        extra = reinterpret_cast<impl::xml_extra_buffer*>((reinterpret_cast<uintptr_t>(extra) + (sizeof(void*) - 1)) & ~(sizeof(void*) - 1));
    #endif
 
        // add extra buffer to the list
        extra->buffer = 0;
        extra->next = doc->extra_buffers;
        doc->extra_buffers = extra;
 
        // name of the root has to be NULL before parsing - otherwise closing node mismatches will not be detected at the top level
        impl::name_null_sentry sentry(_root);
 
        return impl::load_buffer_impl(doc, _root, const_cast<void*>(contents), size, options, encoding, false, false, &extra->buffer);
    }
 
    PUGI__FN xml_node xml_node::find_child_by_attribute(const char_t* name_, const char_t* attr_name, const char_t* attr_value) const
    {
        if (!_root) return xml_node();
 
        for (xml_node_struct* i = _root->first_child; i; i = i->next_sibling)
            if (i->name && impl::strequal(name_, i->name))
            {
                for (xml_attribute_struct* a = i->first_attribute; a; a = a->next_attribute)
                    if (a->name && impl::strequal(attr_name, a->name) && impl::strequal(attr_value, a->value ? a->value + 0 : PUGIXML_TEXT("")))
                        return xml_node(i);
            }
 
        return xml_node();
    }
 
    PUGI__FN xml_node xml_node::find_child_by_attribute(const char_t* attr_name, const char_t* attr_value) const
    {
        if (!_root) return xml_node();
 
        for (xml_node_struct* i = _root->first_child; i; i = i->next_sibling)
            for (xml_attribute_struct* a = i->first_attribute; a; a = a->next_attribute)
                if (a->name && impl::strequal(attr_name, a->name) && impl::strequal(attr_value, a->value ? a->value + 0 : PUGIXML_TEXT("")))
                    return xml_node(i);
 
        return xml_node();
    }
 
#ifndef PUGIXML_NO_STL
    PUGI__FN string_t xml_node::path(char_t delimiter) const
    {
        if (!_root) return string_t();
 
        size_t offset = 0;
 
        for (xml_node_struct* i = _root; i; i = i->parent)
        {
            offset += (i != _root);
            offset += i->name ? impl::strlength(i->name) : 0;
        }
 
        string_t result;
        result.resize(offset);
 
        for (xml_node_struct* j = _root; j; j = j->parent)
        {
            if (j != _root)
                result[--offset] = delimiter;
 
            if (j->name)
            {
                size_t length = impl::strlength(j->name);
 
                offset -= length;
                memcpy(&result[offset], j->name, length * sizeof(char_t));
            }
        }
 
        assert(offset == 0);
 
        return result;
    }
#endif
 
    PUGI__FN xml_node xml_node::first_element_by_path(const char_t* path_, char_t delimiter) const
    {
        xml_node found = *this; // Current search context.
 
        if (!_root || !path_[0]) return found;
 
        if (path_[0] == delimiter)
        {
            // Absolute path; e.g. '/foo/bar'
            found = found.root();
            ++path_;
        }
 
        const char_t* path_segment = path_;
 
        while (*path_segment == delimiter) ++path_segment;
 
        const char_t* path_segment_end = path_segment;
 
        while (*path_segment_end && *path_segment_end != delimiter) ++path_segment_end;
 
        if (path_segment == path_segment_end) return found;
 
        const char_t* next_segment = path_segment_end;
 
        while (*next_segment == delimiter) ++next_segment;
 
        if (*path_segment == '.' && path_segment + 1 == path_segment_end)
            return found.first_element_by_path(next_segment, delimiter);
        else if (*path_segment == '.' && *(path_segment+1) == '.' && path_segment + 2 == path_segment_end)
            return found.parent().first_element_by_path(next_segment, delimiter);
        else
        {
            for (xml_node_struct* j = found._root->first_child; j; j = j->next_sibling)
            {
                if (j->name && impl::strequalrange(j->name, path_segment, static_cast<size_t>(path_segment_end - path_segment)))
                {
                    xml_node subsearch = xml_node(j).first_element_by_path(next_segment, delimiter);
 
                    if (subsearch) return subsearch;
                }
            }
 
            return xml_node();
        }
    }
 
    PUGI__FN bool xml_node::traverse(xml_tree_walker& walker)
    {
        walker._depth = -1;
 
        xml_node arg_begin(_root);
        if (!walker.begin(arg_begin)) return false;
 
        xml_node_struct* cur = _root ? _root->first_child + 0 : 0;
 
        if (cur)
        {
            ++walker._depth;
 
            do
            {
                xml_node arg_for_each(cur);
                if (!walker.for_each(arg_for_each))
                    return false;
 
                if (cur->first_child)
                {
                    ++walker._depth;
                    cur = cur->first_child;
                }
                else if (cur->next_sibling)
                    cur = cur->next_sibling;
                else
                {
                    while (!cur->next_sibling && cur != _root && cur->parent)
                    {
                        --walker._depth;
                        cur = cur->parent;
                    }
 
                    if (cur != _root)
                        cur = cur->next_sibling;
                }
            }
            while (cur && cur != _root);
        }
 
        assert(walker._depth == -1);
 
        xml_node arg_end(_root);
        return walker.end(arg_end);
    }
 
    PUGI__FN size_t xml_node::hash_value() const
    {
        return static_cast<size_t>(reinterpret_cast<uintptr_t>(_root) / sizeof(xml_node_struct));
    }
 
    PUGI__FN xml_node_struct* xml_node::internal_object() const
    {
        return _root;
    }
 
    PUGI__FN void xml_node::print(xml_writer& writer, const char_t* indent, unsigned int flags, xml_encoding encoding, unsigned int depth) const
    {
        if (!_root) return;
 
        impl::xml_buffered_writer buffered_writer(writer, encoding);
 
        impl::node_output(buffered_writer, _root, indent, flags, depth);
 
        buffered_writer.flush();
    }
 
#ifndef PUGIXML_NO_STL
    PUGI__FN void xml_node::print(std::basic_ostream<char, std::char_traits<char> >& stream, const char_t* indent, unsigned int flags, xml_encoding encoding, unsigned int depth) const
    {
        xml_writer_stream writer(stream);
 
        print(writer, indent, flags, encoding, depth);
    }
 
    PUGI__FN void xml_node::print(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >& stream, const char_t* indent, unsigned int flags, unsigned int depth) const
    {
        xml_writer_stream writer(stream);
 
        print(writer, indent, flags, encoding_wchar, depth);
    }
#endif
 
    PUGI__FN ptrdiff_t xml_node::offset_debug() const
    {
        if (!_root) return -1;
 
        impl::xml_document_struct& doc = impl::get_document(_root);
 
        // we can determine the offset reliably only if there is exactly once parse buffer
        if (!doc.buffer || doc.extra_buffers) return -1;
 
        switch (type())
        {
        case node_document:
            return 0;
 
        case node_element:
        case node_declaration:
        case node_pi:
            return _root->name && (_root->header & impl::xml_memory_page_name_allocated_or_shared_mask) == 0 ? _root->name - doc.buffer : -1;
 
        case node_pcdata:
        case node_cdata:
        case node_comment:
        case node_doctype:
            return _root->value && (_root->header & impl::xml_memory_page_value_allocated_or_shared_mask) == 0 ? _root->value - doc.buffer : -1;
 
        default:
            assert(false && "Invalid node type"); // unreachable
            return -1;
        }
    }
 
#ifdef __BORLANDC__
    PUGI__FN bool operator&&(const xml_node& lhs, bool rhs)
    {
        return (bool)lhs && rhs;
    }
 
    PUGI__FN bool operator||(const xml_node& lhs, bool rhs)
    {
        return (bool)lhs || rhs;
    }
#endif
 
    PUGI__FN xml_text::xml_text(xml_node_struct* root): _root(root)
    {
    }
 
    PUGI__FN xml_node_struct* xml_text::_data() const
    {
        if (!_root || impl::is_text_node(_root)) return _root;
 
        // element nodes can have value if parse_embed_pcdata was used
        if (PUGI__NODETYPE(_root) == node_element && _root->value)
            return _root;
 
        for (xml_node_struct* node = _root->first_child; node; node = node->next_sibling)
            if (impl::is_text_node(node))
                return node;
 
        return 0;
    }
 
    PUGI__FN xml_node_struct* xml_text::_data_new()
    {
        xml_node_struct* d = _data();
        if (d) return d;
 
        return xml_node(_root).append_child(node_pcdata).internal_object();
    }
 
    PUGI__FN xml_text::xml_text(): _root(0)
    {
    }
 
    PUGI__FN static void unspecified_bool_xml_text(xml_text***)
    {
    }
 
    PUGI__FN xml_text::operator xml_text::unspecified_bool_type() const
    {
        return _data() ? unspecified_bool_xml_text : 0;
    }
 
    PUGI__FN bool xml_text::operator!() const
    {
        return !_data();
    }
 
    PUGI__FN bool xml_text::empty() const
    {
        return _data() == 0;
    }
 
    PUGI__FN const char_t* xml_text::get() const
    {
        xml_node_struct* d = _data();
 
        return (d && d->value) ? d->value + 0 : PUGIXML_TEXT("");
    }
 
    PUGI__FN const char_t* xml_text::as_string(const char_t* def) const
    {
        xml_node_struct* d = _data();
 
        return (d && d->value) ? d->value + 0 : def;
    }
 
    PUGI__FN int xml_text::as_int(int def) const
    {
        xml_node_struct* d = _data();
 
        return (d && d->value) ? impl::get_value_int(d->value) : def;
    }
 
    PUGI__FN unsigned int xml_text::as_uint(unsigned int def) const
    {
        xml_node_struct* d = _data();
 
        return (d && d->value) ? impl::get_value_uint(d->value) : def;
    }
 
    PUGI__FN double xml_text::as_double(double def) const
    {
        xml_node_struct* d = _data();
 
        return (d && d->value) ? impl::get_value_double(d->value) : def;
    }
 
    PUGI__FN float xml_text::as_float(float def) const
    {
        xml_node_struct* d = _data();
 
        return (d && d->value) ? impl::get_value_float(d->value) : def;
    }
 
    PUGI__FN bool xml_text::as_bool(bool def) const
    {
        xml_node_struct* d = _data();
 
        return (d && d->value) ? impl::get_value_bool(d->value) : def;
    }
 
#ifdef PUGIXML_HAS_LONG_LONG
    PUGI__FN long long xml_text::as_llong(long long def) const
    {
        xml_node_struct* d = _data();
 
        return (d && d->value) ? impl::get_value_llong(d->value) : def;
    }
 
    PUGI__FN unsigned long long xml_text::as_ullong(unsigned long long def) const
    {
        xml_node_struct* d = _data();
 
        return (d && d->value) ? impl::get_value_ullong(d->value) : def;
    }
#endif
 
    PUGI__FN bool xml_text::set(const char_t* rhs)
    {
        xml_node_struct* dn = _data_new();
 
        return dn ? impl::strcpy_insitu(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs, impl::strlength(rhs)) : false;
    }
 
    PUGI__FN bool xml_text::set(int rhs)
    {
        xml_node_struct* dn = _data_new();
 
        return dn ? impl::set_value_integer<unsigned int>(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs, rhs < 0) : false;
    }
 
    PUGI__FN bool xml_text::set(unsigned int rhs)
    {
        xml_node_struct* dn = _data_new();
 
        return dn ? impl::set_value_integer<unsigned int>(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs, false) : false;
    }
 
    PUGI__FN bool xml_text::set(long rhs)
    {
        xml_node_struct* dn = _data_new();
 
        return dn ? impl::set_value_integer<unsigned long>(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs, rhs < 0) : false;
    }
 
    PUGI__FN bool xml_text::set(unsigned long rhs)
    {
        xml_node_struct* dn = _data_new();
 
        return dn ? impl::set_value_integer<unsigned long>(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs, false) : false;
    }
 
    PUGI__FN bool xml_text::set(float rhs)
    {
        xml_node_struct* dn = _data_new();
 
        return dn ? impl::set_value_convert(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs) : false;
    }
 
    PUGI__FN bool xml_text::set(double rhs)
    {
        xml_node_struct* dn = _data_new();
 
        return dn ? impl::set_value_convert(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs) : false;
    }
 
    PUGI__FN bool xml_text::set(bool rhs)
    {
        xml_node_struct* dn = _data_new();
 
        return dn ? impl::set_value_bool(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs) : false;
    }
 
#ifdef PUGIXML_HAS_LONG_LONG
    PUGI__FN bool xml_text::set(long long rhs)
    {
        xml_node_struct* dn = _data_new();
 
        return dn ? impl::set_value_integer<unsigned long long>(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs, rhs < 0) : false;
    }
 
    PUGI__FN bool xml_text::set(unsigned long long rhs)
    {
        xml_node_struct* dn = _data_new();
 
        return dn ? impl::set_value_integer<unsigned long long>(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs, false) : false;
    }
#endif
 
    PUGI__FN xml_text& xml_text::operator=(const char_t* rhs)
    {
        set(rhs);
        return *this;
    }
 
    PUGI__FN xml_text& xml_text::operator=(int rhs)
    {
        set(rhs);
        return *this;
    }
 
    PUGI__FN xml_text& xml_text::operator=(unsigned int rhs)
    {
        set(rhs);
        return *this;
    }
 
    PUGI__FN xml_text& xml_text::operator=(long rhs)
    {
        set(rhs);
        return *this;
    }
 
    PUGI__FN xml_text& xml_text::operator=(unsigned long rhs)
    {
        set(rhs);
        return *this;
    }
 
    PUGI__FN xml_text& xml_text::operator=(double rhs)
    {
        set(rhs);
        return *this;
    }
 
    PUGI__FN xml_text& xml_text::operator=(float rhs)
    {
        set(rhs);
        return *this;
    }
 
    PUGI__FN xml_text& xml_text::operator=(bool rhs)
    {
        set(rhs);
        return *this;
    }
 
#ifdef PUGIXML_HAS_LONG_LONG
    PUGI__FN xml_text& xml_text::operator=(long long rhs)
    {
        set(rhs);
        return *this;
    }
 
    PUGI__FN xml_text& xml_text::operator=(unsigned long long rhs)
    {
        set(rhs);
        return *this;
    }
#endif
 
    PUGI__FN xml_node xml_text::data() const
    {
        return xml_node(_data());
    }
 
#ifdef __BORLANDC__
    PUGI__FN bool operator&&(const xml_text& lhs, bool rhs)
    {
        return (bool)lhs && rhs;
    }
 
    PUGI__FN bool operator||(const xml_text& lhs, bool rhs)
    {
        return (bool)lhs || rhs;
    }
#endif
 
    PUGI__FN xml_node_iterator::xml_node_iterator()
    {
    }
 
    PUGI__FN xml_node_iterator::xml_node_iterator(const xml_node& node): _wrap(node), _parent(node.parent())
    {
    }
 
    PUGI__FN xml_node_iterator::xml_node_iterator(xml_node_struct* ref, xml_node_struct* parent): _wrap(ref), _parent(parent)
    {
    }
 
    PUGI__FN bool xml_node_iterator::operator==(const xml_node_iterator& rhs) const
    {
        return _wrap._root == rhs._wrap._root && _parent._root == rhs._parent._root;
    }
 
    PUGI__FN bool xml_node_iterator::operator!=(const xml_node_iterator& rhs) const
    {
        return _wrap._root != rhs._wrap._root || _parent._root != rhs._parent._root;
    }
 
    PUGI__FN xml_node& xml_node_iterator::operator*() const
    {
        assert(_wrap._root);
        return _wrap;
    }
 
    PUGI__FN xml_node* xml_node_iterator::operator->() const
    {
        assert(_wrap._root);
        return const_cast<xml_node*>(&_wrap); // BCC5 workaround
    }
 
    PUGI__FN const xml_node_iterator& xml_node_iterator::operator++()
    {
        assert(_wrap._root);
        _wrap._root = _wrap._root->next_sibling;
        return *this;
    }
 
    PUGI__FN xml_node_iterator xml_node_iterator::operator++(int)
    {
        xml_node_iterator temp = *this;
        ++*this;
        return temp;
    }
 
    PUGI__FN const xml_node_iterator& xml_node_iterator::operator--()
    {
        _wrap = _wrap._root ? _wrap.previous_sibling() : _parent.last_child();
        return *this;
    }
 
    PUGI__FN xml_node_iterator xml_node_iterator::operator--(int)
    {
        xml_node_iterator temp = *this;
        --*this;
        return temp;
    }
 
    PUGI__FN xml_attribute_iterator::xml_attribute_iterator()
    {
    }
 
    PUGI__FN xml_attribute_iterator::xml_attribute_iterator(const xml_attribute& attr, const xml_node& parent): _wrap(attr), _parent(parent)
    {
    }
 
    PUGI__FN xml_attribute_iterator::xml_attribute_iterator(xml_attribute_struct* ref, xml_node_struct* parent): _wrap(ref), _parent(parent)
    {
    }
 
    PUGI__FN bool xml_attribute_iterator::operator==(const xml_attribute_iterator& rhs) const
    {
        return _wrap._attr == rhs._wrap._attr && _parent._root == rhs._parent._root;
    }
 
    PUGI__FN bool xml_attribute_iterator::operator!=(const xml_attribute_iterator& rhs) const
    {
        return _wrap._attr != rhs._wrap._attr || _parent._root != rhs._parent._root;
    }
 
    PUGI__FN xml_attribute& xml_attribute_iterator::operator*() const
    {
        assert(_wrap._attr);
        return _wrap;
    }
 
    PUGI__FN xml_attribute* xml_attribute_iterator::operator->() const
    {
        assert(_wrap._attr);
        return const_cast<xml_attribute*>(&_wrap); // BCC5 workaround
    }
 
    PUGI__FN const xml_attribute_iterator& xml_attribute_iterator::operator++()
    {
        assert(_wrap._attr);
        _wrap._attr = _wrap._attr->next_attribute;
        return *this;
    }
 
    PUGI__FN xml_attribute_iterator xml_attribute_iterator::operator++(int)
    {
        xml_attribute_iterator temp = *this;
        ++*this;
        return temp;
    }
 
    PUGI__FN const xml_attribute_iterator& xml_attribute_iterator::operator--()
    {
        _wrap = _wrap._attr ? _wrap.previous_attribute() : _parent.last_attribute();
        return *this;
    }
 
    PUGI__FN xml_attribute_iterator xml_attribute_iterator::operator--(int)
    {
        xml_attribute_iterator temp = *this;
        --*this;
        return temp;
    }
 
    PUGI__FN xml_named_node_iterator::xml_named_node_iterator(): _name(0)
    {
    }
 
    PUGI__FN xml_named_node_iterator::xml_named_node_iterator(const xml_node& node, const char_t* name): _wrap(node), _parent(node.parent()), _name(name)
    {
    }
 
    PUGI__FN xml_named_node_iterator::xml_named_node_iterator(xml_node_struct* ref, xml_node_struct* parent, const char_t* name): _wrap(ref), _parent(parent), _name(name)
    {
    }
 
    PUGI__FN bool xml_named_node_iterator::operator==(const xml_named_node_iterator& rhs) const
    {
        return _wrap._root == rhs._wrap._root && _parent._root == rhs._parent._root;
    }
 
    PUGI__FN bool xml_named_node_iterator::operator!=(const xml_named_node_iterator& rhs) const
    {
        return _wrap._root != rhs._wrap._root || _parent._root != rhs._parent._root;
    }
 
    PUGI__FN xml_node& xml_named_node_iterator::operator*() const
    {
        assert(_wrap._root);
        return _wrap;
    }
 
    PUGI__FN xml_node* xml_named_node_iterator::operator->() const
    {
        assert(_wrap._root);
        return const_cast<xml_node*>(&_wrap); // BCC5 workaround
    }
 
    PUGI__FN const xml_named_node_iterator& xml_named_node_iterator::operator++()
    {
        assert(_wrap._root);
        _wrap = _wrap.next_sibling(_name);
        return *this;
    }
 
    PUGI__FN xml_named_node_iterator xml_named_node_iterator::operator++(int)
    {
        xml_named_node_iterator temp = *this;
        ++*this;
        return temp;
    }
 
    PUGI__FN const xml_named_node_iterator& xml_named_node_iterator::operator--()
    {
        if (_wrap._root)
            _wrap = _wrap.previous_sibling(_name);
        else
        {
            _wrap = _parent.last_child();
 
            if (!impl::strequal(_wrap.name(), _name))
                _wrap = _wrap.previous_sibling(_name);
        }
 
        return *this;
    }
 
    PUGI__FN xml_named_node_iterator xml_named_node_iterator::operator--(int)
    {
        xml_named_node_iterator temp = *this;
        --*this;
        return temp;
    }
 
    PUGI__FN xml_parse_result::xml_parse_result(): status(status_internal_error), offset(0), encoding(encoding_auto)
    {
    }
 
    PUGI__FN xml_parse_result::operator bool() const
    {
        return status == status_ok;
    }
 
    PUGI__FN const char* xml_parse_result::description() const
    {
        switch (status)
        {
        case status_ok: return "No error";
 
        case status_file_not_found: return "File was not found";
        case status_io_error: return "Error reading from file/stream";
        case status_out_of_memory: return "Could not allocate memory";
        case status_internal_error: return "Internal error occurred";
 
        case status_unrecognized_tag: return "Could not determine tag type";
 
        case status_bad_pi: return "Error parsing document declaration/processing instruction";
        case status_bad_comment: return "Error parsing comment";
        case status_bad_cdata: return "Error parsing CDATA section";
        case status_bad_doctype: return "Error parsing document type declaration";
        case status_bad_pcdata: return "Error parsing PCDATA section";
        case status_bad_start_element: return "Error parsing start element tag";
        case status_bad_attribute: return "Error parsing element attribute";
        case status_bad_end_element: return "Error parsing end element tag";
        case status_end_element_mismatch: return "Start-end tags mismatch";
 
        case status_append_invalid_root: return "Unable to append nodes: root is not an element or document";
 
        case status_no_document_element: return "No document element found";
 
        default: return "Unknown error";
        }
    }
 
    PUGI__FN xml_document::xml_document(): _buffer(0)
    {
        _create();
    }
 
    PUGI__FN xml_document::~xml_document()
    {
        _destroy();
    }
 
#ifdef PUGIXML_HAS_MOVE
    PUGI__FN xml_document::xml_document(xml_document&& rhs) PUGIXML_NOEXCEPT_IF_NOT_COMPACT: _buffer(0)
    {
        _create();
        _move(rhs);
    }
 
    PUGI__FN xml_document& xml_document::operator=(xml_document&& rhs) PUGIXML_NOEXCEPT_IF_NOT_COMPACT
    {
        if (this == &rhs) return *this;
 
        _destroy();
        _create();
        _move(rhs);
 
        return *this;
    }
#endif
 
    PUGI__FN void xml_document::reset()
    {
        _destroy();
        _create();
    }
 
    PUGI__FN void xml_document::reset(const xml_document& proto)
    {
        reset();
 
        for (xml_node cur = proto.first_child(); cur; cur = cur.next_sibling())
            append_copy(cur);
    }
 
    PUGI__FN void xml_document::_create()
    {
        assert(!_root);
 
    #ifdef PUGIXML_COMPACT
        // space for page marker for the first page (uint32_t), rounded up to pointer size; assumes pointers are at least 32-bit
        const size_t page_offset = sizeof(void*);
    #else
        const size_t page_offset = 0;
    #endif
 
        // initialize sentinel page
        PUGI__STATIC_ASSERT(sizeof(impl::xml_memory_page) + sizeof(impl::xml_document_struct) + page_offset <= sizeof(_memory));
 
        // prepare page structure
        impl::xml_memory_page* page = impl::xml_memory_page::construct(_memory);
        assert(page);
 
        page->busy_size = impl::xml_memory_page_size;
 
        // setup first page marker
    #ifdef PUGIXML_COMPACT
        // round-trip through void* to avoid 'cast increases required alignment of target type' warning
        page->compact_page_marker = reinterpret_cast<uint32_t*>(static_cast<void*>(reinterpret_cast<char*>(page) + sizeof(impl::xml_memory_page)));
        *page->compact_page_marker = sizeof(impl::xml_memory_page);
    #endif
 
        // allocate new root
        _root = new (reinterpret_cast<char*>(page) + sizeof(impl::xml_memory_page) + page_offset) impl::xml_document_struct(page);
        _root->prev_sibling_c = _root;
 
        // setup sentinel page
        page->allocator = static_cast<impl::xml_document_struct*>(_root);
 
        // setup hash table pointer in allocator
    #ifdef PUGIXML_COMPACT
        page->allocator->_hash = &static_cast<impl::xml_document_struct*>(_root)->hash;
    #endif
 
        // verify the document allocation
        assert(reinterpret_cast<char*>(_root) + sizeof(impl::xml_document_struct) <= _memory + sizeof(_memory));
    }
 
    PUGI__FN void xml_document::_destroy()
    {
        assert(_root);
 
        // destroy static storage
        if (_buffer)
        {
            impl::xml_memory::deallocate(_buffer);
            _buffer = 0;
        }
 
        // destroy extra buffers (note: no need to destroy linked list nodes, they're allocated using document allocator)
        for (impl::xml_extra_buffer* extra = static_cast<impl::xml_document_struct*>(_root)->extra_buffers; extra; extra = extra->next)
        {
            if (extra->buffer) impl::xml_memory::deallocate(extra->buffer);
        }
 
        // destroy dynamic storage, leave sentinel page (it's in static memory)
        impl::xml_memory_page* root_page = PUGI__GETPAGE(_root);
        assert(root_page && !root_page->prev);
        assert(reinterpret_cast<char*>(root_page) >= _memory && reinterpret_cast<char*>(root_page) < _memory + sizeof(_memory));
 
        for (impl::xml_memory_page* page = root_page->next; page; )
        {
            impl::xml_memory_page* next = page->next;
 
            impl::xml_allocator::deallocate_page(page);
 
            page = next;
        }
 
    #ifdef PUGIXML_COMPACT
        // destroy hash table
        static_cast<impl::xml_document_struct*>(_root)->hash.clear();
    #endif
 
        _root = 0;
    }
 
#ifdef PUGIXML_HAS_MOVE
    PUGI__FN void xml_document::_move(xml_document& rhs) PUGIXML_NOEXCEPT_IF_NOT_COMPACT
    {
        impl::xml_document_struct* doc = static_cast<impl::xml_document_struct*>(_root);
        impl::xml_document_struct* other = static_cast<impl::xml_document_struct*>(rhs._root);
 
        // save first child pointer for later; this needs hash access
        xml_node_struct* other_first_child = other->first_child;
 
    #ifdef PUGIXML_COMPACT
        // reserve space for the hash table up front; this is the only operation that can fail
        // if it does, we have no choice but to throw (if we have exceptions)
        if (other_first_child)
        {
            size_t other_children = 0;
            for (xml_node_struct* node = other_first_child; node; node = node->next_sibling)
                other_children++;
 
            // in compact mode, each pointer assignment could result in a hash table request
            // during move, we have to relocate document first_child and parents of all children
            // normally there's just one child and its parent has a pointerless encoding but
            // we assume the worst here
            if (!other->_hash->reserve(other_children + 1))
            {
            #ifdef PUGIXML_NO_EXCEPTIONS
                return;
            #else
                throw std::bad_alloc();
            #endif
            }
        }
    #endif
 
        // move allocation state
        doc->_root = other->_root;
        doc->_busy_size = other->_busy_size;
 
        // move buffer state
        doc->buffer = other->buffer;
        doc->extra_buffers = other->extra_buffers;
        _buffer = rhs._buffer;
 
    #ifdef PUGIXML_COMPACT
        // move compact hash; note that the hash table can have pointers to other but they will be "inactive", similarly to nodes removed with remove_child
        doc->hash = other->hash;
        doc->_hash = &doc->hash;
 
        // make sure we don't access other hash up until the end when we reinitialize other document
        other->_hash = 0;
    #endif
 
        // move page structure
        impl::xml_memory_page* doc_page = PUGI__GETPAGE(doc);
        assert(doc_page && !doc_page->prev && !doc_page->next);
 
        impl::xml_memory_page* other_page = PUGI__GETPAGE(other);
        assert(other_page && !other_page->prev);
 
        // relink pages since root page is embedded into xml_document
        if (impl::xml_memory_page* page = other_page->next)
        {
            assert(page->prev == other_page);
 
            page->prev = doc_page;
 
            doc_page->next = page;
            other_page->next = 0;
        }
 
        // make sure pages point to the correct document state
        for (impl::xml_memory_page* page = doc_page->next; page; page = page->next)
        {
            assert(page->allocator == other);
 
            page->allocator = doc;
 
        #ifdef PUGIXML_COMPACT
            // this automatically migrates most children between documents and prevents ->parent assignment from allocating
            if (page->compact_shared_parent == other)
                page->compact_shared_parent = doc;
        #endif
        }
 
        // move tree structure
        assert(!doc->first_child);
 
        doc->first_child = other_first_child;
 
        for (xml_node_struct* node = other_first_child; node; node = node->next_sibling)
        {
        #ifdef PUGIXML_COMPACT
            // most children will have migrated when we reassigned compact_shared_parent
            assert(node->parent == other || node->parent == doc);
 
            node->parent = doc;
        #else
            assert(node->parent == other);
            node->parent = doc;
        #endif
        }
 
        // reset other document
        new (other) impl::xml_document_struct(PUGI__GETPAGE(other));
        rhs._buffer = 0;
    }
#endif
 
#ifndef PUGIXML_NO_STL
    PUGI__FN xml_parse_result xml_document::load(std::basic_istream<char, std::char_traits<char> >& stream, unsigned int options, xml_encoding encoding)
    {
        reset();
 
        return impl::load_stream_impl(static_cast<impl::xml_document_struct*>(_root), stream, options, encoding, &_buffer);
    }
 
    PUGI__FN xml_parse_result xml_document::load(std::basic_istream<wchar_t, std::char_traits<wchar_t> >& stream, unsigned int options)
    {
        reset();
 
        return impl::load_stream_impl(static_cast<impl::xml_document_struct*>(_root), stream, options, encoding_wchar, &_buffer);
    }
#endif
 
    PUGI__FN xml_parse_result xml_document::load_string(const char_t* contents, unsigned int options)
    {
        // Force native encoding (skip autodetection)
    #ifdef PUGIXML_WCHAR_MODE
        xml_encoding encoding = encoding_wchar;
    #else
        xml_encoding encoding = encoding_utf8;
    #endif
 
        return load_buffer(contents, impl::strlength(contents) * sizeof(char_t), options, encoding);
    }
 
    PUGI__FN xml_parse_result xml_document::load(const char_t* contents, unsigned int options)
    {
        return load_string(contents, options);
    }
 
    PUGI__FN xml_parse_result xml_document::load_file(const char* path_, unsigned int options, xml_encoding encoding)
    {
        reset();
 
        using impl::auto_deleter; // MSVC7 workaround
        auto_deleter<FILE> file(fopen(path_, "rb"), impl::close_file);
 
        return impl::load_file_impl(static_cast<impl::xml_document_struct*>(_root), file.data, options, encoding, &_buffer);
    }
 
    PUGI__FN xml_parse_result xml_document::load_file(const wchar_t* path_, unsigned int options, xml_encoding encoding)
    {
        reset();
 
        using impl::auto_deleter; // MSVC7 workaround
        auto_deleter<FILE> file(impl::open_file_wide(path_, L"rb"), impl::close_file);
 
        return impl::load_file_impl(static_cast<impl::xml_document_struct*>(_root), file.data, options, encoding, &_buffer);
    }
 
    PUGI__FN xml_parse_result xml_document::load_buffer(const void* contents, size_t size, unsigned int options, xml_encoding encoding)
    {
        reset();
 
        return impl::load_buffer_impl(static_cast<impl::xml_document_struct*>(_root), _root, const_cast<void*>(contents), size, options, encoding, false, false, &_buffer);
    }
 
    PUGI__FN xml_parse_result xml_document::load_buffer_inplace(void* contents, size_t size, unsigned int options, xml_encoding encoding)
    {
        reset();
 
        return impl::load_buffer_impl(static_cast<impl::xml_document_struct*>(_root), _root, contents, size, options, encoding, true, false, &_buffer);
    }
 
    PUGI__FN xml_parse_result xml_document::load_buffer_inplace_own(void* contents, size_t size, unsigned int options, xml_encoding encoding)
    {
        reset();
 
        return impl::load_buffer_impl(static_cast<impl::xml_document_struct*>(_root), _root, contents, size, options, encoding, true, true, &_buffer);
    }
 
    PUGI__FN void xml_document::save(xml_writer& writer, const char_t* indent, unsigned int flags, xml_encoding encoding) const
    {
        impl::xml_buffered_writer buffered_writer(writer, encoding);
 
        if ((flags & format_write_bom) && encoding != encoding_latin1)
        {
            // BOM always represents the codepoint U+FEFF, so just write it in native encoding
        #ifdef PUGIXML_WCHAR_MODE
            unsigned int bom = 0xfeff;
            buffered_writer.write(static_cast<wchar_t>(bom));
        #else
            buffered_writer.write('\xef', '\xbb', '\xbf');
        #endif
        }
 
        if (!(flags & format_no_declaration) && !impl::has_declaration(_root))
        {
            buffered_writer.write_string(PUGIXML_TEXT("<?xml version=\"1.0\""));
            if (encoding == encoding_latin1) buffered_writer.write_string(PUGIXML_TEXT(" encoding=\"ISO-8859-1\""));
            buffered_writer.write('?', '>');
            if (!(flags & format_raw)) buffered_writer.write('\n');
        }
 
        impl::node_output(buffered_writer, _root, indent, flags, 0);
 
        buffered_writer.flush();
    }
 
#ifndef PUGIXML_NO_STL
    PUGI__FN void xml_document::save(std::basic_ostream<char, std::char_traits<char> >& stream, const char_t* indent, unsigned int flags, xml_encoding encoding) const
    {
        xml_writer_stream writer(stream);
 
        save(writer, indent, flags, encoding);
    }
 
    PUGI__FN void xml_document::save(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >& stream, const char_t* indent, unsigned int flags) const
    {
        xml_writer_stream writer(stream);
 
        save(writer, indent, flags, encoding_wchar);
    }
#endif
 
    PUGI__FN bool xml_document::save_file(const char* path_, const char_t* indent, unsigned int flags, xml_encoding encoding) const
    {
        using impl::auto_deleter; // MSVC7 workaround
        auto_deleter<FILE> file(fopen(path_, (flags & format_save_file_text) ? "w" : "wb"), impl::close_file);
 
        return impl::save_file_impl(*this, file.data, indent, flags, encoding);
    }
 
    PUGI__FN bool xml_document::save_file(const wchar_t* path_, const char_t* indent, unsigned int flags, xml_encoding encoding) const
    {
        using impl::auto_deleter; // MSVC7 workaround
        auto_deleter<FILE> file(impl::open_file_wide(path_, (flags & format_save_file_text) ? L"w" : L"wb"), impl::close_file);
 
        return impl::save_file_impl(*this, file.data, indent, flags, encoding);
    }
 
    PUGI__FN xml_node xml_document::document_element() const
    {
        assert(_root);
 
        for (xml_node_struct* i = _root->first_child; i; i = i->next_sibling)
            if (PUGI__NODETYPE(i) == node_element)
                return xml_node(i);
 
        return xml_node();
    }
 
#ifndef PUGIXML_NO_STL
    PUGI__FN std::string PUGIXML_FUNCTION as_utf8(const wchar_t* str)
    {
        assert(str);
 
        return impl::as_utf8_impl(str, impl::strlength_wide(str));
    }
 
    PUGI__FN std::string PUGIXML_FUNCTION as_utf8(const std::basic_string<wchar_t>& str)
    {
        return impl::as_utf8_impl(str.c_str(), str.size());
    }
 
    PUGI__FN std::basic_string<wchar_t> PUGIXML_FUNCTION as_wide(const char* str)
    {
        assert(str);
 
        return impl::as_wide_impl(str, strlen(str));
    }
 
    PUGI__FN std::basic_string<wchar_t> PUGIXML_FUNCTION as_wide(const std::string& str)
    {
        return impl::as_wide_impl(str.c_str(), str.size());
    }
#endif
 
    PUGI__FN void PUGIXML_FUNCTION set_memory_management_functions(allocation_function allocate, deallocation_function deallocate)
    {
        impl::xml_memory::allocate = allocate;
        impl::xml_memory::deallocate = deallocate;
    }
 
    PUGI__FN allocation_function PUGIXML_FUNCTION get_memory_allocation_function()
    {
        return impl::xml_memory::allocate;
    }
 
    PUGI__FN deallocation_function PUGIXML_FUNCTION get_memory_deallocation_function()
    {
        return impl::xml_memory::deallocate;
    }
}
 
#if !defined(PUGIXML_NO_STL) && (defined(_MSC_VER) || defined(__ICC))
namespace std
{
    // Workarounds for (non-standard) iterator category detection for older versions (MSVC7/IC8 and earlier)
    PUGI__FN std::bidirectional_iterator_tag _Iter_cat(const pugi::xml_node_iterator&)
    {
        return std::bidirectional_iterator_tag();
    }
 
    PUGI__FN std::bidirectional_iterator_tag _Iter_cat(const pugi::xml_attribute_iterator&)
    {
        return std::bidirectional_iterator_tag();
    }
 
    PUGI__FN std::bidirectional_iterator_tag _Iter_cat(const pugi::xml_named_node_iterator&)
    {
        return std::bidirectional_iterator_tag();
    }
}
#endif
 
#if !defined(PUGIXML_NO_STL) && defined(__SUNPRO_CC)
namespace std
{
    // Workarounds for (non-standard) iterator category detection
    PUGI__FN std::bidirectional_iterator_tag __iterator_category(const pugi::xml_node_iterator&)
    {
        return std::bidirectional_iterator_tag();
    }
 
    PUGI__FN std::bidirectional_iterator_tag __iterator_category(const pugi::xml_attribute_iterator&)
    {
        return std::bidirectional_iterator_tag();
    }
 
    PUGI__FN std::bidirectional_iterator_tag __iterator_category(const pugi::xml_named_node_iterator&)
    {
        return std::bidirectional_iterator_tag();
    }
}
#endif
 
#ifndef PUGIXML_NO_XPATH
// STL replacements
PUGI__NS_BEGIN
    struct equal_to
    {
        template <typename T> bool operator()(const T& lhs, const T& rhs) const
        {
            return lhs == rhs;
        }
    };
 
    struct not_equal_to
    {
        template <typename T> bool operator()(const T& lhs, const T& rhs) const
        {
            return lhs != rhs;
        }
    };
 
    struct less
    {
        template <typename T> bool operator()(const T& lhs, const T& rhs) const
        {
            return lhs < rhs;
        }
    };
 
    struct less_equal
    {
        template <typename T> bool operator()(const T& lhs, const T& rhs) const
        {
            return lhs <= rhs;
        }
    };
 
    template <typename T> void swap(T& lhs, T& rhs)
    {
        T temp = lhs;
        lhs = rhs;
        rhs = temp;
    }
 
    template <typename I, typename Pred> I min_element(I begin, I end, const Pred& pred)
    {
        I result = begin;
 
        for (I it = begin + 1; it != end; ++it)
            if (pred(*it, *result))
                result = it;
 
        return result;
    }
 
    template <typename I> void reverse(I begin, I end)
    {
        while (end - begin > 1) swap(*begin++, *--end);
    }
 
    template <typename I> I unique(I begin, I end)
    {
        // fast skip head
        while (end - begin > 1 && *begin != *(begin + 1)) begin++;
 
        if (begin == end) return begin;
 
        // last written element
        I write = begin++;
 
        // merge unique elements
        while (begin != end)
        {
            if (*begin != *write)
                *++write = *begin++;
            else
                begin++;
        }
 
        // past-the-end (write points to live element)
        return write + 1;
    }
 
    template <typename T, typename Pred> void insertion_sort(T* begin, T* end, const Pred& pred)
    {
        if (begin == end)
            return;
 
        for (T* it = begin + 1; it != end; ++it)
        {
            T val = *it;
            T* hole = it;
 
            // move hole backwards
            while (hole > begin && pred(val, *(hole - 1)))
            {
                *hole = *(hole - 1);
                hole--;
            }
 
            // fill hole with element
            *hole = val;
        }
    }
 
    template <typename I, typename Pred> I median3(I first, I middle, I last, const Pred& pred)
    {
        if (pred(*middle, *first)) swap(middle, first);
        if (pred(*last, *middle)) swap(last, middle);
        if (pred(*middle, *first)) swap(middle, first);
 
        return middle;
    }
 
    template <typename T, typename Pred> void partition3(T* begin, T* end, T pivot, const Pred& pred, T** out_eqbeg, T** out_eqend)
    {
        // invariant: array is split into 4 groups: = < ? > (each variable denotes the boundary between the groups)
        T* eq = begin;
        T* lt = begin;
        T* gt = end;
 
        while (lt < gt)
        {
            if (pred(*lt, pivot))
                lt++;
            else if (*lt == pivot)
                swap(*eq++, *lt++);
            else
                swap(*lt, *--gt);
        }
 
        // we now have just 4 groups: = < >; move equal elements to the middle
        T* eqbeg = gt;
 
        for (T* it = begin; it != eq; ++it)
            swap(*it, *--eqbeg);
 
        *out_eqbeg = eqbeg;
        *out_eqend = gt;
    }
 
    template <typename I, typename Pred> void sort(I begin, I end, const Pred& pred)
    {
        // sort large chunks
        while (end - begin > 16)
        {
            // find median element
            I middle = begin + (end - begin) / 2;
            I median = median3(begin, middle, end - 1, pred);
 
            // partition in three chunks (< = >)
            I eqbeg, eqend;
            partition3(begin, end, *median, pred, &eqbeg, &eqend);
 
            // loop on larger half
            if (eqbeg - begin > end - eqend)
            {
                sort(eqend, end, pred);
                end = eqbeg;
            }
            else
            {
                sort(begin, eqbeg, pred);
                begin = eqend;
            }
        }
 
        // insertion sort small chunk
        insertion_sort(begin, end, pred);
    }
PUGI__NS_END
 
// Allocator used for AST and evaluation stacks
PUGI__NS_BEGIN
    static const size_t xpath_memory_page_size =
    #ifdef PUGIXML_MEMORY_XPATH_PAGE_SIZE
        PUGIXML_MEMORY_XPATH_PAGE_SIZE
    #else
        4096
    #endif
        ;
 
    static const uintptr_t xpath_memory_block_alignment = sizeof(double) > sizeof(void*) ? sizeof(double) : sizeof(void*);
 
    struct xpath_memory_block
    {
        xpath_memory_block* next;
        size_t capacity;
 
        union
        {
            char data[xpath_memory_page_size];
            double alignment;
        };
    };
 
    struct xpath_allocator
    {
        xpath_memory_block* _root;
        size_t _root_size;
        bool* _error;
 
        xpath_allocator(xpath_memory_block* root, bool* error = 0): _root(root), _root_size(0), _error(error)
        {
        }
 
        void* allocate(size_t size)
        {
            // round size up to block alignment boundary
            size = (size + xpath_memory_block_alignment - 1) & ~(xpath_memory_block_alignment - 1);
 
            if (_root_size + size <= _root->capacity)
            {
                void* buf = &_root->data[0] + _root_size;
                _root_size += size;
                return buf;
            }
            else
            {
                // make sure we have at least 1/4th of the page free after allocation to satisfy subsequent allocation requests
                size_t block_capacity_base = sizeof(_root->data);
                size_t block_capacity_req = size + block_capacity_base / 4;
                size_t block_capacity = (block_capacity_base > block_capacity_req) ? block_capacity_base : block_capacity_req;
 
                size_t block_size = block_capacity + offsetof(xpath_memory_block, data);
 
                xpath_memory_block* block = static_cast<xpath_memory_block*>(xml_memory::allocate(block_size));
                if (!block)
                {
                    if (_error) *_error = true;
                    return 0;
                }
 
                block->next = _root;
                block->capacity = block_capacity;
 
                _root = block;
                _root_size = size;
 
                return block->data;
            }
        }
 
        void* reallocate(void* ptr, size_t old_size, size_t new_size)
        {
            // round size up to block alignment boundary
            old_size = (old_size + xpath_memory_block_alignment - 1) & ~(xpath_memory_block_alignment - 1);
            new_size = (new_size + xpath_memory_block_alignment - 1) & ~(xpath_memory_block_alignment - 1);
 
            // we can only reallocate the last object
            assert(ptr == 0 || static_cast<char*>(ptr) + old_size == &_root->data[0] + _root_size);
 
            // try to reallocate the object inplace
            if (ptr && _root_size - old_size + new_size <= _root->capacity)
            {
                _root_size = _root_size - old_size + new_size;
                return ptr;
            }
 
            // allocate a new block
            void* result = allocate(new_size);
            if (!result) return 0;
 
            // we have a new block
            if (ptr)
            {
                // copy old data (we only support growing)
                assert(new_size >= old_size);
                memcpy(result, ptr, old_size);
 
                // free the previous page if it had no other objects
                assert(_root->data == result);
                assert(_root->next);
 
                if (_root->next->data == ptr)
                {
                    // deallocate the whole page, unless it was the first one
                    xpath_memory_block* next = _root->next->next;
 
                    if (next)
                    {
                        xml_memory::deallocate(_root->next);
                        _root->next = next;
                    }
                }
            }
 
            return result;
        }
 
        void revert(const xpath_allocator& state)
        {
            // free all new pages
            xpath_memory_block* cur = _root;
 
            while (cur != state._root)
            {
                xpath_memory_block* next = cur->next;
 
                xml_memory::deallocate(cur);
 
                cur = next;
            }
 
            // restore state
            _root = state._root;
            _root_size = state._root_size;
        }
 
        void release()
        {
            xpath_memory_block* cur = _root;
            assert(cur);
 
            while (cur->next)
            {
                xpath_memory_block* next = cur->next;
 
                xml_memory::deallocate(cur);
 
                cur = next;
            }
        }
    };
 
    struct xpath_allocator_capture
    {
        xpath_allocator_capture(xpath_allocator* alloc): _target(alloc), _state(*alloc)
        {
        }
 
        ~xpath_allocator_capture()
        {
            _target->revert(_state);
        }
 
        xpath_allocator* _target;
        xpath_allocator _state;
    };
 
    struct xpath_stack
    {
        xpath_allocator* result;
        xpath_allocator* temp;
    };
 
    struct xpath_stack_data
    {
        xpath_memory_block blocks[2];
        xpath_allocator result;
        xpath_allocator temp;
        xpath_stack stack;
        bool oom;
 
        xpath_stack_data(): result(blocks + 0, &oom), temp(blocks + 1, &oom), oom(false)
        {
            blocks[0].next = blocks[1].next = 0;
            blocks[0].capacity = blocks[1].capacity = sizeof(blocks[0].data);
 
            stack.result = &result;
            stack.temp = &temp;
        }
 
        ~xpath_stack_data()
        {
            result.release();
            temp.release();
        }
    };
PUGI__NS_END
 
// String class
PUGI__NS_BEGIN
    class xpath_string
    {
        const char_t* _buffer;
        bool _uses_heap;
        size_t _length_heap;
 
        static char_t* duplicate_string(const char_t* string, size_t length, xpath_allocator* alloc)
        {
            char_t* result = static_cast<char_t*>(alloc->allocate((length + 1) * sizeof(char_t)));
            if (!result) return 0;
 
            memcpy(result, string, length * sizeof(char_t));
            result[length] = 0;
 
            return result;
        }
 
        xpath_string(const char_t* buffer, bool uses_heap_, size_t length_heap): _buffer(buffer), _uses_heap(uses_heap_), _length_heap(length_heap)
        {
        }
 
    public:
        static xpath_string from_const(const char_t* str)
        {
            return xpath_string(str, false, 0);
        }
 
        static xpath_string from_heap_preallocated(const char_t* begin, const char_t* end)
        {
            assert(begin <= end && *end == 0);
 
            return xpath_string(begin, true, static_cast<size_t>(end - begin));
        }
 
        static xpath_string from_heap(const char_t* begin, const char_t* end, xpath_allocator* alloc)
        {
            assert(begin <= end);
 
            if (begin == end)
                return xpath_string();
 
            size_t length = static_cast<size_t>(end - begin);
            const char_t* data = duplicate_string(begin, length, alloc);
 
            return data ? xpath_string(data, true, length) : xpath_string();
        }
 
        xpath_string(): _buffer(PUGIXML_TEXT("")), _uses_heap(false), _length_heap(0)
        {
        }
 
        void append(const xpath_string& o, xpath_allocator* alloc)
        {
            // skip empty sources
            if (!*o._buffer) return;
 
            // fast append for constant empty target and constant source
            if (!*_buffer && !_uses_heap && !o._uses_heap)
            {
                _buffer = o._buffer;
            }
            else
            {
                // need to make heap copy
                size_t target_length = length();
                size_t source_length = o.length();
                size_t result_length = target_length + source_length;
 
                // allocate new buffer
                char_t* result = static_cast<char_t*>(alloc->reallocate(_uses_heap ? const_cast<char_t*>(_buffer) : 0, (target_length + 1) * sizeof(char_t), (result_length + 1) * sizeof(char_t)));
                if (!result) return;
 
                // append first string to the new buffer in case there was no reallocation
                if (!_uses_heap) memcpy(result, _buffer, target_length * sizeof(char_t));
 
                // append second string to the new buffer
                memcpy(result + target_length, o._buffer, source_length * sizeof(char_t));
                result[result_length] = 0;
 
                // finalize
                _buffer = result;
                _uses_heap = true;
                _length_heap = result_length;
            }
        }
 
        const char_t* c_str() const
        {
            return _buffer;
        }
 
        size_t length() const
        {
            return _uses_heap ? _length_heap : strlength(_buffer);
        }
 
        char_t* data(xpath_allocator* alloc)
        {
            // make private heap copy
            if (!_uses_heap)
            {
                size_t length_ = strlength(_buffer);
                const char_t* data_ = duplicate_string(_buffer, length_, alloc);
 
                if (!data_) return 0;
 
                _buffer = data_;
                _uses_heap = true;
                _length_heap = length_;
            }
 
            return const_cast<char_t*>(_buffer);
        }
 
        bool empty() const
        {
            return *_buffer == 0;
        }
 
        bool operator==(const xpath_string& o) const
        {
            return strequal(_buffer, o._buffer);
        }
 
        bool operator!=(const xpath_string& o) const
        {
            return !strequal(_buffer, o._buffer);
        }
 
        bool uses_heap() const
        {
            return _uses_heap;
        }
    };
PUGI__NS_END
 
PUGI__NS_BEGIN
    PUGI__FN bool starts_with(const char_t* string, const char_t* pattern)
    {
        while (*pattern && *string == *pattern)
        {
            string++;
            pattern++;
        }
 
        return *pattern == 0;
    }
 
    PUGI__FN const char_t* find_char(const char_t* s, char_t c)
    {
    #ifdef PUGIXML_WCHAR_MODE
        return wcschr(s, c);
    #else
        return strchr(s, c);
    #endif
    }
 
    PUGI__FN const char_t* find_substring(const char_t* s, const char_t* p)
    {
    #ifdef PUGIXML_WCHAR_MODE
        // MSVC6 wcsstr bug workaround (if s is empty it always returns 0)
        return (*p == 0) ? s : wcsstr(s, p);
    #else
        return strstr(s, p);
    #endif
    }
 
    // Converts symbol to lower case, if it is an ASCII one
    PUGI__FN char_t tolower_ascii(char_t ch)
    {
        return static_cast<unsigned int>(ch - 'A') < 26 ? static_cast<char_t>(ch | ' ') : ch;
    }
 
    PUGI__FN xpath_string string_value(const xpath_node& na, xpath_allocator* alloc)
    {
        if (na.attribute())
            return xpath_string::from_const(na.attribute().value());
        else
        {
            xml_node n = na.node();
 
            switch (n.type())
            {
            case node_pcdata:
            case node_cdata:
            case node_comment:
            case node_pi:
                return xpath_string::from_const(n.value());
 
            case node_document:
            case node_element:
            {
                xpath_string result;
 
                // element nodes can have value if parse_embed_pcdata was used
                if (n.value()[0])
                    result.append(xpath_string::from_const(n.value()), alloc);
 
                xml_node cur = n.first_child();
 
                while (cur && cur != n)
                {
                    if (cur.type() == node_pcdata || cur.type() == node_cdata)
                        result.append(xpath_string::from_const(cur.value()), alloc);
 
                    if (cur.first_child())
                        cur = cur.first_child();
                    else if (cur.next_sibling())
                        cur = cur.next_sibling();
                    else
                    {
                        while (!cur.next_sibling() && cur != n)
                            cur = cur.parent();
 
                        if (cur != n) cur = cur.next_sibling();
                    }
                }
 
                return result;
            }
 
            default:
                return xpath_string();
            }
        }
    }
 
    PUGI__FN bool node_is_before_sibling(xml_node_struct* ln, xml_node_struct* rn)
    {
        assert(ln->parent == rn->parent);
 
        // there is no common ancestor (the shared parent is null), nodes are from different documents
        if (!ln->parent) return ln < rn;
 
        // determine sibling order
        xml_node_struct* ls = ln;
        xml_node_struct* rs = rn;
 
        while (ls && rs)
        {
            if (ls == rn) return true;
            if (rs == ln) return false;
 
            ls = ls->next_sibling;
            rs = rs->next_sibling;
        }
 
        // if rn sibling chain ended ln must be before rn
        return !rs;
    }
 
    PUGI__FN bool node_is_before(xml_node_struct* ln, xml_node_struct* rn)
    {
        // find common ancestor at the same depth, if any
        xml_node_struct* lp = ln;
        xml_node_struct* rp = rn;
 
        while (lp && rp && lp->parent != rp->parent)
        {
            lp = lp->parent;
            rp = rp->parent;
        }
 
        // parents are the same!
        if (lp && rp) return node_is_before_sibling(lp, rp);
 
        // nodes are at different depths, need to normalize heights
        bool left_higher = !lp;
 
        while (lp)
        {
            lp = lp->parent;
            ln = ln->parent;
        }
 
        while (rp)
        {
            rp = rp->parent;
            rn = rn->parent;
        }
 
        // one node is the ancestor of the other
        if (ln == rn) return left_higher;
 
        // find common ancestor... again
        while (ln->parent != rn->parent)
        {
            ln = ln->parent;
            rn = rn->parent;
        }
 
        return node_is_before_sibling(ln, rn);
    }
 
    PUGI__FN bool node_is_ancestor(xml_node_struct* parent, xml_node_struct* node)
    {
        while (node && node != parent) node = node->parent;
 
        return parent && node == parent;
    }
 
    PUGI__FN const void* document_buffer_order(const xpath_node& xnode)
    {
        xml_node_struct* node = xnode.node().internal_object();
 
        if (node)
        {
            if ((get_document(node).header & xml_memory_page_contents_shared_mask) == 0)
            {
                if (node->name && (node->header & impl::xml_memory_page_name_allocated_or_shared_mask) == 0) return node->name;
                if (node->value && (node->header & impl::xml_memory_page_value_allocated_or_shared_mask) == 0) return node->value;
            }
 
            return 0;
        }
 
        xml_attribute_struct* attr = xnode.attribute().internal_object();
 
        if (attr)
        {
            if ((get_document(attr).header & xml_memory_page_contents_shared_mask) == 0)
            {
                if ((attr->header & impl::xml_memory_page_name_allocated_or_shared_mask) == 0) return attr->name;
                if ((attr->header & impl::xml_memory_page_value_allocated_or_shared_mask) == 0) return attr->value;
            }
 
            return 0;
        }
 
        return 0;
    }
 
    struct document_order_comparator
    {
        bool operator()(const xpath_node& lhs, const xpath_node& rhs) const
        {
            // optimized document order based check
            const void* lo = document_buffer_order(lhs);
            const void* ro = document_buffer_order(rhs);
 
            if (lo && ro) return lo < ro;
 
            // slow comparison
            xml_node ln = lhs.node(), rn = rhs.node();
 
            // compare attributes
            if (lhs.attribute() && rhs.attribute())
            {
                // shared parent
                if (lhs.parent() == rhs.parent())
                {
                    // determine sibling order
                    for (xml_attribute a = lhs.attribute(); a; a = a.next_attribute())
                        if (a == rhs.attribute())
                            return true;
 
                    return false;
                }
 
                // compare attribute parents
                ln = lhs.parent();
                rn = rhs.parent();
            }
            else if (lhs.attribute())
            {
                // attributes go after the parent element
                if (lhs.parent() == rhs.node()) return false;
 
                ln = lhs.parent();
            }
            else if (rhs.attribute())
            {
                // attributes go after the parent element
                if (rhs.parent() == lhs.node()) return true;
 
                rn = rhs.parent();
            }
 
            if (ln == rn) return false;
 
            if (!ln || !rn) return ln < rn;
 
            return node_is_before(ln.internal_object(), rn.internal_object());
        }
    };
 
    struct duplicate_comparator
    {
        bool operator()(const xpath_node& lhs, const xpath_node& rhs) const
        {
            if (lhs.attribute()) return rhs.attribute() ? lhs.attribute() < rhs.attribute() : true;
            else return rhs.attribute() ? false : lhs.node() < rhs.node();
        }
    };
 
    PUGI__FN double gen_nan()
    {
    #if defined(__STDC_IEC_559__) || ((FLT_RADIX - 0 == 2) && (FLT_MAX_EXP - 0 == 128) && (FLT_MANT_DIG - 0 == 24))
        PUGI__STATIC_ASSERT(sizeof(float) == sizeof(uint32_t));
        typedef uint32_t UI; // BCC5 workaround
        union { float f; UI i; } u;
        u.i = 0x7fc00000;
        return u.f;
    #else
        // fallback
        const volatile double zero = 0.0;
        return zero / zero;
    #endif
    }
 
    PUGI__FN bool is_nan(double value)
    {
    #if defined(PUGI__MSVC_CRT_VERSION) || defined(__BORLANDC__)
        return !!_isnan(value);
    #elif defined(fpclassify) && defined(FP_NAN)
        return fpclassify(value) == FP_NAN;
    #else
        // fallback
        const volatile double v = value;
        return v != v;
    #endif
    }
 
    PUGI__FN const char_t* convert_number_to_string_special(double value)
    {
    #if defined(PUGI__MSVC_CRT_VERSION) || defined(__BORLANDC__)
        if (_finite(value)) return (value == 0) ? PUGIXML_TEXT("0") : 0;
        if (_isnan(value)) return PUGIXML_TEXT("NaN");
        return value > 0 ? PUGIXML_TEXT("Infinity") : PUGIXML_TEXT("-Infinity");
    #elif defined(fpclassify) && defined(FP_NAN) && defined(FP_INFINITE) && defined(FP_ZERO)
        switch (fpclassify(value))
        {
        case FP_NAN:
            return PUGIXML_TEXT("NaN");
 
        case FP_INFINITE:
            return value > 0 ? PUGIXML_TEXT("Infinity") : PUGIXML_TEXT("-Infinity");
 
        case FP_ZERO:
            return PUGIXML_TEXT("0");
 
        default:
            return 0;
        }
    #else
        // fallback
        const volatile double v = value;
 
        if (v == 0) return PUGIXML_TEXT("0");
        if (v != v) return PUGIXML_TEXT("NaN");
        if (v * 2 == v) return value > 0 ? PUGIXML_TEXT("Infinity") : PUGIXML_TEXT("-Infinity");
        return 0;
    #endif
    }
 
    PUGI__FN bool convert_number_to_boolean(double value)
    {
        return (value != 0 && !is_nan(value));
    }
 
    PUGI__FN void truncate_zeros(char* begin, char* end)
    {
        while (begin != end && end[-1] == '0') end--;
 
        *end = 0;
    }
 
    // gets mantissa digits in the form of 0.xxxxx with 0. implied and the exponent
#if defined(PUGI__MSVC_CRT_VERSION) && PUGI__MSVC_CRT_VERSION >= 1400 && !defined(_WIN32_WCE)
    PUGI__FN void convert_number_to_mantissa_exponent(double value, char (&buffer)[32], char** out_mantissa, int* out_exponent)
    {
        // get base values
        int sign, exponent;
        _ecvt_s(buffer, sizeof(buffer), value, DBL_DIG + 1, &exponent, &sign);
 
        // truncate redundant zeros
        truncate_zeros(buffer, buffer + strlen(buffer));
 
        // fill results
        *out_mantissa = buffer;
        *out_exponent = exponent;
    }
#else
    PUGI__FN void convert_number_to_mantissa_exponent(double value, char (&buffer)[32], char** out_mantissa, int* out_exponent)
    {
        // get a scientific notation value with IEEE DBL_DIG decimals
        PUGI__SNPRINTF(buffer, "%.*e", DBL_DIG, value);
 
        // get the exponent (possibly negative)
        char* exponent_string = strchr(buffer, 'e');
        assert(exponent_string);
 
        int exponent = atoi(exponent_string + 1);
 
        // extract mantissa string: skip sign
        char* mantissa = buffer[0] == '-' ? buffer + 1 : buffer;
        assert(mantissa[0] != '0' && mantissa[1] == '.');
 
        // divide mantissa by 10 to eliminate integer part
        mantissa[1] = mantissa[0];
        mantissa++;
        exponent++;
 
        // remove extra mantissa digits and zero-terminate mantissa
        truncate_zeros(mantissa, exponent_string);
 
        // fill results
        *out_mantissa = mantissa;
        *out_exponent = exponent;
    }
#endif
 
    PUGI__FN xpath_string convert_number_to_string(double value, xpath_allocator* alloc)
    {
        // try special number conversion
        const char_t* special = convert_number_to_string_special(value);
        if (special) return xpath_string::from_const(special);
 
        // get mantissa + exponent form
        char mantissa_buffer[32];
 
        char* mantissa;
        int exponent;
        convert_number_to_mantissa_exponent(value, mantissa_buffer, &mantissa, &exponent);
 
        // allocate a buffer of suitable length for the number
        size_t result_size = strlen(mantissa_buffer) + (exponent > 0 ? exponent : -exponent) + 4;
        char_t* result = static_cast<char_t*>(alloc->allocate(sizeof(char_t) * result_size));
        if (!result) return xpath_string();
 
        // make the number!
        char_t* s = result;
 
        // sign
        if (value < 0) *s++ = '-';
 
        // integer part
        if (exponent <= 0)
        {
            *s++ = '0';
        }
        else
        {
            while (exponent > 0)
            {
                assert(*mantissa == 0 || static_cast<unsigned int>(*mantissa - '0') <= 9);
                *s++ = *mantissa ? *mantissa++ : '0';
                exponent--;
            }
        }
 
        // fractional part
        if (*mantissa)
        {
            // decimal point
            *s++ = '.';
 
            // extra zeroes from negative exponent
            while (exponent < 0)
            {
                *s++ = '0';
                exponent++;
            }
 
            // extra mantissa digits
            while (*mantissa)
            {
                assert(static_cast<unsigned int>(*mantissa - '0') <= 9);
                *s++ = *mantissa++;
            }
        }
 
        // zero-terminate
        assert(s < result + result_size);
        *s = 0;
 
        return xpath_string::from_heap_preallocated(result, s);
    }
 
    PUGI__FN bool check_string_to_number_format(const char_t* string)
    {
        // parse leading whitespace
        while (PUGI__IS_CHARTYPE(*string, ct_space)) ++string;
 
        // parse sign
        if (*string == '-') ++string;
 
        if (!*string) return false;
 
        // if there is no integer part, there should be a decimal part with at least one digit
        if (!PUGI__IS_CHARTYPEX(string[0], ctx_digit) && (string[0] != '.' || !PUGI__IS_CHARTYPEX(string[1], ctx_digit))) return false;
 
        // parse integer part
        while (PUGI__IS_CHARTYPEX(*string, ctx_digit)) ++string;
 
        // parse decimal part
        if (*string == '.')
        {
            ++string;
 
            while (PUGI__IS_CHARTYPEX(*string, ctx_digit)) ++string;
        }
 
        // parse trailing whitespace
        while (PUGI__IS_CHARTYPE(*string, ct_space)) ++string;
 
        return *string == 0;
    }
 
    PUGI__FN double convert_string_to_number(const char_t* string)
    {
        // check string format
        if (!check_string_to_number_format(string)) return gen_nan();
 
        // parse string
    #ifdef PUGIXML_WCHAR_MODE
        return wcstod(string, 0);
    #else
        return strtod(string, 0);
    #endif
    }
 
    PUGI__FN bool convert_string_to_number_scratch(char_t (&buffer)[32], const char_t* begin, const char_t* end, double* out_result)
    {
        size_t length = static_cast<size_t>(end - begin);
        char_t* scratch = buffer;
 
        if (length >= sizeof(buffer) / sizeof(buffer[0]))
        {
            // need to make dummy on-heap copy
            scratch = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
            if (!scratch) return false;
        }
 
        // copy string to zero-terminated buffer and perform conversion
        memcpy(scratch, begin, length * sizeof(char_t));
        scratch[length] = 0;
 
        *out_result = convert_string_to_number(scratch);
 
        // free dummy buffer
        if (scratch != buffer) xml_memory::deallocate(scratch);
 
        return true;
    }
 
    PUGI__FN double round_nearest(double value)
    {
        return floor(value + 0.5);
    }
 
    PUGI__FN double round_nearest_nzero(double value)
    {
        // same as round_nearest, but returns -0 for [-0.5, -0]
        // ceil is used to differentiate between +0 and -0 (we return -0 for [-0.5, -0] and +0 for +0)
        return (value >= -0.5 && value <= 0) ? ceil(value) : floor(value + 0.5);
    }
 
    PUGI__FN const char_t* qualified_name(const xpath_node& node)
    {
        return node.attribute() ? node.attribute().name() : node.node().name();
    }
 
    PUGI__FN const char_t* local_name(const xpath_node& node)
    {
        const char_t* name = qualified_name(node);
        const char_t* p = find_char(name, ':');
 
        return p ? p + 1 : name;
    }
 
    struct namespace_uri_predicate
    {
        const char_t* prefix;
        size_t prefix_length;
 
        namespace_uri_predicate(const char_t* name)
        {
            const char_t* pos = find_char(name, ':');
 
            prefix = pos ? name : 0;
            prefix_length = pos ? static_cast<size_t>(pos - name) : 0;
        }
 
        bool operator()(xml_attribute a) const
        {
            const char_t* name = a.name();
 
            if (!starts_with(name, PUGIXML_TEXT("xmlns"))) return false;
 
            return prefix ? name[5] == ':' && strequalrange(name + 6, prefix, prefix_length) : name[5] == 0;
        }
    };
 
    PUGI__FN const char_t* namespace_uri(xml_node node)
    {
        namespace_uri_predicate pred = node.name();
 
        xml_node p = node;
 
        while (p)
        {
            xml_attribute a = p.find_attribute(pred);
 
            if (a) return a.value();
 
            p = p.parent();
        }
 
        return PUGIXML_TEXT("");
    }
 
    PUGI__FN const char_t* namespace_uri(xml_attribute attr, xml_node parent)
    {
        namespace_uri_predicate pred = attr.name();
 
        // Default namespace does not apply to attributes
        if (!pred.prefix) return PUGIXML_TEXT("");
 
        xml_node p = parent;
 
        while (p)
        {
            xml_attribute a = p.find_attribute(pred);
 
            if (a) return a.value();
 
            p = p.parent();
        }
 
        return PUGIXML_TEXT("");
    }
 
    PUGI__FN const char_t* namespace_uri(const xpath_node& node)
    {
        return node.attribute() ? namespace_uri(node.attribute(), node.parent()) : namespace_uri(node.node());
    }
 
    PUGI__FN char_t* normalize_space(char_t* buffer)
    {
        char_t* write = buffer;
 
        for (char_t* it = buffer; *it; )
        {
            char_t ch = *it++;
 
            if (PUGI__IS_CHARTYPE(ch, ct_space))
            {
                // replace whitespace sequence with single space
                while (PUGI__IS_CHARTYPE(*it, ct_space)) it++;
 
                // avoid leading spaces
                if (write != buffer) *write++ = ' ';
            }
            else *write++ = ch;
        }
 
        // remove trailing space
        if (write != buffer && PUGI__IS_CHARTYPE(write[-1], ct_space)) write--;
 
        // zero-terminate
        *write = 0;
 
        return write;
    }
 
    PUGI__FN char_t* translate(char_t* buffer, const char_t* from, const char_t* to, size_t to_length)
    {
        char_t* write = buffer;
 
        while (*buffer)
        {
            PUGI__DMC_VOLATILE char_t ch = *buffer++;
 
            const char_t* pos = find_char(from, ch);
 
            if (!pos)
                *write++ = ch; // do not process
            else if (static_cast<size_t>(pos - from) < to_length)
                *write++ = to[pos - from]; // replace
        }
 
        // zero-terminate
        *write = 0;
 
        return write;
    }
 
    PUGI__FN unsigned char* translate_table_generate(xpath_allocator* alloc, const char_t* from, const char_t* to)
    {
        unsigned char table[128] = {0};
 
        while (*from)
        {
            unsigned int fc = static_cast<unsigned int>(*from);
            unsigned int tc = static_cast<unsigned int>(*to);
 
            if (fc >= 128 || tc >= 128)
                return 0;
 
            // code=128 means "skip character"
            if (!table[fc])
                table[fc] = static_cast<unsigned char>(tc ? tc : 128);
 
            from++;
            if (tc) to++;
        }
 
        for (int i = 0; i < 128; ++i)
            if (!table[i])
                table[i] = static_cast<unsigned char>(i);
 
        void* result = alloc->allocate(sizeof(table));
        if (!result) return 0;
 
        memcpy(result, table, sizeof(table));
 
        return static_cast<unsigned char*>(result);
    }
 
    PUGI__FN char_t* translate_table(char_t* buffer, const unsigned char* table)
    {
        char_t* write = buffer;
 
        while (*buffer)
        {
            char_t ch = *buffer++;
            unsigned int index = static_cast<unsigned int>(ch);
 
            if (index < 128)
            {
                unsigned char code = table[index];
 
                // code=128 means "skip character" (table size is 128 so 128 can be a special value)
                // this code skips these characters without extra branches
                *write = static_cast<char_t>(code);
                write += 1 - (code >> 7);
            }
            else
            {
                *write++ = ch;
            }
        }
 
        // zero-terminate
        *write = 0;
 
        return write;
    }
 
    inline bool is_xpath_attribute(const char_t* name)
    {
        return !(starts_with(name, PUGIXML_TEXT("xmlns")) && (name[5] == 0 || name[5] == ':'));
    }
 
    struct xpath_variable_boolean: xpath_variable
    {
        xpath_variable_boolean(): xpath_variable(xpath_type_boolean), value(false)
        {
        }
 
        bool value;
        char_t name[1];
    };
 
    struct xpath_variable_number: xpath_variable
    {
        xpath_variable_number(): xpath_variable(xpath_type_number), value(0)
        {
        }
 
        double value;
        char_t name[1];
    };
 
    struct xpath_variable_string: xpath_variable
    {
        xpath_variable_string(): xpath_variable(xpath_type_string), value(0)
        {
        }
 
        ~xpath_variable_string()
        {
            if (value) xml_memory::deallocate(value);
        }
 
        char_t* value;
        char_t name[1];
    };
 
    struct xpath_variable_node_set: xpath_variable
    {
        xpath_variable_node_set(): xpath_variable(xpath_type_node_set)
        {
        }
 
        xpath_node_set value;
        char_t name[1];
    };
 
    static const xpath_node_set dummy_node_set;
 
    PUGI__FN PUGI__UNSIGNED_OVERFLOW unsigned int hash_string(const char_t* str)
    {
        // Jenkins one-at-a-time hash (http://en.wikipedia.org/wiki/Jenkins_hash_function#one-at-a-time)
        unsigned int result = 0;
 
        while (*str)
        {
            result += static_cast<unsigned int>(*str++);
            result += result << 10;
            result ^= result >> 6;
        }
 
        result += result << 3;
        result ^= result >> 11;
        result += result << 15;
 
        return result;
    }
 
    template <typename T> PUGI__FN T* new_xpath_variable(const char_t* name)
    {
        size_t length = strlength(name);
        if (length == 0) return 0; // empty variable names are invalid
 
        // $$ we can't use offsetof(T, name) because T is non-POD, so we just allocate additional length characters
        void* memory = xml_memory::allocate(sizeof(T) + length * sizeof(char_t));
        if (!memory) return 0;
 
        T* result = new (memory) T();
 
        memcpy(result->name, name, (length + 1) * sizeof(char_t));
 
        return result;
    }
 
    PUGI__FN xpath_variable* new_xpath_variable(xpath_value_type type, const char_t* name)
    {
        switch (type)
        {
        case xpath_type_node_set:
            return new_xpath_variable<xpath_variable_node_set>(name);
 
        case xpath_type_number:
            return new_xpath_variable<xpath_variable_number>(name);
 
        case xpath_type_string:
            return new_xpath_variable<xpath_variable_string>(name);
 
        case xpath_type_boolean:
            return new_xpath_variable<xpath_variable_boolean>(name);
 
        default:
            return 0;
        }
    }
 
    template <typename T> PUGI__FN void delete_xpath_variable(T* var)
    {
        var->~T();
        xml_memory::deallocate(var);
    }
 
    PUGI__FN void delete_xpath_variable(xpath_value_type type, xpath_variable* var)
    {
        switch (type)
        {
        case xpath_type_node_set:
            delete_xpath_variable(static_cast<xpath_variable_node_set*>(var));
            break;
 
        case xpath_type_number:
            delete_xpath_variable(static_cast<xpath_variable_number*>(var));
            break;
 
        case xpath_type_string:
            delete_xpath_variable(static_cast<xpath_variable_string*>(var));
            break;
 
        case xpath_type_boolean:
            delete_xpath_variable(static_cast<xpath_variable_boolean*>(var));
            break;
 
        default:
            assert(false && "Invalid variable type"); // unreachable
        }
    }
 
    PUGI__FN bool copy_xpath_variable(xpath_variable* lhs, const xpath_variable* rhs)
    {
        switch (rhs->type())
        {
        case xpath_type_node_set:
            return lhs->set(static_cast<const xpath_variable_node_set*>(rhs)->value);
 
        case xpath_type_number:
            return lhs->set(static_cast<const xpath_variable_number*>(rhs)->value);
 
        case xpath_type_string:
            return lhs->set(static_cast<const xpath_variable_string*>(rhs)->value);
 
        case xpath_type_boolean:
            return lhs->set(static_cast<const xpath_variable_boolean*>(rhs)->value);
 
        default:
            assert(false && "Invalid variable type"); // unreachable
            return false;
        }
    }
 
    PUGI__FN bool get_variable_scratch(char_t (&buffer)[32], xpath_variable_set* set, const char_t* begin, const char_t* end, xpath_variable** out_result)
    {
        size_t length = static_cast<size_t>(end - begin);
        char_t* scratch = buffer;
 
        if (length >= sizeof(buffer) / sizeof(buffer[0]))
        {
            // need to make dummy on-heap copy
            scratch = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
            if (!scratch) return false;
        }
 
        // copy string to zero-terminated buffer and perform lookup
        memcpy(scratch, begin, length * sizeof(char_t));
        scratch[length] = 0;
 
        *out_result = set->get(scratch);
 
        // free dummy buffer
        if (scratch != buffer) xml_memory::deallocate(scratch);
 
        return true;
    }
PUGI__NS_END
 
// Internal node set class
PUGI__NS_BEGIN
    PUGI__FN xpath_node_set::type_t xpath_get_order(const xpath_node* begin, const xpath_node* end)
    {
        if (end - begin < 2)
            return xpath_node_set::type_sorted;
 
        document_order_comparator cmp;
 
        bool first = cmp(begin[0], begin[1]);
 
        for (const xpath_node* it = begin + 1; it + 1 < end; ++it)
            if (cmp(it[0], it[1]) != first)
                return xpath_node_set::type_unsorted;
 
        return first ? xpath_node_set::type_sorted : xpath_node_set::type_sorted_reverse;
    }
 
    PUGI__FN xpath_node_set::type_t xpath_sort(xpath_node* begin, xpath_node* end, xpath_node_set::type_t type, bool rev)
    {
        xpath_node_set::type_t order = rev ? xpath_node_set::type_sorted_reverse : xpath_node_set::type_sorted;
 
        if (type == xpath_node_set::type_unsorted)
        {
            xpath_node_set::type_t sorted = xpath_get_order(begin, end);
 
            if (sorted == xpath_node_set::type_unsorted)
            {
                sort(begin, end, document_order_comparator());
 
                type = xpath_node_set::type_sorted;
            }
            else
                type = sorted;
        }
 
        if (type != order) reverse(begin, end);
 
        return order;
    }
 
    PUGI__FN xpath_node xpath_first(const xpath_node* begin, const xpath_node* end, xpath_node_set::type_t type)
    {
        if (begin == end) return xpath_node();
 
        switch (type)
        {
        case xpath_node_set::type_sorted:
            return *begin;
 
        case xpath_node_set::type_sorted_reverse:
            return *(end - 1);
 
        case xpath_node_set::type_unsorted:
            return *min_element(begin, end, document_order_comparator());
 
        default:
            assert(false && "Invalid node set type"); // unreachable
            return xpath_node();
        }
    }
 
    class xpath_node_set_raw
    {
        xpath_node_set::type_t _type;
 
        xpath_node* _begin;
        xpath_node* _end;
        xpath_node* _eos;
 
    public:
        xpath_node_set_raw(): _type(xpath_node_set::type_unsorted), _begin(0), _end(0), _eos(0)
        {
        }
 
        xpath_node* begin() const
        {
            return _begin;
        }
 
        xpath_node* end() const
        {
            return _end;
        }
 
        bool empty() const
        {
            return _begin == _end;
        }
 
        size_t size() const
        {
            return static_cast<size_t>(_end - _begin);
        }
 
        xpath_node first() const
        {
            return xpath_first(_begin, _end, _type);
        }
 
        void push_back_grow(const xpath_node& node, xpath_allocator* alloc);
 
        void push_back(const xpath_node& node, xpath_allocator* alloc)
        {
            if (_end != _eos)
                *_end++ = node;
            else
                push_back_grow(node, alloc);
        }
 
        void append(const xpath_node* begin_, const xpath_node* end_, xpath_allocator* alloc)
        {
            if (begin_ == end_) return;
 
            size_t size_ = static_cast<size_t>(_end - _begin);
            size_t capacity = static_cast<size_t>(_eos - _begin);
            size_t count = static_cast<size_t>(end_ - begin_);
 
            if (size_ + count > capacity)
            {
                // reallocate the old array or allocate a new one
                xpath_node* data = static_cast<xpath_node*>(alloc->reallocate(_begin, capacity * sizeof(xpath_node), (size_ + count) * sizeof(xpath_node)));
                if (!data) return;
 
                // finalize
                _begin = data;
                _end = data + size_;
                _eos = data + size_ + count;
            }
 
            memcpy(_end, begin_, count * sizeof(xpath_node));
            _end += count;
        }
 
        void sort_do()
        {
            _type = xpath_sort(_begin, _end, _type, false);
        }
 
        void truncate(xpath_node* pos)
        {
            assert(_begin <= pos && pos <= _end);
 
            _end = pos;
        }
 
        void remove_duplicates()
        {
            if (_type == xpath_node_set::type_unsorted)
                sort(_begin, _end, duplicate_comparator());
 
            _end = unique(_begin, _end);
        }
 
        xpath_node_set::type_t type() const
        {
            return _type;
        }
 
        void set_type(xpath_node_set::type_t value)
        {
            _type = value;
        }
    };
 
    PUGI__FN_NO_INLINE void xpath_node_set_raw::push_back_grow(const xpath_node& node, xpath_allocator* alloc)
    {
        size_t capacity = static_cast<size_t>(_eos - _begin);
 
        // get new capacity (1.5x rule)
        size_t new_capacity = capacity + capacity / 2 + 1;
 
        // reallocate the old array or allocate a new one
        xpath_node* data = static_cast<xpath_node*>(alloc->reallocate(_begin, capacity * sizeof(xpath_node), new_capacity * sizeof(xpath_node)));
        if (!data) return;
 
        // finalize
        _begin = data;
        _end = data + capacity;
        _eos = data + new_capacity;
 
        // push
        *_end++ = node;
    }
PUGI__NS_END
 
PUGI__NS_BEGIN
    struct xpath_context
    {
        xpath_node n;
        size_t position, size;
 
        xpath_context(const xpath_node& n_, size_t position_, size_t size_): n(n_), position(position_), size(size_)
        {
        }
    };
 
    enum lexeme_t
    {
        lex_none = 0,
        lex_equal,
        lex_not_equal,
        lex_less,
        lex_greater,
        lex_less_or_equal,
        lex_greater_or_equal,
        lex_plus,
        lex_minus,
        lex_multiply,
        lex_union,
        lex_var_ref,
        lex_open_brace,
        lex_close_brace,
        lex_quoted_string,
        lex_number,
        lex_slash,
        lex_double_slash,
        lex_open_square_brace,
        lex_close_square_brace,
        lex_string,
        lex_comma,
        lex_axis_attribute,
        lex_dot,
        lex_double_dot,
        lex_double_colon,
        lex_eof
    };
 
    struct xpath_lexer_string
    {
        const char_t* begin;
        const char_t* end;
 
        xpath_lexer_string(): begin(0), end(0)
        {
        }
 
        bool operator==(const char_t* other) const
        {
            size_t length = static_cast<size_t>(end - begin);
 
            return strequalrange(other, begin, length);
        }
    };
 
    class xpath_lexer
    {
        const char_t* _cur;
        const char_t* _cur_lexeme_pos;
        xpath_lexer_string _cur_lexeme_contents;
 
        lexeme_t _cur_lexeme;
 
    public:
        explicit xpath_lexer(const char_t* query): _cur(query)
        {
            next();
        }
 
        const char_t* state() const
        {
            return _cur;
        }
 
        void next()
        {
            const char_t* cur = _cur;
 
            while (PUGI__IS_CHARTYPE(*cur, ct_space)) ++cur;
 
            // save lexeme position for error reporting
            _cur_lexeme_pos = cur;
 
            switch (*cur)
            {
            case 0:
                _cur_lexeme = lex_eof;
                break;
 
            case '>':
                if (*(cur+1) == '=')
                {
                    cur += 2;
                    _cur_lexeme = lex_greater_or_equal;
                }
                else
                {
                    cur += 1;
                    _cur_lexeme = lex_greater;
                }
                break;
 
            case '<':
                if (*(cur+1) == '=')
                {
                    cur += 2;
                    _cur_lexeme = lex_less_or_equal;
                }
                else
                {
                    cur += 1;
                    _cur_lexeme = lex_less;
                }
                break;
 
            case '!':
                if (*(cur+1) == '=')
                {
                    cur += 2;
                    _cur_lexeme = lex_not_equal;
                }
                else
                {
                    _cur_lexeme = lex_none;
                }
                break;
 
            case '=':
                cur += 1;
                _cur_lexeme = lex_equal;
 
                break;
 
            case '+':
                cur += 1;
                _cur_lexeme = lex_plus;
 
                break;
 
            case '-':
                cur += 1;
                _cur_lexeme = lex_minus;
 
                break;
 
            case '*':
                cur += 1;
                _cur_lexeme = lex_multiply;
 
                break;
 
            case '|':
                cur += 1;
                _cur_lexeme = lex_union;
 
                break;
 
            case '$':
                cur += 1;
 
                if (PUGI__IS_CHARTYPEX(*cur, ctx_start_symbol))
                {
                    _cur_lexeme_contents.begin = cur;
 
                    while (PUGI__IS_CHARTYPEX(*cur, ctx_symbol)) cur++;
 
                    if (cur[0] == ':' && PUGI__IS_CHARTYPEX(cur[1], ctx_symbol)) // qname
                    {
                        cur++; // :
 
                        while (PUGI__IS_CHARTYPEX(*cur, ctx_symbol)) cur++;
                    }
 
                    _cur_lexeme_contents.end = cur;
 
                    _cur_lexeme = lex_var_ref;
                }
                else
                {
                    _cur_lexeme = lex_none;
                }
 
                break;
 
            case '(':
                cur += 1;
                _cur_lexeme = lex_open_brace;
 
                break;
 
            case ')':
                cur += 1;
                _cur_lexeme = lex_close_brace;
 
                break;
 
            case '[':
                cur += 1;
                _cur_lexeme = lex_open_square_brace;
 
                break;
 
            case ']':
                cur += 1;
                _cur_lexeme = lex_close_square_brace;
 
                break;
 
            case ',':
                cur += 1;
                _cur_lexeme = lex_comma;
 
                break;
 
            case '/':
                if (*(cur+1) == '/')
                {
                    cur += 2;
                    _cur_lexeme = lex_double_slash;
                }
                else
                {
                    cur += 1;
                    _cur_lexeme = lex_slash;
                }
                break;
 
            case '.':
                if (*(cur+1) == '.')
                {
                    cur += 2;
                    _cur_lexeme = lex_double_dot;
                }
                else if (PUGI__IS_CHARTYPEX(*(cur+1), ctx_digit))
                {
                    _cur_lexeme_contents.begin = cur; // .
 
                    ++cur;
 
                    while (PUGI__IS_CHARTYPEX(*cur, ctx_digit)) cur++;
 
                    _cur_lexeme_contents.end = cur;
 
                    _cur_lexeme = lex_number;
                }
                else
                {
                    cur += 1;
                    _cur_lexeme = lex_dot;
                }
                break;
 
            case '@':
                cur += 1;
                _cur_lexeme = lex_axis_attribute;
 
                break;
 
            case '"':
            case '\'':
            {
                char_t terminator = *cur;
 
                ++cur;
 
                _cur_lexeme_contents.begin = cur;
                while (*cur && *cur != terminator) cur++;
                _cur_lexeme_contents.end = cur;
 
                if (!*cur)
                    _cur_lexeme = lex_none;
                else
                {
                    cur += 1;
                    _cur_lexeme = lex_quoted_string;
                }
 
                break;
            }
 
            case ':':
                if (*(cur+1) == ':')
                {
                    cur += 2;
                    _cur_lexeme = lex_double_colon;
                }
                else
                {
                    _cur_lexeme = lex_none;
                }
                break;
 
            default:
                if (PUGI__IS_CHARTYPEX(*cur, ctx_digit))
                {
                    _cur_lexeme_contents.begin = cur;
 
                    while (PUGI__IS_CHARTYPEX(*cur, ctx_digit)) cur++;
 
                    if (*cur == '.')
                    {
                        cur++;
 
                        while (PUGI__IS_CHARTYPEX(*cur, ctx_digit)) cur++;
                    }
 
                    _cur_lexeme_contents.end = cur;
 
                    _cur_lexeme = lex_number;
                }
                else if (PUGI__IS_CHARTYPEX(*cur, ctx_start_symbol))
                {
                    _cur_lexeme_contents.begin = cur;
 
                    while (PUGI__IS_CHARTYPEX(*cur, ctx_symbol)) cur++;
 
                    if (cur[0] == ':')
                    {
                        if (cur[1] == '*') // namespace test ncname:*
                        {
                            cur += 2; // :*
                        }
                        else if (PUGI__IS_CHARTYPEX(cur[1], ctx_symbol)) // namespace test qname
                        {
                            cur++; // :
 
                            while (PUGI__IS_CHARTYPEX(*cur, ctx_symbol)) cur++;
                        }
                    }
 
                    _cur_lexeme_contents.end = cur;
 
                    _cur_lexeme = lex_string;
                }
                else
                {
                    _cur_lexeme = lex_none;
                }
            }
 
            _cur = cur;
        }
 
        lexeme_t current() const
        {
            return _cur_lexeme;
        }
 
        const char_t* current_pos() const
        {
            return _cur_lexeme_pos;
        }
 
        const xpath_lexer_string& contents() const
        {
            assert(_cur_lexeme == lex_var_ref || _cur_lexeme == lex_number || _cur_lexeme == lex_string || _cur_lexeme == lex_quoted_string);
 
            return _cur_lexeme_contents;
        }
    };
 
    enum ast_type_t
    {
        ast_unknown,
        ast_op_or,                      // left or right
        ast_op_and,                     // left and right
        ast_op_equal,                   // left = right
        ast_op_not_equal,               // left != right
        ast_op_less,                    // left < right
        ast_op_greater,                 // left > right
        ast_op_less_or_equal,           // left <= right
        ast_op_greater_or_equal,        // left >= right
        ast_op_add,                     // left + right
        ast_op_subtract,                // left - right
        ast_op_multiply,                // left * right
        ast_op_divide,                  // left / right
        ast_op_mod,                     // left % right
        ast_op_negate,                  // left - right
        ast_op_union,                   // left | right
        ast_predicate,                  // apply predicate to set; next points to next predicate
        ast_filter,                     // select * from left where right
        ast_string_constant,            // string constant
        ast_number_constant,            // number constant
        ast_variable,                   // variable
        ast_func_last,                  // last()
        ast_func_position,              // position()
        ast_func_count,                 // count(left)
        ast_func_id,                    // id(left)
        ast_func_local_name_0,          // local-name()
        ast_func_local_name_1,          // local-name(left)
        ast_func_namespace_uri_0,       // namespace-uri()
        ast_func_namespace_uri_1,       // namespace-uri(left)
        ast_func_name_0,                // name()
        ast_func_name_1,                // name(left)
        ast_func_string_0,              // string()
        ast_func_string_1,              // string(left)
        ast_func_concat,                // concat(left, right, siblings)
        ast_func_starts_with,           // starts_with(left, right)
        ast_func_contains,              // contains(left, right)
        ast_func_substring_before,      // substring-before(left, right)
        ast_func_substring_after,       // substring-after(left, right)
        ast_func_substring_2,           // substring(left, right)
        ast_func_substring_3,           // substring(left, right, third)
        ast_func_string_length_0,       // string-length()
        ast_func_string_length_1,       // string-length(left)
        ast_func_normalize_space_0,     // normalize-space()
        ast_func_normalize_space_1,     // normalize-space(left)
        ast_func_translate,             // translate(left, right, third)
        ast_func_boolean,               // boolean(left)
        ast_func_not,                   // not(left)
        ast_func_true,                  // true()
        ast_func_false,                 // false()
        ast_func_lang,                  // lang(left)
        ast_func_number_0,              // number()
        ast_func_number_1,              // number(left)
        ast_func_sum,                   // sum(left)
        ast_func_floor,                 // floor(left)
        ast_func_ceiling,               // ceiling(left)
        ast_func_round,                 // round(left)
        ast_step,                       // process set left with step
        ast_step_root,                  // select root node
 
        ast_opt_translate_table,        // translate(left, right, third) where right/third are constants
        ast_opt_compare_attribute       // @name = 'string'
    };
 
    enum axis_t
    {
        axis_ancestor,
        axis_ancestor_or_self,
        axis_attribute,
        axis_child,
        axis_descendant,
        axis_descendant_or_self,
        axis_following,
        axis_following_sibling,
        axis_namespace,
        axis_parent,
        axis_preceding,
        axis_preceding_sibling,
        axis_self
    };
 
    enum nodetest_t
    {
        nodetest_none,
        nodetest_name,
        nodetest_type_node,
        nodetest_type_comment,
        nodetest_type_pi,
        nodetest_type_text,
        nodetest_pi,
        nodetest_all,
        nodetest_all_in_namespace
    };
 
    enum predicate_t
    {
        predicate_default,
        predicate_posinv,
        predicate_constant,
        predicate_constant_one
    };
 
    enum nodeset_eval_t
    {
        nodeset_eval_all,
        nodeset_eval_any,
        nodeset_eval_first
    };
 
    template <axis_t N> struct axis_to_type
    {
        static const axis_t axis;
    };
 
    template <axis_t N> const axis_t axis_to_type<N>::axis = N;
 
    class xpath_ast_node
    {
    private:
        // node type
        char _type;
        char _rettype;
 
        // for ast_step
        char _axis;
 
        // for ast_step/ast_predicate/ast_filter
        char _test;
 
        // tree node structure
        xpath_ast_node* _left;
        xpath_ast_node* _right;
        xpath_ast_node* _next;
 
        union
        {
            // value for ast_string_constant
            const char_t* string;
            // value for ast_number_constant
            double number;
            // variable for ast_variable
            xpath_variable* variable;
            // node test for ast_step (node name/namespace/node type/pi target)
            const char_t* nodetest;
            // table for ast_opt_translate_table
            const unsigned char* table;
        } _data;
 
        xpath_ast_node(const xpath_ast_node&);
        xpath_ast_node& operator=(const xpath_ast_node&);
 
        template <class Comp> static bool compare_eq(xpath_ast_node* lhs, xpath_ast_node* rhs, const xpath_context& c, const xpath_stack& stack, const Comp& comp)
        {
            xpath_value_type lt = lhs->rettype(), rt = rhs->rettype();
 
            if (lt != xpath_type_node_set && rt != xpath_type_node_set)
            {
                if (lt == xpath_type_boolean || rt == xpath_type_boolean)
                    return comp(lhs->eval_boolean(c, stack), rhs->eval_boolean(c, stack));
                else if (lt == xpath_type_number || rt == xpath_type_number)
                    return comp(lhs->eval_number(c, stack), rhs->eval_number(c, stack));
                else if (lt == xpath_type_string || rt == xpath_type_string)
                {
                    xpath_allocator_capture cr(stack.result);
 
                    xpath_string ls = lhs->eval_string(c, stack);
                    xpath_string rs = rhs->eval_string(c, stack);
 
                    return comp(ls, rs);
                }
            }
            else if (lt == xpath_type_node_set && rt == xpath_type_node_set)
            {
                xpath_allocator_capture cr(stack.result);
 
                xpath_node_set_raw ls = lhs->eval_node_set(c, stack, nodeset_eval_all);
                xpath_node_set_raw rs = rhs->eval_node_set(c, stack, nodeset_eval_all);
 
                for (const xpath_node* li = ls.begin(); li != ls.end(); ++li)
                    for (const xpath_node* ri = rs.begin(); ri != rs.end(); ++ri)
                    {
                        xpath_allocator_capture cri(stack.result);
 
                        if (comp(string_value(*li, stack.result), string_value(*ri, stack.result)))
                            return true;
                    }
 
                return false;
            }
            else
            {
                if (lt == xpath_type_node_set)
                {
                    swap(lhs, rhs);
                    swap(lt, rt);
                }
 
                if (lt == xpath_type_boolean)
                    return comp(lhs->eval_boolean(c, stack), rhs->eval_boolean(c, stack));
                else if (lt == xpath_type_number)
                {
                    xpath_allocator_capture cr(stack.result);
 
                    double l = lhs->eval_number(c, stack);
                    xpath_node_set_raw rs = rhs->eval_node_set(c, stack, nodeset_eval_all);
 
                    for (const xpath_node* ri = rs.begin(); ri != rs.end(); ++ri)
                    {
                        xpath_allocator_capture cri(stack.result);
 
                        if (comp(l, convert_string_to_number(string_value(*ri, stack.result).c_str())))
                            return true;
                    }
 
                    return false;
                }
                else if (lt == xpath_type_string)
                {
                    xpath_allocator_capture cr(stack.result);
 
                    xpath_string l = lhs->eval_string(c, stack);
                    xpath_node_set_raw rs = rhs->eval_node_set(c, stack, nodeset_eval_all);
 
                    for (const xpath_node* ri = rs.begin(); ri != rs.end(); ++ri)
                    {
                        xpath_allocator_capture cri(stack.result);
 
                        if (comp(l, string_value(*ri, stack.result)))
                            return true;
                    }
 
                    return false;
                }
            }
 
            assert(false && "Wrong types"); // unreachable
            return false;
        }
 
        static bool eval_once(xpath_node_set::type_t type, nodeset_eval_t eval)
        {
            return type == xpath_node_set::type_sorted ? eval != nodeset_eval_all : eval == nodeset_eval_any;
        }
 
        template <class Comp> static bool compare_rel(xpath_ast_node* lhs, xpath_ast_node* rhs, const xpath_context& c, const xpath_stack& stack, const Comp& comp)
        {
            xpath_value_type lt = lhs->rettype(), rt = rhs->rettype();
 
            if (lt != xpath_type_node_set && rt != xpath_type_node_set)
                return comp(lhs->eval_number(c, stack), rhs->eval_number(c, stack));
            else if (lt == xpath_type_node_set && rt == xpath_type_node_set)
            {
                xpath_allocator_capture cr(stack.result);
 
                xpath_node_set_raw ls = lhs->eval_node_set(c, stack, nodeset_eval_all);
                xpath_node_set_raw rs = rhs->eval_node_set(c, stack, nodeset_eval_all);
 
                for (const xpath_node* li = ls.begin(); li != ls.end(); ++li)
                {
                    xpath_allocator_capture cri(stack.result);
 
                    double l = convert_string_to_number(string_value(*li, stack.result).c_str());
 
                    for (const xpath_node* ri = rs.begin(); ri != rs.end(); ++ri)
                    {
                        xpath_allocator_capture crii(stack.result);
 
                        if (comp(l, convert_string_to_number(string_value(*ri, stack.result).c_str())))
                            return true;
                    }
                }
 
                return false;
            }
            else if (lt != xpath_type_node_set && rt == xpath_type_node_set)
            {
                xpath_allocator_capture cr(stack.result);
 
                double l = lhs->eval_number(c, stack);
                xpath_node_set_raw rs = rhs->eval_node_set(c, stack, nodeset_eval_all);
 
                for (const xpath_node* ri = rs.begin(); ri != rs.end(); ++ri)
                {
                    xpath_allocator_capture cri(stack.result);
 
                    if (comp(l, convert_string_to_number(string_value(*ri, stack.result).c_str())))
                        return true;
                }
 
                return false;
            }
            else if (lt == xpath_type_node_set && rt != xpath_type_node_set)
            {
                xpath_allocator_capture cr(stack.result);
 
                xpath_node_set_raw ls = lhs->eval_node_set(c, stack, nodeset_eval_all);
                double r = rhs->eval_number(c, stack);
 
                for (const xpath_node* li = ls.begin(); li != ls.end(); ++li)
                {
                    xpath_allocator_capture cri(stack.result);
 
                    if (comp(convert_string_to_number(string_value(*li, stack.result).c_str()), r))
                        return true;
                }
 
                return false;
            }
            else
            {
                assert(false && "Wrong types"); // unreachable
                return false;
            }
        }
 
        static void apply_predicate_boolean(xpath_node_set_raw& ns, size_t first, xpath_ast_node* expr, const xpath_stack& stack, bool once)
        {
            assert(ns.size() >= first);
            assert(expr->rettype() != xpath_type_number);
 
            size_t i = 1;
            size_t size = ns.size() - first;
 
            xpath_node* last = ns.begin() + first;
 
            // remove_if... or well, sort of
            for (xpath_node* it = last; it != ns.end(); ++it, ++i)
            {
                xpath_context c(*it, i, size);
 
                if (expr->eval_boolean(c, stack))
                {
                    *last++ = *it;
 
                    if (once) break;
                }
            }
 
            ns.truncate(last);
        }
 
        static void apply_predicate_number(xpath_node_set_raw& ns, size_t first, xpath_ast_node* expr, const xpath_stack& stack, bool once)
        {
            assert(ns.size() >= first);
            assert(expr->rettype() == xpath_type_number);
 
            size_t i = 1;
            size_t size = ns.size() - first;
 
            xpath_node* last = ns.begin() + first;
 
            // remove_if... or well, sort of
            for (xpath_node* it = last; it != ns.end(); ++it, ++i)
            {
                xpath_context c(*it, i, size);
 
                if (expr->eval_number(c, stack) == i)
                {
                    *last++ = *it;
 
                    if (once) break;
                }
            }
 
            ns.truncate(last);
        }
 
        static void apply_predicate_number_const(xpath_node_set_raw& ns, size_t first, xpath_ast_node* expr, const xpath_stack& stack)
        {
            assert(ns.size() >= first);
            assert(expr->rettype() == xpath_type_number);
 
            size_t size = ns.size() - first;
 
            xpath_node* last = ns.begin() + first;
 
            xpath_context c(xpath_node(), 1, size);
 
            double er = expr->eval_number(c, stack);
 
            if (er >= 1.0 && er <= size)
            {
                size_t eri = static_cast<size_t>(er);
 
                if (er == eri)
                {
                    xpath_node r = last[eri - 1];
 
                    *last++ = r;
                }
            }
 
            ns.truncate(last);
        }
 
        void apply_predicate(xpath_node_set_raw& ns, size_t first, const xpath_stack& stack, bool once)
        {
            if (ns.size() == first) return;
 
            assert(_type == ast_filter || _type == ast_predicate);
 
            if (_test == predicate_constant || _test == predicate_constant_one)
                apply_predicate_number_const(ns, first, _right, stack);
            else if (_right->rettype() == xpath_type_number)
                apply_predicate_number(ns, first, _right, stack, once);
            else
                apply_predicate_boolean(ns, first, _right, stack, once);
        }
 
        void apply_predicates(xpath_node_set_raw& ns, size_t first, const xpath_stack& stack, nodeset_eval_t eval)
        {
            if (ns.size() == first) return;
 
            bool last_once = eval_once(ns.type(), eval);
 
            for (xpath_ast_node* pred = _right; pred; pred = pred->_next)
                pred->apply_predicate(ns, first, stack, !pred->_next && last_once);
        }
 
        bool step_push(xpath_node_set_raw& ns, xml_attribute_struct* a, xml_node_struct* parent, xpath_allocator* alloc)
        {
            assert(a);
 
            const char_t* name = a->name ? a->name + 0 : PUGIXML_TEXT("");
 
            switch (_test)
            {
            case nodetest_name:
                if (strequal(name, _data.nodetest) && is_xpath_attribute(name))
                {
                    ns.push_back(xpath_node(xml_attribute(a), xml_node(parent)), alloc);
                    return true;
                }
                break;
 
            case nodetest_type_node:
            case nodetest_all:
                if (is_xpath_attribute(name))
                {
                    ns.push_back(xpath_node(xml_attribute(a), xml_node(parent)), alloc);
                    return true;
                }
                break;
 
            case nodetest_all_in_namespace:
                if (starts_with(name, _data.nodetest) && is_xpath_attribute(name))
                {
                    ns.push_back(xpath_node(xml_attribute(a), xml_node(parent)), alloc);
                    return true;
                }
                break;
 
            default:
                ;
            }
 
            return false;
        }
 
        bool step_push(xpath_node_set_raw& ns, xml_node_struct* n, xpath_allocator* alloc)
        {
            assert(n);
 
            xml_node_type type = PUGI__NODETYPE(n);
 
            switch (_test)
            {
            case nodetest_name:
                if (type == node_element && n->name && strequal(n->name, _data.nodetest))
                {
                    ns.push_back(xml_node(n), alloc);
                    return true;
                }
                break;
 
            case nodetest_type_node:
                ns.push_back(xml_node(n), alloc);
                return true;
 
            case nodetest_type_comment:
                if (type == node_comment)
                {
                    ns.push_back(xml_node(n), alloc);
                    return true;
                }
                break;
 
            case nodetest_type_text:
                if (type == node_pcdata || type == node_cdata)
                {
                    ns.push_back(xml_node(n), alloc);
                    return true;
                }
                break;
 
            case nodetest_type_pi:
                if (type == node_pi)
                {
                    ns.push_back(xml_node(n), alloc);
                    return true;
                }
                break;
 
            case nodetest_pi:
                if (type == node_pi && n->name && strequal(n->name, _data.nodetest))
                {
                    ns.push_back(xml_node(n), alloc);
                    return true;
                }
                break;
 
            case nodetest_all:
                if (type == node_element)
                {
                    ns.push_back(xml_node(n), alloc);
                    return true;
                }
                break;
 
            case nodetest_all_in_namespace:
                if (type == node_element && n->name && starts_with(n->name, _data.nodetest))
                {
                    ns.push_back(xml_node(n), alloc);
                    return true;
                }
                break;
 
            default:
                assert(false && "Unknown axis"); // unreachable
            }
 
            return false;
        }
 
        template <class T> void step_fill(xpath_node_set_raw& ns, xml_node_struct* n, xpath_allocator* alloc, bool once, T)
        {
            const axis_t axis = T::axis;
 
            switch (axis)
            {
            case axis_attribute:
            {
                for (xml_attribute_struct* a = n->first_attribute; a; a = a->next_attribute)
                    if (step_push(ns, a, n, alloc) & once)
                        return;
 
                break;
            }
 
            case axis_child:
            {
                for (xml_node_struct* c = n->first_child; c; c = c->next_sibling)
                    if (step_push(ns, c, alloc) & once)
                        return;
 
                break;
            }
 
            case axis_descendant:
            case axis_descendant_or_self:
            {
                if (axis == axis_descendant_or_self)
                    if (step_push(ns, n, alloc) & once)
                        return;
 
                xml_node_struct* cur = n->first_child;
 
                while (cur)
                {
                    if (step_push(ns, cur, alloc) & once)
                        return;
 
                    if (cur->first_child)
                        cur = cur->first_child;
                    else
                    {
                        while (!cur->next_sibling)
                        {
                            cur = cur->parent;
 
                            if (cur == n) return;
                        }
 
                        cur = cur->next_sibling;
                    }
                }
 
                break;
            }
 
            case axis_following_sibling:
            {
                for (xml_node_struct* c = n->next_sibling; c; c = c->next_sibling)
                    if (step_push(ns, c, alloc) & once)
                        return;
 
                break;
            }
 
            case axis_preceding_sibling:
            {
                for (xml_node_struct* c = n->prev_sibling_c; c->next_sibling; c = c->prev_sibling_c)
                    if (step_push(ns, c, alloc) & once)
                        return;
 
                break;
            }
 
            case axis_following:
            {
                xml_node_struct* cur = n;
 
                // exit from this node so that we don't include descendants
                while (!cur->next_sibling)
                {
                    cur = cur->parent;
 
                    if (!cur) return;
                }
 
                cur = cur->next_sibling;
 
                while (cur)
                {
                    if (step_push(ns, cur, alloc) & once)
                        return;
 
                    if (cur->first_child)
                        cur = cur->first_child;
                    else
                    {
                        while (!cur->next_sibling)
                        {
                            cur = cur->parent;
 
                            if (!cur) return;
                        }
 
                        cur = cur->next_sibling;
                    }
                }
 
                break;
            }
 
            case axis_preceding:
            {
                xml_node_struct* cur = n;
 
                // exit from this node so that we don't include descendants
                while (!cur->prev_sibling_c->next_sibling)
                {
                    cur = cur->parent;
 
                    if (!cur) return;
                }
 
                cur = cur->prev_sibling_c;
 
                while (cur)
                {
                    if (cur->first_child)
                        cur = cur->first_child->prev_sibling_c;
                    else
                    {
                        // leaf node, can't be ancestor
                        if (step_push(ns, cur, alloc) & once)
                            return;
 
                        while (!cur->prev_sibling_c->next_sibling)
                        {
                            cur = cur->parent;
 
                            if (!cur) return;
 
                            if (!node_is_ancestor(cur, n))
                                if (step_push(ns, cur, alloc) & once)
                                    return;
                        }
 
                        cur = cur->prev_sibling_c;
                    }
                }
 
                break;
            }
 
            case axis_ancestor:
            case axis_ancestor_or_self:
            {
                if (axis == axis_ancestor_or_self)
                    if (step_push(ns, n, alloc) & once)
                        return;
 
                xml_node_struct* cur = n->parent;
 
                while (cur)
                {
                    if (step_push(ns, cur, alloc) & once)
                        return;
 
                    cur = cur->parent;
                }
 
                break;
            }
 
            case axis_self:
            {
                step_push(ns, n, alloc);
 
                break;
            }
 
            case axis_parent:
            {
                if (n->parent)
                    step_push(ns, n->parent, alloc);
 
                break;
            }
 
            default:
                assert(false && "Unimplemented axis"); // unreachable
            }
        }
 
        template <class T> void step_fill(xpath_node_set_raw& ns, xml_attribute_struct* a, xml_node_struct* p, xpath_allocator* alloc, bool once, T v)
        {
            const axis_t axis = T::axis;
 
            switch (axis)
            {
            case axis_ancestor:
            case axis_ancestor_or_self:
            {
                if (axis == axis_ancestor_or_self && _test == nodetest_type_node) // reject attributes based on principal node type test
                    if (step_push(ns, a, p, alloc) & once)
                        return;
 
                xml_node_struct* cur = p;
 
                while (cur)
                {
                    if (step_push(ns, cur, alloc) & once)
                        return;
 
                    cur = cur->parent;
                }
 
                break;
            }
 
            case axis_descendant_or_self:
            case axis_self:
            {
                if (_test == nodetest_type_node) // reject attributes based on principal node type test
                    step_push(ns, a, p, alloc);
 
                break;
            }
 
            case axis_following:
            {
                xml_node_struct* cur = p;
 
                while (cur)
                {
                    if (cur->first_child)
                        cur = cur->first_child;
                    else
                    {
                        while (!cur->next_sibling)
                        {
                            cur = cur->parent;
 
                            if (!cur) return;
                        }
 
                        cur = cur->next_sibling;
                    }
 
                    if (step_push(ns, cur, alloc) & once)
                        return;
                }
 
                break;
            }
 
            case axis_parent:
            {
                step_push(ns, p, alloc);
 
                break;
            }
 
            case axis_preceding:
            {
                // preceding:: axis does not include attribute nodes and attribute ancestors (they are the same as parent's ancestors), so we can reuse node preceding
                step_fill(ns, p, alloc, once, v);
                break;
            }
 
            default:
                assert(false && "Unimplemented axis"); // unreachable
            }
        }
 
        template <class T> void step_fill(xpath_node_set_raw& ns, const xpath_node& xn, xpath_allocator* alloc, bool once, T v)
        {
            const axis_t axis = T::axis;
            const bool axis_has_attributes = (axis == axis_ancestor || axis == axis_ancestor_or_self || axis == axis_descendant_or_self || axis == axis_following || axis == axis_parent || axis == axis_preceding || axis == axis_self);
 
            if (xn.node())
                step_fill(ns, xn.node().internal_object(), alloc, once, v);
            else if (axis_has_attributes && xn.attribute() && xn.parent())
                step_fill(ns, xn.attribute().internal_object(), xn.parent().internal_object(), alloc, once, v);
        }
 
        template <class T> xpath_node_set_raw step_do(const xpath_context& c, const xpath_stack& stack, nodeset_eval_t eval, T v)
        {
            const axis_t axis = T::axis;
            const bool axis_reverse = (axis == axis_ancestor || axis == axis_ancestor_or_self || axis == axis_preceding || axis == axis_preceding_sibling);
            const xpath_node_set::type_t axis_type = axis_reverse ? xpath_node_set::type_sorted_reverse : xpath_node_set::type_sorted;
 
            bool once =
                (axis == axis_attribute && _test == nodetest_name) ||
                (!_right && eval_once(axis_type, eval)) ||
                (_right && !_right->_next && _right->_test == predicate_constant_one);
 
            xpath_node_set_raw ns;
            ns.set_type(axis_type);
 
            if (_left)
            {
                xpath_node_set_raw s = _left->eval_node_set(c, stack, nodeset_eval_all);
 
                // self axis preserves the original order
                if (axis == axis_self) ns.set_type(s.type());
 
                for (const xpath_node* it = s.begin(); it != s.end(); ++it)
                {
                    size_t size = ns.size();
 
                    // in general, all axes generate elements in a particular order, but there is no order guarantee if axis is applied to two nodes
                    if (axis != axis_self && size != 0) ns.set_type(xpath_node_set::type_unsorted);
 
                    step_fill(ns, *it, stack.result, once, v);
                    if (_right) apply_predicates(ns, size, stack, eval);
                }
            }
            else
            {
                step_fill(ns, c.n, stack.result, once, v);
                if (_right) apply_predicates(ns, 0, stack, eval);
            }
 
            // child, attribute and self axes always generate unique set of nodes
            // for other axis, if the set stayed sorted, it stayed unique because the traversal algorithms do not visit the same node twice
            if (axis != axis_child && axis != axis_attribute && axis != axis_self && ns.type() == xpath_node_set::type_unsorted)
                ns.remove_duplicates();
 
            return ns;
        }
 
    public:
        xpath_ast_node(ast_type_t type, xpath_value_type rettype_, const char_t* value):
            _type(static_cast<char>(type)), _rettype(static_cast<char>(rettype_)), _axis(0), _test(0), _left(0), _right(0), _next(0)
        {
            assert(type == ast_string_constant);
            _data.string = value;
        }
 
        xpath_ast_node(ast_type_t type, xpath_value_type rettype_, double value):
            _type(static_cast<char>(type)), _rettype(static_cast<char>(rettype_)), _axis(0), _test(0), _left(0), _right(0), _next(0)
        {
            assert(type == ast_number_constant);
            _data.number = value;
        }
 
        xpath_ast_node(ast_type_t type, xpath_value_type rettype_, xpath_variable* value):
            _type(static_cast<char>(type)), _rettype(static_cast<char>(rettype_)), _axis(0), _test(0), _left(0), _right(0), _next(0)
        {
            assert(type == ast_variable);
            _data.variable = value;
        }
 
        xpath_ast_node(ast_type_t type, xpath_value_type rettype_, xpath_ast_node* left = 0, xpath_ast_node* right = 0):
            _type(static_cast<char>(type)), _rettype(static_cast<char>(rettype_)), _axis(0), _test(0), _left(left), _right(right), _next(0)
        {
        }
 
        xpath_ast_node(ast_type_t type, xpath_ast_node* left, axis_t axis, nodetest_t test, const char_t* contents):
            _type(static_cast<char>(type)), _rettype(xpath_type_node_set), _axis(static_cast<char>(axis)), _test(static_cast<char>(test)), _left(left), _right(0), _next(0)
        {
            assert(type == ast_step);
            _data.nodetest = contents;
        }
 
        xpath_ast_node(ast_type_t type, xpath_ast_node* left, xpath_ast_node* right, predicate_t test):
            _type(static_cast<char>(type)), _rettype(xpath_type_node_set), _axis(0), _test(static_cast<char>(test)), _left(left), _right(right), _next(0)
        {
            assert(type == ast_filter || type == ast_predicate);
        }
 
        void set_next(xpath_ast_node* value)
        {
            _next = value;
        }
 
        void set_right(xpath_ast_node* value)
        {
            _right = value;
        }
 
        bool eval_boolean(const xpath_context& c, const xpath_stack& stack)
        {
            switch (_type)
            {
            case ast_op_or:
                return _left->eval_boolean(c, stack) || _right->eval_boolean(c, stack);
 
            case ast_op_and:
                return _left->eval_boolean(c, stack) && _right->eval_boolean(c, stack);
 
            case ast_op_equal:
                return compare_eq(_left, _right, c, stack, equal_to());
 
            case ast_op_not_equal:
                return compare_eq(_left, _right, c, stack, not_equal_to());
 
            case ast_op_less:
                return compare_rel(_left, _right, c, stack, less());
 
            case ast_op_greater:
                return compare_rel(_right, _left, c, stack, less());
 
            case ast_op_less_or_equal:
                return compare_rel(_left, _right, c, stack, less_equal());
 
            case ast_op_greater_or_equal:
                return compare_rel(_right, _left, c, stack, less_equal());
 
            case ast_func_starts_with:
            {
                xpath_allocator_capture cr(stack.result);
 
                xpath_string lr = _left->eval_string(c, stack);
                xpath_string rr = _right->eval_string(c, stack);
 
                return starts_with(lr.c_str(), rr.c_str());
            }
 
            case ast_func_contains:
            {
                xpath_allocator_capture cr(stack.result);
 
                xpath_string lr = _left->eval_string(c, stack);
                xpath_string rr = _right->eval_string(c, stack);
 
                return find_substring(lr.c_str(), rr.c_str()) != 0;
            }
 
            case ast_func_boolean:
                return _left->eval_boolean(c, stack);
 
            case ast_func_not:
                return !_left->eval_boolean(c, stack);
 
            case ast_func_true:
                return true;
 
            case ast_func_false:
                return false;
 
            case ast_func_lang:
            {
                if (c.n.attribute()) return false;
 
                xpath_allocator_capture cr(stack.result);
 
                xpath_string lang = _left->eval_string(c, stack);
 
                for (xml_node n = c.n.node(); n; n = n.parent())
                {
                    xml_attribute a = n.attribute(PUGIXML_TEXT("xml:lang"));
 
                    if (a)
                    {
                        const char_t* value = a.value();
 
                        // strnicmp / strncasecmp is not portable
                        for (const char_t* lit = lang.c_str(); *lit; ++lit)
                        {
                            if (tolower_ascii(*lit) != tolower_ascii(*value)) return false;
                            ++value;
                        }
 
                        return *value == 0 || *value == '-';
                    }
                }
 
                return false;
            }
 
            case ast_opt_compare_attribute:
            {
                const char_t* value = (_right->_type == ast_string_constant) ? _right->_data.string : _right->_data.variable->get_string();
 
                xml_attribute attr = c.n.node().attribute(_left->_data.nodetest);
 
                return attr && strequal(attr.value(), value) && is_xpath_attribute(attr.name());
            }
 
            case ast_variable:
            {
                assert(_rettype == _data.variable->type());
 
                if (_rettype == xpath_type_boolean)
                    return _data.variable->get_boolean();
            }
 
            // fallthrough
            default:
            {
                switch (_rettype)
                {
                case xpath_type_number:
                    return convert_number_to_boolean(eval_number(c, stack));
 
                case xpath_type_string:
                {
                    xpath_allocator_capture cr(stack.result);
 
                    return !eval_string(c, stack).empty();
                }
 
                case xpath_type_node_set:
                {
                    xpath_allocator_capture cr(stack.result);
 
                    return !eval_node_set(c, stack, nodeset_eval_any).empty();
                }
 
                default:
                    assert(false && "Wrong expression for return type boolean"); // unreachable
                    return false;
                }
            }
            }
        }
 
        double eval_number(const xpath_context& c, const xpath_stack& stack)
        {
            switch (_type)
            {
            case ast_op_add:
                return _left->eval_number(c, stack) + _right->eval_number(c, stack);
 
            case ast_op_subtract:
                return _left->eval_number(c, stack) - _right->eval_number(c, stack);
 
            case ast_op_multiply:
                return _left->eval_number(c, stack) * _right->eval_number(c, stack);
 
            case ast_op_divide:
                return _left->eval_number(c, stack) / _right->eval_number(c, stack);
 
            case ast_op_mod:
                return fmod(_left->eval_number(c, stack), _right->eval_number(c, stack));
 
            case ast_op_negate:
                return -_left->eval_number(c, stack);
 
            case ast_number_constant:
                return _data.number;
 
            case ast_func_last:
                return static_cast<double>(c.size);
 
            case ast_func_position:
                return static_cast<double>(c.position);
 
            case ast_func_count:
            {
                xpath_allocator_capture cr(stack.result);
 
                return static_cast<double>(_left->eval_node_set(c, stack, nodeset_eval_all).size());
            }
 
            case ast_func_string_length_0:
            {
                xpath_allocator_capture cr(stack.result);
 
                return static_cast<double>(string_value(c.n, stack.result).length());
            }
 
            case ast_func_string_length_1:
            {
                xpath_allocator_capture cr(stack.result);
 
                return static_cast<double>(_left->eval_string(c, stack).length());
            }
 
            case ast_func_number_0:
            {
                xpath_allocator_capture cr(stack.result);
 
                return convert_string_to_number(string_value(c.n, stack.result).c_str());
            }
 
            case ast_func_number_1:
                return _left->eval_number(c, stack);
 
            case ast_func_sum:
            {
                xpath_allocator_capture cr(stack.result);
 
                double r = 0;
 
                xpath_node_set_raw ns = _left->eval_node_set(c, stack, nodeset_eval_all);
 
                for (const xpath_node* it = ns.begin(); it != ns.end(); ++it)
                {
                    xpath_allocator_capture cri(stack.result);
 
                    r += convert_string_to_number(string_value(*it, stack.result).c_str());
                }
 
                return r;
            }
 
            case ast_func_floor:
            {
                double r = _left->eval_number(c, stack);
 
                return r == r ? floor(r) : r;
            }
 
            case ast_func_ceiling:
            {
                double r = _left->eval_number(c, stack);
 
                return r == r ? ceil(r) : r;
            }
 
            case ast_func_round:
                return round_nearest_nzero(_left->eval_number(c, stack));
 
            case ast_variable:
            {
                assert(_rettype == _data.variable->type());
 
                if (_rettype == xpath_type_number)
                    return _data.variable->get_number();
            }
 
            // fallthrough
            default:
            {
                switch (_rettype)
                {
                case xpath_type_boolean:
                    return eval_boolean(c, stack) ? 1 : 0;
 
                case xpath_type_string:
                {
                    xpath_allocator_capture cr(stack.result);
 
                    return convert_string_to_number(eval_string(c, stack).c_str());
                }
 
                case xpath_type_node_set:
                {
                    xpath_allocator_capture cr(stack.result);
 
                    return convert_string_to_number(eval_string(c, stack).c_str());
                }
 
                default:
                    assert(false && "Wrong expression for return type number"); // unreachable
                    return 0;
                }
 
            }
            }
        }
 
        xpath_string eval_string_concat(const xpath_context& c, const xpath_stack& stack)
        {
            assert(_type == ast_func_concat);
 
            xpath_allocator_capture ct(stack.temp);
 
            // count the string number
            size_t count = 1;
            for (xpath_ast_node* nc = _right; nc; nc = nc->_next) count++;
 
            // allocate a buffer for temporary string objects
            xpath_string* buffer = static_cast<xpath_string*>(stack.temp->allocate(count * sizeof(xpath_string)));
            if (!buffer) return xpath_string();
 
            // evaluate all strings to temporary stack
            xpath_stack swapped_stack = {stack.temp, stack.result};
 
            buffer[0] = _left->eval_string(c, swapped_stack);
 
            size_t pos = 1;
            for (xpath_ast_node* n = _right; n; n = n->_next, ++pos) buffer[pos] = n->eval_string(c, swapped_stack);
            assert(pos == count);
 
            // get total length
            size_t length = 0;
            for (size_t i = 0; i < count; ++i) length += buffer[i].length();
 
            // create final string
            char_t* result = static_cast<char_t*>(stack.result->allocate((length + 1) * sizeof(char_t)));
            if (!result) return xpath_string();
 
            char_t* ri = result;
 
            for (size_t j = 0; j < count; ++j)
                for (const char_t* bi = buffer[j].c_str(); *bi; ++bi)
                    *ri++ = *bi;
 
            *ri = 0;
 
            return xpath_string::from_heap_preallocated(result, ri);
        }
 
        xpath_string eval_string(const xpath_context& c, const xpath_stack& stack)
        {
            switch (_type)
            {
            case ast_string_constant:
                return xpath_string::from_const(_data.string);
 
            case ast_func_local_name_0:
            {
                xpath_node na = c.n;
 
                return xpath_string::from_const(local_name(na));
            }
 
            case ast_func_local_name_1:
            {
                xpath_allocator_capture cr(stack.result);
 
                xpath_node_set_raw ns = _left->eval_node_set(c, stack, nodeset_eval_first);
                xpath_node na = ns.first();
 
                return xpath_string::from_const(local_name(na));
            }
 
            case ast_func_name_0:
            {
                xpath_node na = c.n;
 
                return xpath_string::from_const(qualified_name(na));
            }
 
            case ast_func_name_1:
            {
                xpath_allocator_capture cr(stack.result);
 
                xpath_node_set_raw ns = _left->eval_node_set(c, stack, nodeset_eval_first);
                xpath_node na = ns.first();
 
                return xpath_string::from_const(qualified_name(na));
            }
 
            case ast_func_namespace_uri_0:
            {
                xpath_node na = c.n;
 
                return xpath_string::from_const(namespace_uri(na));
            }
 
            case ast_func_namespace_uri_1:
            {
                xpath_allocator_capture cr(stack.result);
 
                xpath_node_set_raw ns = _left->eval_node_set(c, stack, nodeset_eval_first);
                xpath_node na = ns.first();
 
                return xpath_string::from_const(namespace_uri(na));
            }
 
            case ast_func_string_0:
                return string_value(c.n, stack.result);
 
            case ast_func_string_1:
                return _left->eval_string(c, stack);
 
            case ast_func_concat:
                return eval_string_concat(c, stack);
 
            case ast_func_substring_before:
            {
                xpath_allocator_capture cr(stack.temp);
 
                xpath_stack swapped_stack = {stack.temp, stack.result};
 
                xpath_string s = _left->eval_string(c, swapped_stack);
                xpath_string p = _right->eval_string(c, swapped_stack);
 
                const char_t* pos = find_substring(s.c_str(), p.c_str());
 
                return pos ? xpath_string::from_heap(s.c_str(), pos, stack.result) : xpath_string();
            }
 
            case ast_func_substring_after:
            {
                xpath_allocator_capture cr(stack.temp);
 
                xpath_stack swapped_stack = {stack.temp, stack.result};
 
                xpath_string s = _left->eval_string(c, swapped_stack);
                xpath_string p = _right->eval_string(c, swapped_stack);
 
                const char_t* pos = find_substring(s.c_str(), p.c_str());
                if (!pos) return xpath_string();
 
                const char_t* rbegin = pos + p.length();
                const char_t* rend = s.c_str() + s.length();
 
                return s.uses_heap() ? xpath_string::from_heap(rbegin, rend, stack.result) : xpath_string::from_const(rbegin);
            }
 
            case ast_func_substring_2:
            {
                xpath_allocator_capture cr(stack.temp);
 
                xpath_stack swapped_stack = {stack.temp, stack.result};
 
                xpath_string s = _left->eval_string(c, swapped_stack);
                size_t s_length = s.length();
 
                double first = round_nearest(_right->eval_number(c, stack));
 
                if (is_nan(first)) return xpath_string(); // NaN
                else if (first >= s_length + 1) return xpath_string();
 
                size_t pos = first < 1 ? 1 : static_cast<size_t>(first);
                assert(1 <= pos && pos <= s_length + 1);
 
                const char_t* rbegin = s.c_str() + (pos - 1);
                const char_t* rend = s.c_str() + s.length();
 
                return s.uses_heap() ? xpath_string::from_heap(rbegin, rend, stack.result) : xpath_string::from_const(rbegin);
            }
 
            case ast_func_substring_3:
            {
                xpath_allocator_capture cr(stack.temp);
 
                xpath_stack swapped_stack = {stack.temp, stack.result};
 
                xpath_string s = _left->eval_string(c, swapped_stack);
                size_t s_length = s.length();
 
                double first = round_nearest(_right->eval_number(c, stack));
                double last = first + round_nearest(_right->_next->eval_number(c, stack));
 
                if (is_nan(first) || is_nan(last)) return xpath_string();
                else if (first >= s_length + 1) return xpath_string();
                else if (first >= last) return xpath_string();
                else if (last < 1) return xpath_string();
 
                size_t pos = first < 1 ? 1 : static_cast<size_t>(first);
                size_t end = last >= s_length + 1 ? s_length + 1 : static_cast<size_t>(last);
 
                assert(1 <= pos && pos <= end && end <= s_length + 1);
                const char_t* rbegin = s.c_str() + (pos - 1);
                const char_t* rend = s.c_str() + (end - 1);
 
                return (end == s_length + 1 && !s.uses_heap()) ? xpath_string::from_const(rbegin) : xpath_string::from_heap(rbegin, rend, stack.result);
            }
 
            case ast_func_normalize_space_0:
            {
                xpath_string s = string_value(c.n, stack.result);
 
                char_t* begin = s.data(stack.result);
                if (!begin) return xpath_string();
 
                char_t* end = normalize_space(begin);
 
                return xpath_string::from_heap_preallocated(begin, end);
            }
 
            case ast_func_normalize_space_1:
            {
                xpath_string s = _left->eval_string(c, stack);
 
                char_t* begin = s.data(stack.result);
                if (!begin) return xpath_string();
 
                char_t* end = normalize_space(begin);
 
                return xpath_string::from_heap_preallocated(begin, end);
            }
 
            case ast_func_translate:
            {
                xpath_allocator_capture cr(stack.temp);
 
                xpath_stack swapped_stack = {stack.temp, stack.result};
 
                xpath_string s = _left->eval_string(c, stack);
                xpath_string from = _right->eval_string(c, swapped_stack);
                xpath_string to = _right->_next->eval_string(c, swapped_stack);
 
                char_t* begin = s.data(stack.result);
                if (!begin) return xpath_string();
 
                char_t* end = translate(begin, from.c_str(), to.c_str(), to.length());
 
                return xpath_string::from_heap_preallocated(begin, end);
            }
 
            case ast_opt_translate_table:
            {
                xpath_string s = _left->eval_string(c, stack);
 
                char_t* begin = s.data(stack.result);
                if (!begin) return xpath_string();
 
                char_t* end = translate_table(begin, _data.table);
 
                return xpath_string::from_heap_preallocated(begin, end);
            }
 
            case ast_variable:
            {
                assert(_rettype == _data.variable->type());
 
                if (_rettype == xpath_type_string)
                    return xpath_string::from_const(_data.variable->get_string());
            }
 
            // fallthrough
            default:
            {
                switch (_rettype)
                {
                case xpath_type_boolean:
                    return xpath_string::from_const(eval_boolean(c, stack) ? PUGIXML_TEXT("true") : PUGIXML_TEXT("false"));
 
                case xpath_type_number:
                    return convert_number_to_string(eval_number(c, stack), stack.result);
 
                case xpath_type_node_set:
                {
                    xpath_allocator_capture cr(stack.temp);
 
                    xpath_stack swapped_stack = {stack.temp, stack.result};
 
                    xpath_node_set_raw ns = eval_node_set(c, swapped_stack, nodeset_eval_first);
                    return ns.empty() ? xpath_string() : string_value(ns.first(), stack.result);
                }
 
                default:
                    assert(false && "Wrong expression for return type string"); // unreachable
                    return xpath_string();
                }
            }
            }
        }
 
        xpath_node_set_raw eval_node_set(const xpath_context& c, const xpath_stack& stack, nodeset_eval_t eval)
        {
            switch (_type)
            {
            case ast_op_union:
            {
                xpath_allocator_capture cr(stack.temp);
 
                xpath_stack swapped_stack = {stack.temp, stack.result};
 
                xpath_node_set_raw ls = _left->eval_node_set(c, swapped_stack, eval);
                xpath_node_set_raw rs = _right->eval_node_set(c, stack, eval);
 
                // we can optimize merging two sorted sets, but this is a very rare operation, so don't bother
                rs.set_type(xpath_node_set::type_unsorted);
 
                rs.append(ls.begin(), ls.end(), stack.result);
                rs.remove_duplicates();
 
                return rs;
            }
 
            case ast_filter:
            {
                xpath_node_set_raw set = _left->eval_node_set(c, stack, _test == predicate_constant_one ? nodeset_eval_first : nodeset_eval_all);
 
                // either expression is a number or it contains position() call; sort by document order
                if (_test != predicate_posinv) set.sort_do();
 
                bool once = eval_once(set.type(), eval);
 
                apply_predicate(set, 0, stack, once);
 
                return set;
            }
 
            case ast_func_id:
                return xpath_node_set_raw();
 
            case ast_step:
            {
                switch (_axis)
                {
                case axis_ancestor:
                    return step_do(c, stack, eval, axis_to_type<axis_ancestor>());
 
                case axis_ancestor_or_self:
                    return step_do(c, stack, eval, axis_to_type<axis_ancestor_or_self>());
 
                case axis_attribute:
                    return step_do(c, stack, eval, axis_to_type<axis_attribute>());
 
                case axis_child:
                    return step_do(c, stack, eval, axis_to_type<axis_child>());
 
                case axis_descendant:
                    return step_do(c, stack, eval, axis_to_type<axis_descendant>());
 
                case axis_descendant_or_self:
                    return step_do(c, stack, eval, axis_to_type<axis_descendant_or_self>());
 
                case axis_following:
                    return step_do(c, stack, eval, axis_to_type<axis_following>());
 
                case axis_following_sibling:
                    return step_do(c, stack, eval, axis_to_type<axis_following_sibling>());
 
                case axis_namespace:
                    // namespaced axis is not supported
                    return xpath_node_set_raw();
 
                case axis_parent:
                    return step_do(c, stack, eval, axis_to_type<axis_parent>());
 
                case axis_preceding:
                    return step_do(c, stack, eval, axis_to_type<axis_preceding>());
 
                case axis_preceding_sibling:
                    return step_do(c, stack, eval, axis_to_type<axis_preceding_sibling>());
 
                case axis_self:
                    return step_do(c, stack, eval, axis_to_type<axis_self>());
 
                default:
                    assert(false && "Unknown axis"); // unreachable
                    return xpath_node_set_raw();
                }
            }
 
            case ast_step_root:
            {
                assert(!_right); // root step can't have any predicates
 
                xpath_node_set_raw ns;
 
                ns.set_type(xpath_node_set::type_sorted);
 
                if (c.n.node()) ns.push_back(c.n.node().root(), stack.result);
                else if (c.n.attribute()) ns.push_back(c.n.parent().root(), stack.result);
 
                return ns;
            }
 
            case ast_variable:
            {
                assert(_rettype == _data.variable->type());
 
                if (_rettype == xpath_type_node_set)
                {
                    const xpath_node_set& s = _data.variable->get_node_set();
 
                    xpath_node_set_raw ns;
 
                    ns.set_type(s.type());
                    ns.append(s.begin(), s.end(), stack.result);
 
                    return ns;
                }
            }
 
            // fallthrough
            default:
                assert(false && "Wrong expression for return type node set"); // unreachable
                return xpath_node_set_raw();
            }
        }
 
        void optimize(xpath_allocator* alloc)
        {
            if (_left)
                _left->optimize(alloc);
 
            if (_right)
                _right->optimize(alloc);
 
            if (_next)
                _next->optimize(alloc);
 
            optimize_self(alloc);
        }
 
        void optimize_self(xpath_allocator* alloc)
        {
            // Rewrite [position()=expr] with [expr]
            // Note that this step has to go before classification to recognize [position()=1]
            if ((_type == ast_filter || _type == ast_predicate) &&
                _right->_type == ast_op_equal && _right->_left->_type == ast_func_position && _right->_right->_rettype == xpath_type_number)
            {
                _right = _right->_right;
            }
 
            // Classify filter/predicate ops to perform various optimizations during evaluation
            if (_type == ast_filter || _type == ast_predicate)
            {
                assert(_test == predicate_default);
 
                if (_right->_type == ast_number_constant && _right->_data.number == 1.0)
                    _test = predicate_constant_one;
                else if (_right->_rettype == xpath_type_number && (_right->_type == ast_number_constant || _right->_type == ast_variable || _right->_type == ast_func_last))
                    _test = predicate_constant;
                else if (_right->_rettype != xpath_type_number && _right->is_posinv_expr())
                    _test = predicate_posinv;
            }
 
            // Rewrite descendant-or-self::node()/child::foo with descendant::foo
            // The former is a full form of //foo, the latter is much faster since it executes the node test immediately
            // Do a similar kind of rewrite for self/descendant/descendant-or-self axes
            // Note that we only rewrite positionally invariant steps (//foo[1] != /descendant::foo[1])
            if (_type == ast_step && (_axis == axis_child || _axis == axis_self || _axis == axis_descendant || _axis == axis_descendant_or_self) && _left &&
                _left->_type == ast_step && _left->_axis == axis_descendant_or_self && _left->_test == nodetest_type_node && !_left->_right &&
                is_posinv_step())
            {
                if (_axis == axis_child || _axis == axis_descendant)
                    _axis = axis_descendant;
                else
                    _axis = axis_descendant_or_self;
 
                _left = _left->_left;
            }
 
            // Use optimized lookup table implementation for translate() with constant arguments
            if (_type == ast_func_translate && _right->_type == ast_string_constant && _right->_next->_type == ast_string_constant)
            {
                unsigned char* table = translate_table_generate(alloc, _right->_data.string, _right->_next->_data.string);
 
                if (table)
                {
                    _type = ast_opt_translate_table;
                    _data.table = table;
                }
            }
 
            // Use optimized path for @attr = 'value' or @attr = $value
            if (_type == ast_op_equal &&
                _left->_type == ast_step && _left->_axis == axis_attribute && _left->_test == nodetest_name && !_left->_left && !_left->_right &&
                (_right->_type == ast_string_constant || (_right->_type == ast_variable && _right->_rettype == xpath_type_string)))
            {
                _type = ast_opt_compare_attribute;
            }
        }
 
        bool is_posinv_expr() const
        {
            switch (_type)
            {
            case ast_func_position:
            case ast_func_last:
                return false;
 
            case ast_string_constant:
            case ast_number_constant:
            case ast_variable:
                return true;
 
            case ast_step:
            case ast_step_root:
                return true;
 
            case ast_predicate:
            case ast_filter:
                return true;
 
            default:
                if (_left && !_left->is_posinv_expr()) return false;
 
                for (xpath_ast_node* n = _right; n; n = n->_next)
                    if (!n->is_posinv_expr()) return false;
 
                return true;
            }
        }
 
        bool is_posinv_step() const
        {
            assert(_type == ast_step);
 
            for (xpath_ast_node* n = _right; n; n = n->_next)
            {
                assert(n->_type == ast_predicate);
 
                if (n->_test != predicate_posinv)
                    return false;
            }
 
            return true;
        }
 
        xpath_value_type rettype() const
        {
            return static_cast<xpath_value_type>(_rettype);
        }
    };
 
    struct xpath_parser
    {
        xpath_allocator* _alloc;
        xpath_lexer _lexer;
 
        const char_t* _query;
        xpath_variable_set* _variables;
 
        xpath_parse_result* _result;
 
        char_t _scratch[32];
 
        xpath_ast_node* error(const char* message)
        {
            _result->error = message;
            _result->offset = _lexer.current_pos() - _query;
 
            return 0;
        }
 
        xpath_ast_node* error_oom()
        {
            assert(_alloc->_error);
            *_alloc->_error = true;
 
            return 0;
        }
 
        void* alloc_node()
        {
            return _alloc->allocate(sizeof(xpath_ast_node));
        }
 
        xpath_ast_node* alloc_node(ast_type_t type, xpath_value_type rettype, const char_t* value)
        {
            void* memory = alloc_node();
            return memory ? new (memory) xpath_ast_node(type, rettype, value) : 0;
        }
 
        xpath_ast_node* alloc_node(ast_type_t type, xpath_value_type rettype, double value)
        {
            void* memory = alloc_node();
            return memory ? new (memory) xpath_ast_node(type, rettype, value) : 0;
        }
 
        xpath_ast_node* alloc_node(ast_type_t type, xpath_value_type rettype, xpath_variable* value)
        {
            void* memory = alloc_node();
            return memory ? new (memory) xpath_ast_node(type, rettype, value) : 0;
        }
 
        xpath_ast_node* alloc_node(ast_type_t type, xpath_value_type rettype, xpath_ast_node* left = 0, xpath_ast_node* right = 0)
        {
            void* memory = alloc_node();
            return memory ? new (memory) xpath_ast_node(type, rettype, left, right) : 0;
        }
 
        xpath_ast_node* alloc_node(ast_type_t type, xpath_ast_node* left, axis_t axis, nodetest_t test, const char_t* contents)
        {
            void* memory = alloc_node();
            return memory ? new (memory) xpath_ast_node(type, left, axis, test, contents) : 0;
        }
 
        xpath_ast_node* alloc_node(ast_type_t type, xpath_ast_node* left, xpath_ast_node* right, predicate_t test)
        {
            void* memory = alloc_node();
            return memory ? new (memory) xpath_ast_node(type, left, right, test) : 0;
        }
 
        const char_t* alloc_string(const xpath_lexer_string& value)
        {
            if (!value.begin)
                return PUGIXML_TEXT("");
 
            size_t length = static_cast<size_t>(value.end - value.begin);
 
            char_t* c = static_cast<char_t*>(_alloc->allocate((length + 1) * sizeof(char_t)));
            if (!c) return 0;
 
            memcpy(c, value.begin, length * sizeof(char_t));
            c[length] = 0;
 
            return c;
        }
 
        xpath_ast_node* parse_function(const xpath_lexer_string& name, size_t argc, xpath_ast_node* args[2])
        {
            switch (name.begin[0])
            {
            case 'b':
                if (name == PUGIXML_TEXT("boolean") && argc == 1)
                    return alloc_node(ast_func_boolean, xpath_type_boolean, args[0]);
 
                break;
 
            case 'c':
                if (name == PUGIXML_TEXT("count") && argc == 1)
                {
                    if (args[0]->rettype() != xpath_type_node_set) return error("Function has to be applied to node set");
                    return alloc_node(ast_func_count, xpath_type_number, args[0]);
                }
                else if (name == PUGIXML_TEXT("contains") && argc == 2)
                    return alloc_node(ast_func_contains, xpath_type_boolean, args[0], args[1]);
                else if (name == PUGIXML_TEXT("concat") && argc >= 2)
                    return alloc_node(ast_func_concat, xpath_type_string, args[0], args[1]);
                else if (name == PUGIXML_TEXT("ceiling") && argc == 1)
                    return alloc_node(ast_func_ceiling, xpath_type_number, args[0]);
 
                break;
 
            case 'f':
                if (name == PUGIXML_TEXT("false") && argc == 0)
                    return alloc_node(ast_func_false, xpath_type_boolean);
                else if (name == PUGIXML_TEXT("floor") && argc == 1)
                    return alloc_node(ast_func_floor, xpath_type_number, args[0]);
 
                break;
 
            case 'i':
                if (name == PUGIXML_TEXT("id") && argc == 1)
                    return alloc_node(ast_func_id, xpath_type_node_set, args[0]);
 
                break;
 
            case 'l':
                if (name == PUGIXML_TEXT("last") && argc == 0)
                    return alloc_node(ast_func_last, xpath_type_number);
                else if (name == PUGIXML_TEXT("lang") && argc == 1)
                    return alloc_node(ast_func_lang, xpath_type_boolean, args[0]);
                else if (name == PUGIXML_TEXT("local-name") && argc <= 1)
                {
                    if (argc == 1 && args[0]->rettype() != xpath_type_node_set) return error("Function has to be applied to node set");
                    return alloc_node(argc == 0 ? ast_func_local_name_0 : ast_func_local_name_1, xpath_type_string, args[0]);
                }
 
                break;
 
            case 'n':
                if (name == PUGIXML_TEXT("name") && argc <= 1)
                {
                    if (argc == 1 && args[0]->rettype() != xpath_type_node_set) return error("Function has to be applied to node set");
                    return alloc_node(argc == 0 ? ast_func_name_0 : ast_func_name_1, xpath_type_string, args[0]);
                }
                else if (name == PUGIXML_TEXT("namespace-uri") && argc <= 1)
                {
                    if (argc == 1 && args[0]->rettype() != xpath_type_node_set) return error("Function has to be applied to node set");
                    return alloc_node(argc == 0 ? ast_func_namespace_uri_0 : ast_func_namespace_uri_1, xpath_type_string, args[0]);
                }
                else if (name == PUGIXML_TEXT("normalize-space") && argc <= 1)
                    return alloc_node(argc == 0 ? ast_func_normalize_space_0 : ast_func_normalize_space_1, xpath_type_string, args[0], args[1]);
                else if (name == PUGIXML_TEXT("not") && argc == 1)
                    return alloc_node(ast_func_not, xpath_type_boolean, args[0]);
                else if (name == PUGIXML_TEXT("number") && argc <= 1)
                    return alloc_node(argc == 0 ? ast_func_number_0 : ast_func_number_1, xpath_type_number, args[0]);
 
                break;
 
            case 'p':
                if (name == PUGIXML_TEXT("position") && argc == 0)
                    return alloc_node(ast_func_position, xpath_type_number);
 
                break;
 
            case 'r':
                if (name == PUGIXML_TEXT("round") && argc == 1)
                    return alloc_node(ast_func_round, xpath_type_number, args[0]);
 
                break;
 
            case 's':
                if (name == PUGIXML_TEXT("string") && argc <= 1)
                    return alloc_node(argc == 0 ? ast_func_string_0 : ast_func_string_1, xpath_type_string, args[0]);
                else if (name == PUGIXML_TEXT("string-length") && argc <= 1)
                    return alloc_node(argc == 0 ? ast_func_string_length_0 : ast_func_string_length_1, xpath_type_number, args[0]);
                else if (name == PUGIXML_TEXT("starts-with") && argc == 2)
                    return alloc_node(ast_func_starts_with, xpath_type_boolean, args[0], args[1]);
                else if (name == PUGIXML_TEXT("substring-before") && argc == 2)
                    return alloc_node(ast_func_substring_before, xpath_type_string, args[0], args[1]);
                else if (name == PUGIXML_TEXT("substring-after") && argc == 2)
                    return alloc_node(ast_func_substring_after, xpath_type_string, args[0], args[1]);
                else if (name == PUGIXML_TEXT("substring") && (argc == 2 || argc == 3))
                    return alloc_node(argc == 2 ? ast_func_substring_2 : ast_func_substring_3, xpath_type_string, args[0], args[1]);
                else if (name == PUGIXML_TEXT("sum") && argc == 1)
                {
                    if (args[0]->rettype() != xpath_type_node_set) return error("Function has to be applied to node set");
                    return alloc_node(ast_func_sum, xpath_type_number, args[0]);
                }
 
                break;
 
            case 't':
                if (name == PUGIXML_TEXT("translate") && argc == 3)
                    return alloc_node(ast_func_translate, xpath_type_string, args[0], args[1]);
                else if (name == PUGIXML_TEXT("true") && argc == 0)
                    return alloc_node(ast_func_true, xpath_type_boolean);
 
                break;
 
            default:
                break;
            }
 
            return error("Unrecognized function or wrong parameter count");
        }
 
        axis_t parse_axis_name(const xpath_lexer_string& name, bool& specified)
        {
            specified = true;
 
            switch (name.begin[0])
            {
            case 'a':
                if (name == PUGIXML_TEXT("ancestor"))
                    return axis_ancestor;
                else if (name == PUGIXML_TEXT("ancestor-or-self"))
                    return axis_ancestor_or_self;
                else if (name == PUGIXML_TEXT("attribute"))
                    return axis_attribute;
 
                break;
 
            case 'c':
                if (name == PUGIXML_TEXT("child"))
                    return axis_child;
 
                break;
 
            case 'd':
                if (name == PUGIXML_TEXT("descendant"))
                    return axis_descendant;
                else if (name == PUGIXML_TEXT("descendant-or-self"))
                    return axis_descendant_or_self;
 
                break;
 
            case 'f':
                if (name == PUGIXML_TEXT("following"))
                    return axis_following;
                else if (name == PUGIXML_TEXT("following-sibling"))
                    return axis_following_sibling;
 
                break;
 
            case 'n':
                if (name == PUGIXML_TEXT("namespace"))
                    return axis_namespace;
 
                break;
 
            case 'p':
                if (name == PUGIXML_TEXT("parent"))
                    return axis_parent;
                else if (name == PUGIXML_TEXT("preceding"))
                    return axis_preceding;
                else if (name == PUGIXML_TEXT("preceding-sibling"))
                    return axis_preceding_sibling;
 
                break;
 
            case 's':
                if (name == PUGIXML_TEXT("self"))
                    return axis_self;
 
                break;
 
            default:
                break;
            }
 
            specified = false;
            return axis_child;
        }
 
        nodetest_t parse_node_test_type(const xpath_lexer_string& name)
        {
            switch (name.begin[0])
            {
            case 'c':
                if (name == PUGIXML_TEXT("comment"))
                    return nodetest_type_comment;
 
                break;
 
            case 'n':
                if (name == PUGIXML_TEXT("node"))
                    return nodetest_type_node;
 
                break;
 
            case 'p':
                if (name == PUGIXML_TEXT("processing-instruction"))
                    return nodetest_type_pi;
 
                break;
 
            case 't':
                if (name == PUGIXML_TEXT("text"))
                    return nodetest_type_text;
 
                break;
 
            default:
                break;
            }
 
            return nodetest_none;
        }
 
        // PrimaryExpr ::= VariableReference | '(' Expr ')' | Literal | Number | FunctionCall
        xpath_ast_node* parse_primary_expression()
        {
            switch (_lexer.current())
            {
            case lex_var_ref:
            {
                xpath_lexer_string name = _lexer.contents();
 
                if (!_variables)
                    return error("Unknown variable: variable set is not provided");
 
                xpath_variable* var = 0;
                if (!get_variable_scratch(_scratch, _variables, name.begin, name.end, &var))
                    return error_oom();
 
                if (!var)
                    return error("Unknown variable: variable set does not contain the given name");
 
                _lexer.next();
 
                return alloc_node(ast_variable, var->type(), var);
            }
 
            case lex_open_brace:
            {
                _lexer.next();
 
                xpath_ast_node* n = parse_expression();
                if (!n) return 0;
 
                if (_lexer.current() != lex_close_brace)
                    return error("Expected ')' to match an opening '('");
 
                _lexer.next();
 
                return n;
            }
 
            case lex_quoted_string:
            {
                const char_t* value = alloc_string(_lexer.contents());
                if (!value) return 0;
 
                _lexer.next();
 
                return alloc_node(ast_string_constant, xpath_type_string, value);
            }
 
            case lex_number:
            {
                double value = 0;
 
                if (!convert_string_to_number_scratch(_scratch, _lexer.contents().begin, _lexer.contents().end, &value))
                    return error_oom();
 
                _lexer.next();
 
                return alloc_node(ast_number_constant, xpath_type_number, value);
            }
 
            case lex_string:
            {
                xpath_ast_node* args[2] = {0};
                size_t argc = 0;
 
                xpath_lexer_string function = _lexer.contents();
                _lexer.next();
 
                xpath_ast_node* last_arg = 0;
 
                if (_lexer.current() != lex_open_brace)
                    return error("Unrecognized function call");
                _lexer.next();
 
                while (_lexer.current() != lex_close_brace)
                {
                    if (argc > 0)
                    {
                        if (_lexer.current() != lex_comma)
                            return error("No comma between function arguments");
                        _lexer.next();
                    }
 
                    xpath_ast_node* n = parse_expression();
                    if (!n) return 0;
 
                    if (argc < 2) args[argc] = n;
                    else last_arg->set_next(n);
 
                    argc++;
                    last_arg = n;
                }
 
                _lexer.next();
 
                return parse_function(function, argc, args);
            }
 
            default:
                return error("Unrecognizable primary expression");
            }
        }
 
        // FilterExpr ::= PrimaryExpr | FilterExpr Predicate
        // Predicate ::= '[' PredicateExpr ']'
        // PredicateExpr ::= Expr
        xpath_ast_node* parse_filter_expression()
        {
            xpath_ast_node* n = parse_primary_expression();
            if (!n) return 0;
 
            while (_lexer.current() == lex_open_square_brace)
            {
                _lexer.next();
 
                if (n->rettype() != xpath_type_node_set)
                    return error("Predicate has to be applied to node set");
 
                xpath_ast_node* expr = parse_expression();
                if (!expr) return 0;
 
                n = alloc_node(ast_filter, n, expr, predicate_default);
                if (!n) return 0;
 
                if (_lexer.current() != lex_close_square_brace)
                    return error("Expected ']' to match an opening '['");
 
                _lexer.next();
            }
 
            return n;
        }
 
        // Step ::= AxisSpecifier NodeTest Predicate* | AbbreviatedStep
        // AxisSpecifier ::= AxisName '::' | '@'?
        // NodeTest ::= NameTest | NodeType '(' ')' | 'processing-instruction' '(' Literal ')'
        // NameTest ::= '*' | NCName ':' '*' | QName
        // AbbreviatedStep ::= '.' | '..'
        xpath_ast_node* parse_step(xpath_ast_node* set)
        {
            if (set && set->rettype() != xpath_type_node_set)
                return error("Step has to be applied to node set");
 
            bool axis_specified = false;
            axis_t axis = axis_child; // implied child axis
 
            if (_lexer.current() == lex_axis_attribute)
            {
                axis = axis_attribute;
                axis_specified = true;
 
                _lexer.next();
            }
            else if (_lexer.current() == lex_dot)
            {
                _lexer.next();
 
                if (_lexer.current() == lex_open_square_brace)
                    return error("Predicates are not allowed after an abbreviated step");
 
                return alloc_node(ast_step, set, axis_self, nodetest_type_node, 0);
            }
            else if (_lexer.current() == lex_double_dot)
            {
                _lexer.next();
 
                if (_lexer.current() == lex_open_square_brace)
                    return error("Predicates are not allowed after an abbreviated step");
 
                return alloc_node(ast_step, set, axis_parent, nodetest_type_node, 0);
            }
 
            nodetest_t nt_type = nodetest_none;
            xpath_lexer_string nt_name;
 
            if (_lexer.current() == lex_string)
            {
                // node name test
                nt_name = _lexer.contents();
                _lexer.next();
 
                // was it an axis name?
                if (_lexer.current() == lex_double_colon)
                {
                    // parse axis name
                    if (axis_specified)
                        return error("Two axis specifiers in one step");
 
                    axis = parse_axis_name(nt_name, axis_specified);
 
                    if (!axis_specified)
                        return error("Unknown axis");
 
                    // read actual node test
                    _lexer.next();
 
                    if (_lexer.current() == lex_multiply)
                    {
                        nt_type = nodetest_all;
                        nt_name = xpath_lexer_string();
                        _lexer.next();
                    }
                    else if (_lexer.current() == lex_string)
                    {
                        nt_name = _lexer.contents();
                        _lexer.next();
                    }
                    else
                    {
                        return error("Unrecognized node test");
                    }
                }
 
                if (nt_type == nodetest_none)
                {
                    // node type test or processing-instruction
                    if (_lexer.current() == lex_open_brace)
                    {
                        _lexer.next();
 
                        if (_lexer.current() == lex_close_brace)
                        {
                            _lexer.next();
 
                            nt_type = parse_node_test_type(nt_name);
 
                            if (nt_type == nodetest_none)
                                return error("Unrecognized node type");
 
                            nt_name = xpath_lexer_string();
                        }
                        else if (nt_name == PUGIXML_TEXT("processing-instruction"))
                        {
                            if (_lexer.current() != lex_quoted_string)
                                return error("Only literals are allowed as arguments to processing-instruction()");
 
                            nt_type = nodetest_pi;
                            nt_name = _lexer.contents();
                            _lexer.next();
 
                            if (_lexer.current() != lex_close_brace)
                                return error("Unmatched brace near processing-instruction()");
                            _lexer.next();
                        }
                        else
                        {
                            return error("Unmatched brace near node type test");
                        }
                    }
                    // QName or NCName:*
                    else
                    {
                        if (nt_name.end - nt_name.begin > 2 && nt_name.end[-2] == ':' && nt_name.end[-1] == '*') // NCName:*
                        {
                            nt_name.end--; // erase *
 
                            nt_type = nodetest_all_in_namespace;
                        }
                        else
                        {
                            nt_type = nodetest_name;
                        }
                    }
                }
            }
            else if (_lexer.current() == lex_multiply)
            {
                nt_type = nodetest_all;
                _lexer.next();
            }
            else
            {
                return error("Unrecognized node test");
            }
 
            const char_t* nt_name_copy = alloc_string(nt_name);
            if (!nt_name_copy) return 0;
 
            xpath_ast_node* n = alloc_node(ast_step, set, axis, nt_type, nt_name_copy);
            if (!n) return 0;
 
            xpath_ast_node* last = 0;
 
            while (_lexer.current() == lex_open_square_brace)
            {
                _lexer.next();
 
                xpath_ast_node* expr = parse_expression();
                if (!expr) return 0;
 
                xpath_ast_node* pred = alloc_node(ast_predicate, 0, expr, predicate_default);
                if (!pred) return 0;
 
                if (_lexer.current() != lex_close_square_brace)
                    return error("Expected ']' to match an opening '['");
                _lexer.next();
 
                if (last) last->set_next(pred);
                else n->set_right(pred);
 
                last = pred;
            }
 
            return n;
        }
 
        // RelativeLocationPath ::= Step | RelativeLocationPath '/' Step | RelativeLocationPath '//' Step
        xpath_ast_node* parse_relative_location_path(xpath_ast_node* set)
        {
            xpath_ast_node* n = parse_step(set);
            if (!n) return 0;
 
            while (_lexer.current() == lex_slash || _lexer.current() == lex_double_slash)
            {
                lexeme_t l = _lexer.current();
                _lexer.next();
 
                if (l == lex_double_slash)
                {
                    n = alloc_node(ast_step, n, axis_descendant_or_self, nodetest_type_node, 0);
                    if (!n) return 0;
                }
 
                n = parse_step(n);
                if (!n) return 0;
            }
 
            return n;
        }
 
        // LocationPath ::= RelativeLocationPath | AbsoluteLocationPath
        // AbsoluteLocationPath ::= '/' RelativeLocationPath? | '//' RelativeLocationPath
        xpath_ast_node* parse_location_path()
        {
            if (_lexer.current() == lex_slash)
            {
                _lexer.next();
 
                xpath_ast_node* n = alloc_node(ast_step_root, xpath_type_node_set);
                if (!n) return 0;
 
                // relative location path can start from axis_attribute, dot, double_dot, multiply and string lexemes; any other lexeme means standalone root path
                lexeme_t l = _lexer.current();
 
                if (l == lex_string || l == lex_axis_attribute || l == lex_dot || l == lex_double_dot || l == lex_multiply)
                    return parse_relative_location_path(n);
                else
                    return n;
            }
            else if (_lexer.current() == lex_double_slash)
            {
                _lexer.next();
 
                xpath_ast_node* n = alloc_node(ast_step_root, xpath_type_node_set);
                if (!n) return 0;
 
                n = alloc_node(ast_step, n, axis_descendant_or_self, nodetest_type_node, 0);
                if (!n) return 0;
 
                return parse_relative_location_path(n);
            }
 
            // else clause moved outside of if because of bogus warning 'control may reach end of non-void function being inlined' in gcc 4.0.1
            return parse_relative_location_path(0);
        }
 
        // PathExpr ::= LocationPath
        //              | FilterExpr
        //              | FilterExpr '/' RelativeLocationPath
        //              | FilterExpr '//' RelativeLocationPath
        // UnionExpr ::= PathExpr | UnionExpr '|' PathExpr
        // UnaryExpr ::= UnionExpr | '-' UnaryExpr
        xpath_ast_node* parse_path_or_unary_expression()
        {
            // Clarification.
            // PathExpr begins with either LocationPath or FilterExpr.
            // FilterExpr begins with PrimaryExpr
            // PrimaryExpr begins with '$' in case of it being a variable reference,
            // '(' in case of it being an expression, string literal, number constant or
            // function call.
            if (_lexer.current() == lex_var_ref || _lexer.current() == lex_open_brace ||
                _lexer.current() == lex_quoted_string || _lexer.current() == lex_number ||
                _lexer.current() == lex_string)
            {
                if (_lexer.current() == lex_string)
                {
                    // This is either a function call, or not - if not, we shall proceed with location path
                    const char_t* state = _lexer.state();
 
                    while (PUGI__IS_CHARTYPE(*state, ct_space)) ++state;
 
                    if (*state != '(')
                        return parse_location_path();
 
                    // This looks like a function call; however this still can be a node-test. Check it.
                    if (parse_node_test_type(_lexer.contents()) != nodetest_none)
                        return parse_location_path();
                }
 
                xpath_ast_node* n = parse_filter_expression();
                if (!n) return 0;
 
                if (_lexer.current() == lex_slash || _lexer.current() == lex_double_slash)
                {
                    lexeme_t l = _lexer.current();
                    _lexer.next();
 
                    if (l == lex_double_slash)
                    {
                        if (n->rettype() != xpath_type_node_set)
                            return error("Step has to be applied to node set");
 
                        n = alloc_node(ast_step, n, axis_descendant_or_self, nodetest_type_node, 0);
                        if (!n) return 0;
                    }
 
                    // select from location path
                    return parse_relative_location_path(n);
                }
 
                return n;
            }
            else if (_lexer.current() == lex_minus)
            {
                _lexer.next();
 
                // precedence 7+ - only parses union expressions
                xpath_ast_node* n = parse_expression(7);
                if (!n) return 0;
 
                return alloc_node(ast_op_negate, xpath_type_number, n);
            }
            else
            {
                return parse_location_path();
            }
        }
 
        struct binary_op_t
        {
            ast_type_t asttype;
            xpath_value_type rettype;
            int precedence;
 
            binary_op_t(): asttype(ast_unknown), rettype(xpath_type_none), precedence(0)
            {
            }
 
            binary_op_t(ast_type_t asttype_, xpath_value_type rettype_, int precedence_): asttype(asttype_), rettype(rettype_), precedence(precedence_)
            {
            }
 
            static binary_op_t parse(xpath_lexer& lexer)
            {
                switch (lexer.current())
                {
                case lex_string:
                    if (lexer.contents() == PUGIXML_TEXT("or"))
                        return binary_op_t(ast_op_or, xpath_type_boolean, 1);
                    else if (lexer.contents() == PUGIXML_TEXT("and"))
                        return binary_op_t(ast_op_and, xpath_type_boolean, 2);
                    else if (lexer.contents() == PUGIXML_TEXT("div"))
                        return binary_op_t(ast_op_divide, xpath_type_number, 6);
                    else if (lexer.contents() == PUGIXML_TEXT("mod"))
                        return binary_op_t(ast_op_mod, xpath_type_number, 6);
                    else
                        return binary_op_t();
 
                case lex_equal:
                    return binary_op_t(ast_op_equal, xpath_type_boolean, 3);
 
                case lex_not_equal:
                    return binary_op_t(ast_op_not_equal, xpath_type_boolean, 3);
 
                case lex_less:
                    return binary_op_t(ast_op_less, xpath_type_boolean, 4);
 
                case lex_greater:
                    return binary_op_t(ast_op_greater, xpath_type_boolean, 4);
 
                case lex_less_or_equal:
                    return binary_op_t(ast_op_less_or_equal, xpath_type_boolean, 4);
 
                case lex_greater_or_equal:
                    return binary_op_t(ast_op_greater_or_equal, xpath_type_boolean, 4);
 
                case lex_plus:
                    return binary_op_t(ast_op_add, xpath_type_number, 5);
 
                case lex_minus:
                    return binary_op_t(ast_op_subtract, xpath_type_number, 5);
 
                case lex_multiply:
                    return binary_op_t(ast_op_multiply, xpath_type_number, 6);
 
                case lex_union:
                    return binary_op_t(ast_op_union, xpath_type_node_set, 7);
 
                default:
                    return binary_op_t();
                }
            }
        };
 
        xpath_ast_node* parse_expression_rec(xpath_ast_node* lhs, int limit)
        {
            binary_op_t op = binary_op_t::parse(_lexer);
 
            while (op.asttype != ast_unknown && op.precedence >= limit)
            {
                _lexer.next();
 
                xpath_ast_node* rhs = parse_path_or_unary_expression();
                if (!rhs) return 0;
 
                binary_op_t nextop = binary_op_t::parse(_lexer);
 
                while (nextop.asttype != ast_unknown && nextop.precedence > op.precedence)
                {
                    rhs = parse_expression_rec(rhs, nextop.precedence);
                    if (!rhs) return 0;
 
                    nextop = binary_op_t::parse(_lexer);
                }
 
                if (op.asttype == ast_op_union && (lhs->rettype() != xpath_type_node_set || rhs->rettype() != xpath_type_node_set))
                    return error("Union operator has to be applied to node sets");
 
                lhs = alloc_node(op.asttype, op.rettype, lhs, rhs);
                if (!lhs) return 0;
 
                op = binary_op_t::parse(_lexer);
            }
 
            return lhs;
        }
 
        // Expr ::= OrExpr
        // OrExpr ::= AndExpr | OrExpr 'or' AndExpr
        // AndExpr ::= EqualityExpr | AndExpr 'and' EqualityExpr
        // EqualityExpr ::= RelationalExpr
        //                  | EqualityExpr '=' RelationalExpr
        //                  | EqualityExpr '!=' RelationalExpr
        // RelationalExpr ::= AdditiveExpr
        //                    | RelationalExpr '<' AdditiveExpr
        //                    | RelationalExpr '>' AdditiveExpr
        //                    | RelationalExpr '<=' AdditiveExpr
        //                    | RelationalExpr '>=' AdditiveExpr
        // AdditiveExpr ::= MultiplicativeExpr
        //                  | AdditiveExpr '+' MultiplicativeExpr
        //                  | AdditiveExpr '-' MultiplicativeExpr
        // MultiplicativeExpr ::= UnaryExpr
        //                        | MultiplicativeExpr '*' UnaryExpr
        //                        | MultiplicativeExpr 'div' UnaryExpr
        //                        | MultiplicativeExpr 'mod' UnaryExpr
        xpath_ast_node* parse_expression(int limit = 0)
        {
            xpath_ast_node* n = parse_path_or_unary_expression();
            if (!n) return 0;
 
            return parse_expression_rec(n, limit);
        }
 
        xpath_parser(const char_t* query, xpath_variable_set* variables, xpath_allocator* alloc, xpath_parse_result* result): _alloc(alloc), _lexer(query), _query(query), _variables(variables), _result(result)
        {
        }
 
        xpath_ast_node* parse()
        {
            xpath_ast_node* n = parse_expression();
            if (!n) return 0;
 
            // check if there are unparsed tokens left
            if (_lexer.current() != lex_eof)
                return error("Incorrect query");
 
            return n;
        }
 
        static xpath_ast_node* parse(const char_t* query, xpath_variable_set* variables, xpath_allocator* alloc, xpath_parse_result* result)
        {
            xpath_parser parser(query, variables, alloc, result);
 
            return parser.parse();
        }
    };
 
    struct xpath_query_impl
    {
        static xpath_query_impl* create()
        {
            void* memory = xml_memory::allocate(sizeof(xpath_query_impl));
            if (!memory) return 0;
 
            return new (memory) xpath_query_impl();
        }
 
        static void destroy(xpath_query_impl* impl)
        {
            // free all allocated pages
            impl->alloc.release();
 
            // free allocator memory (with the first page)
            xml_memory::deallocate(impl);
        }
 
        xpath_query_impl(): root(0), alloc(&block, &oom), oom(false)
        {
            block.next = 0;
            block.capacity = sizeof(block.data);
        }
 
        xpath_ast_node* root;
        xpath_allocator alloc;
        xpath_memory_block block;
        bool oom;
    };
 
    PUGI__FN impl::xpath_ast_node* evaluate_node_set_prepare(xpath_query_impl* impl)
    {
        if (!impl) return 0;
 
        if (impl->root->rettype() != xpath_type_node_set)
        {
        #ifdef PUGIXML_NO_EXCEPTIONS
            return 0;
        #else
            xpath_parse_result res;
            res.error = "Expression does not evaluate to node set";
 
            throw xpath_exception(res);
        #endif
        }
 
        return impl->root;
    }
PUGI__NS_END
 
namespace pugi
{
#ifndef PUGIXML_NO_EXCEPTIONS
    PUGI__FN xpath_exception::xpath_exception(const xpath_parse_result& result_): _result(result_)
    {
        assert(_result.error);
    }
 
    PUGI__FN const char* xpath_exception::what() const noexcept
    {
        return _result.error;
    }
 
    PUGI__FN const xpath_parse_result& xpath_exception::result() const
    {
        return _result;
    }
#endif
 
    PUGI__FN xpath_node::xpath_node()
    {
    }
 
    PUGI__FN xpath_node::xpath_node(const xml_node& node_): _node(node_)
    {
    }
 
    PUGI__FN xpath_node::xpath_node(const xml_attribute& attribute_, const xml_node& parent_): _node(attribute_ ? parent_ : xml_node()), _attribute(attribute_)
    {
    }
 
    PUGI__FN xml_node xpath_node::node() const
    {
        return _attribute ? xml_node() : _node;
    }
 
    PUGI__FN xml_attribute xpath_node::attribute() const
    {
        return _attribute;
    }
 
    PUGI__FN xml_node xpath_node::parent() const
    {
        return _attribute ? _node : _node.parent();
    }
 
    PUGI__FN static void unspecified_bool_xpath_node(xpath_node***)
    {
    }
 
    PUGI__FN xpath_node::operator xpath_node::unspecified_bool_type() const
    {
        return (_node || _attribute) ? unspecified_bool_xpath_node : 0;
    }
 
    PUGI__FN bool xpath_node::operator!() const
    {
        return !(_node || _attribute);
    }
 
    PUGI__FN bool xpath_node::operator==(const xpath_node& n) const
    {
        return _node == n._node && _attribute == n._attribute;
    }
 
    PUGI__FN bool xpath_node::operator!=(const xpath_node& n) const
    {
        return _node != n._node || _attribute != n._attribute;
    }
 
#ifdef __BORLANDC__
    PUGI__FN bool operator&&(const xpath_node& lhs, bool rhs)
    {
        return (bool)lhs && rhs;
    }
 
    PUGI__FN bool operator||(const xpath_node& lhs, bool rhs)
    {
        return (bool)lhs || rhs;
    }
#endif
 
    PUGI__FN void xpath_node_set::_assign(const_iterator begin_, const_iterator end_, type_t type_)
    {
        assert(begin_ <= end_);
 
        size_t size_ = static_cast<size_t>(end_ - begin_);
 
        if (size_ <= 1)
        {
            // deallocate old buffer
            if (_begin != &_storage) impl::xml_memory::deallocate(_begin);
 
            // use internal buffer
            if (begin_ != end_) _storage = *begin_;
 
            _begin = &_storage;
            _end = &_storage + size_;
            _type = type_;
        }
        else
        {
            // make heap copy
            xpath_node* storage = static_cast<xpath_node*>(impl::xml_memory::allocate(size_ * sizeof(xpath_node)));
 
            if (!storage)
            {
            #ifdef PUGIXML_NO_EXCEPTIONS
                return;
            #else
                throw std::bad_alloc();
            #endif
            }
 
            memcpy(storage, begin_, size_ * sizeof(xpath_node));
 
            // deallocate old buffer
            if (_begin != &_storage) impl::xml_memory::deallocate(_begin);
 
            // finalize
            _begin = storage;
            _end = storage + size_;
            _type = type_;
        }
    }
 
#ifdef PUGIXML_HAS_MOVE
    PUGI__FN void xpath_node_set::_move(xpath_node_set& rhs) PUGIXML_NOEXCEPT
    {
        _type = rhs._type;
        _storage = rhs._storage;
        _begin = (rhs._begin == &rhs._storage) ? &_storage : rhs._begin;
        _end = _begin + (rhs._end - rhs._begin);
 
        rhs._type = type_unsorted;
        rhs._begin = &rhs._storage;
        rhs._end = rhs._begin;
    }
#endif
 
    PUGI__FN xpath_node_set::xpath_node_set(): _type(type_unsorted), _begin(&_storage), _end(&_storage)
    {
    }
 
    PUGI__FN xpath_node_set::xpath_node_set(const_iterator begin_, const_iterator end_, type_t type_): _type(type_unsorted), _begin(&_storage), _end(&_storage)
    {
        _assign(begin_, end_, type_);
    }
 
    PUGI__FN xpath_node_set::~xpath_node_set()
    {
        if (_begin != &_storage)
            impl::xml_memory::deallocate(_begin);
    }
 
    PUGI__FN xpath_node_set::xpath_node_set(const xpath_node_set& ns): _type(type_unsorted), _begin(&_storage), _end(&_storage)
    {
        _assign(ns._begin, ns._end, ns._type);
    }
 
    PUGI__FN xpath_node_set& xpath_node_set::operator=(const xpath_node_set& ns)
    {
        if (this == &ns) return *this;
 
        _assign(ns._begin, ns._end, ns._type);
 
        return *this;
    }
 
#ifdef PUGIXML_HAS_MOVE
    PUGI__FN xpath_node_set::xpath_node_set(xpath_node_set&& rhs) PUGIXML_NOEXCEPT: _type(type_unsorted), _begin(&_storage), _end(&_storage)
    {
        _move(rhs);
    }
 
    PUGI__FN xpath_node_set& xpath_node_set::operator=(xpath_node_set&& rhs) PUGIXML_NOEXCEPT
    {
        if (this == &rhs) return *this;
 
        if (_begin != &_storage)
            impl::xml_memory::deallocate(_begin);
 
        _move(rhs);
 
        return *this;
    }
#endif
 
    PUGI__FN xpath_node_set::type_t xpath_node_set::type() const
    {
        return _type;
    }
 
    PUGI__FN size_t xpath_node_set::size() const
    {
        return _end - _begin;
    }
 
    PUGI__FN bool xpath_node_set::empty() const
    {
        return _begin == _end;
    }
 
    PUGI__FN const xpath_node& xpath_node_set::operator[](size_t index) const
    {
        assert(index < size());
        return _begin[index];
    }
 
    PUGI__FN xpath_node_set::const_iterator xpath_node_set::begin() const
    {
        return _begin;
    }
 
    PUGI__FN xpath_node_set::const_iterator xpath_node_set::end() const
    {
        return _end;
    }
 
    PUGI__FN void xpath_node_set::sort(bool reverse)
    {
        _type = impl::xpath_sort(_begin, _end, _type, reverse);
    }
 
    PUGI__FN xpath_node xpath_node_set::first() const
    {
        return impl::xpath_first(_begin, _end, _type);
    }
 
    PUGI__FN xpath_parse_result::xpath_parse_result(): error("Internal error"), offset(0)
    {
    }
 
    PUGI__FN xpath_parse_result::operator bool() const
    {
        return error == 0;
    }
 
    PUGI__FN const char* xpath_parse_result::description() const
    {
        return error ? error : "No error";
    }
 
    PUGI__FN xpath_variable::xpath_variable(xpath_value_type type_): _type(type_), _next(0)
    {
    }
 
    PUGI__FN const char_t* xpath_variable::name() const
    {
        switch (_type)
        {
        case xpath_type_node_set:
            return static_cast<const impl::xpath_variable_node_set*>(this)->name;
 
        case xpath_type_number:
            return static_cast<const impl::xpath_variable_number*>(this)->name;
 
        case xpath_type_string:
            return static_cast<const impl::xpath_variable_string*>(this)->name;
 
        case xpath_type_boolean:
            return static_cast<const impl::xpath_variable_boolean*>(this)->name;
 
        default:
            assert(false && "Invalid variable type"); // unreachable
            return 0;
        }
    }
 
    PUGI__FN xpath_value_type xpath_variable::type() const
    {
        return _type;
    }
 
    PUGI__FN bool xpath_variable::get_boolean() const
    {
        return (_type == xpath_type_boolean) ? static_cast<const impl::xpath_variable_boolean*>(this)->value : false;
    }
 
    PUGI__FN double xpath_variable::get_number() const
    {
        return (_type == xpath_type_number) ? static_cast<const impl::xpath_variable_number*>(this)->value : impl::gen_nan();
    }
 
    PUGI__FN const char_t* xpath_variable::get_string() const
    {
        const char_t* value = (_type == xpath_type_string) ? static_cast<const impl::xpath_variable_string*>(this)->value : 0;
        return value ? value : PUGIXML_TEXT("");
    }
 
    PUGI__FN const xpath_node_set& xpath_variable::get_node_set() const
    {
        return (_type == xpath_type_node_set) ? static_cast<const impl::xpath_variable_node_set*>(this)->value : impl::dummy_node_set;
    }
 
    PUGI__FN bool xpath_variable::set(bool value)
    {
        if (_type != xpath_type_boolean) return false;
 
        static_cast<impl::xpath_variable_boolean*>(this)->value = value;
        return true;
    }
 
    PUGI__FN bool xpath_variable::set(double value)
    {
        if (_type != xpath_type_number) return false;
 
        static_cast<impl::xpath_variable_number*>(this)->value = value;
        return true;
    }
 
    PUGI__FN bool xpath_variable::set(const char_t* value)
    {
        if (_type != xpath_type_string) return false;
 
        impl::xpath_variable_string* var = static_cast<impl::xpath_variable_string*>(this);
 
        // duplicate string
        size_t size = (impl::strlength(value) + 1) * sizeof(char_t);
 
        char_t* copy = static_cast<char_t*>(impl::xml_memory::allocate(size));
        if (!copy) return false;
 
        memcpy(copy, value, size);
 
        // replace old string
        if (var->value) impl::xml_memory::deallocate(var->value);
        var->value = copy;
 
        return true;
    }
 
    PUGI__FN bool xpath_variable::set(const xpath_node_set& value)
    {
        if (_type != xpath_type_node_set) return false;
 
        static_cast<impl::xpath_variable_node_set*>(this)->value = value;
        return true;
    }
 
    PUGI__FN xpath_variable_set::xpath_variable_set()
    {
        for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
            _data[i] = 0;
    }
 
    PUGI__FN xpath_variable_set::~xpath_variable_set()
    {
        for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
            _destroy(_data[i]);
    }
 
    PUGI__FN xpath_variable_set::xpath_variable_set(const xpath_variable_set& rhs)
    {
        for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
            _data[i] = 0;
 
        _assign(rhs);
    }
 
    PUGI__FN xpath_variable_set& xpath_variable_set::operator=(const xpath_variable_set& rhs)
    {
        if (this == &rhs) return *this;
 
        _assign(rhs);
 
        return *this;
    }
 
#ifdef PUGIXML_HAS_MOVE
    PUGI__FN xpath_variable_set::xpath_variable_set(xpath_variable_set&& rhs) PUGIXML_NOEXCEPT
    {
        for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
        {
            _data[i] = rhs._data[i];
            rhs._data[i] = 0;
        }
    }
 
    PUGI__FN xpath_variable_set& xpath_variable_set::operator=(xpath_variable_set&& rhs) PUGIXML_NOEXCEPT
    {
        for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
        {
            _destroy(_data[i]);
 
            _data[i] = rhs._data[i];
            rhs._data[i] = 0;
        }
 
        return *this;
    }
#endif
 
    PUGI__FN void xpath_variable_set::_assign(const xpath_variable_set& rhs)
    {
        xpath_variable_set temp;
 
        for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
            if (rhs._data[i] && !_clone(rhs._data[i], &temp._data[i]))
                return;
 
        _swap(temp);
    }
 
    PUGI__FN void xpath_variable_set::_swap(xpath_variable_set& rhs)
    {
        for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
        {
            xpath_variable* chain = _data[i];
 
            _data[i] = rhs._data[i];
            rhs._data[i] = chain;
        }
    }
 
    PUGI__FN xpath_variable* xpath_variable_set::_find(const char_t* name) const
    {
        const size_t hash_size = sizeof(_data) / sizeof(_data[0]);
        size_t hash = impl::hash_string(name) % hash_size;
 
        // look for existing variable
        for (xpath_variable* var = _data[hash]; var; var = var->_next)
            if (impl::strequal(var->name(), name))
                return var;
 
        return 0;
    }
 
    PUGI__FN bool xpath_variable_set::_clone(xpath_variable* var, xpath_variable** out_result)
    {
        xpath_variable* last = 0;
 
        while (var)
        {
            // allocate storage for new variable
            xpath_variable* nvar = impl::new_xpath_variable(var->_type, var->name());
            if (!nvar) return false;
 
            // link the variable to the result immediately to handle failures gracefully
            if (last)
                last->_next = nvar;
            else
                *out_result = nvar;
 
            last = nvar;
 
            // copy the value; this can fail due to out-of-memory conditions
            if (!impl::copy_xpath_variable(nvar, var)) return false;
 
            var = var->_next;
        }
 
        return true;
    }
 
    PUGI__FN void xpath_variable_set::_destroy(xpath_variable* var)
    {
        while (var)
        {
            xpath_variable* next = var->_next;
 
            impl::delete_xpath_variable(var->_type, var);
 
            var = next;
        }
    }
 
    PUGI__FN xpath_variable* xpath_variable_set::add(const char_t* name, xpath_value_type type)
    {
        const size_t hash_size = sizeof(_data) / sizeof(_data[0]);
        size_t hash = impl::hash_string(name) % hash_size;
 
        // look for existing variable
        for (xpath_variable* var = _data[hash]; var; var = var->_next)
            if (impl::strequal(var->name(), name))
                return var->type() == type ? var : 0;
 
        // add new variable
        xpath_variable* result = impl::new_xpath_variable(type, name);
 
        if (result)
        {
            result->_next = _data[hash];
 
            _data[hash] = result;
        }
 
        return result;
    }
 
    PUGI__FN bool xpath_variable_set::set(const char_t* name, bool value)
    {
        xpath_variable* var = add(name, xpath_type_boolean);
        return var ? var->set(value) : false;
    }
 
    PUGI__FN bool xpath_variable_set::set(const char_t* name, double value)
    {
        xpath_variable* var = add(name, xpath_type_number);
        return var ? var->set(value) : false;
    }
 
    PUGI__FN bool xpath_variable_set::set(const char_t* name, const char_t* value)
    {
        xpath_variable* var = add(name, xpath_type_string);
        return var ? var->set(value) : false;
    }
 
    PUGI__FN bool xpath_variable_set::set(const char_t* name, const xpath_node_set& value)
    {
        xpath_variable* var = add(name, xpath_type_node_set);
        return var ? var->set(value) : false;
    }
 
    PUGI__FN xpath_variable* xpath_variable_set::get(const char_t* name)
    {
        return _find(name);
    }
 
    PUGI__FN const xpath_variable* xpath_variable_set::get(const char_t* name) const
    {
        return _find(name);
    }
 
    PUGI__FN xpath_query::xpath_query(const char_t* query, xpath_variable_set* variables): _impl(0)
    {
        impl::xpath_query_impl* qimpl = impl::xpath_query_impl::create();
 
        if (!qimpl)
        {
        #ifdef PUGIXML_NO_EXCEPTIONS
            _result.error = "Out of memory";
        #else
            throw std::bad_alloc();
        #endif
        }
        else
        {
            using impl::auto_deleter; // MSVC7 workaround
            auto_deleter<impl::xpath_query_impl> impl(qimpl, impl::xpath_query_impl::destroy);
 
            qimpl->root = impl::xpath_parser::parse(query, variables, &qimpl->alloc, &_result);
 
            if (qimpl->root)
            {
                qimpl->root->optimize(&qimpl->alloc);
 
                _impl = impl.release();
                _result.error = 0;
            }
            else
            {
            #ifdef PUGIXML_NO_EXCEPTIONS
                if (qimpl->oom) _result.error = "Out of memory";
            #else
                if (qimpl->oom) throw std::bad_alloc();
                throw xpath_exception(_result);
            #endif
            }
        }
    }
 
    PUGI__FN xpath_query::xpath_query(): _impl(0)
    {
    }
 
    PUGI__FN xpath_query::~xpath_query()
    {
        if (_impl)
            impl::xpath_query_impl::destroy(static_cast<impl::xpath_query_impl*>(_impl));
    }
 
#ifdef PUGIXML_HAS_MOVE
    PUGI__FN xpath_query::xpath_query(xpath_query&& rhs) PUGIXML_NOEXCEPT
    {
        _impl = rhs._impl;
        _result = rhs._result;
        rhs._impl = 0;
        rhs._result = xpath_parse_result();
    }
 
    PUGI__FN xpath_query& xpath_query::operator=(xpath_query&& rhs) PUGIXML_NOEXCEPT
    {
        if (this == &rhs) return *this;
 
        if (_impl)
            impl::xpath_query_impl::destroy(static_cast<impl::xpath_query_impl*>(_impl));
 
        _impl = rhs._impl;
        _result = rhs._result;
        rhs._impl = 0;
        rhs._result = xpath_parse_result();
 
        return *this;
    }
#endif
 
    PUGI__FN xpath_value_type xpath_query::return_type() const
    {
        if (!_impl) return xpath_type_none;
 
        return static_cast<impl::xpath_query_impl*>(_impl)->root->rettype();
    }
 
    PUGI__FN bool xpath_query::evaluate_boolean(const xpath_node& n) const
    {
        if (!_impl) return false;
 
        impl::xpath_context c(n, 1, 1);
        impl::xpath_stack_data sd;
 
        bool r = static_cast<impl::xpath_query_impl*>(_impl)->root->eval_boolean(c, sd.stack);
 
        if (sd.oom)
        {
        #ifdef PUGIXML_NO_EXCEPTIONS
            return false;
        #else
            throw std::bad_alloc();
        #endif
        }
 
        return r;
    }
 
    PUGI__FN double xpath_query::evaluate_number(const xpath_node& n) const
    {
        if (!_impl) return impl::gen_nan();
 
        impl::xpath_context c(n, 1, 1);
        impl::xpath_stack_data sd;
 
        double r = static_cast<impl::xpath_query_impl*>(_impl)->root->eval_number(c, sd.stack);
 
        if (sd.oom)
        {
        #ifdef PUGIXML_NO_EXCEPTIONS
            return impl::gen_nan();
        #else
            throw std::bad_alloc();
        #endif
        }
 
        return r;
    }
 
#ifndef PUGIXML_NO_STL
    PUGI__FN string_t xpath_query::evaluate_string(const xpath_node& n) const
    {
        if (!_impl) return string_t();
 
        impl::xpath_context c(n, 1, 1);
        impl::xpath_stack_data sd;
 
        impl::xpath_string r = static_cast<impl::xpath_query_impl*>(_impl)->root->eval_string(c, sd.stack);
 
        if (sd.oom)
        {
        #ifdef PUGIXML_NO_EXCEPTIONS
            return string_t();
        #else
            throw std::bad_alloc();
        #endif
        }
 
        return string_t(r.c_str(), r.length());
    }
#endif
 
    PUGI__FN size_t xpath_query::evaluate_string(char_t* buffer, size_t capacity, const xpath_node& n) const
    {
        impl::xpath_context c(n, 1, 1);
        impl::xpath_stack_data sd;
 
        impl::xpath_string r = _impl ? static_cast<impl::xpath_query_impl*>(_impl)->root->eval_string(c, sd.stack) : impl::xpath_string();
 
        if (sd.oom)
        {
        #ifdef PUGIXML_NO_EXCEPTIONS
            r = impl::xpath_string();
        #else
            throw std::bad_alloc();
        #endif
        }
 
        size_t full_size = r.length() + 1;
 
        if (capacity > 0)
        {
            size_t size = (full_size < capacity) ? full_size : capacity;
            assert(size > 0);
 
            memcpy(buffer, r.c_str(), (size - 1) * sizeof(char_t));
            buffer[size - 1] = 0;
        }
 
        return full_size;
    }
 
    PUGI__FN xpath_node_set xpath_query::evaluate_node_set(const xpath_node& n) const
    {
        impl::xpath_ast_node* root = impl::evaluate_node_set_prepare(static_cast<impl::xpath_query_impl*>(_impl));
        if (!root) return xpath_node_set();
 
        impl::xpath_context c(n, 1, 1);
        impl::xpath_stack_data sd;
 
        impl::xpath_node_set_raw r = root->eval_node_set(c, sd.stack, impl::nodeset_eval_all);
 
        if (sd.oom)
        {
        #ifdef PUGIXML_NO_EXCEPTIONS
            return xpath_node_set();
        #else
            throw std::bad_alloc();
        #endif
        }
 
        return xpath_node_set(r.begin(), r.end(), r.type());
    }
 
    PUGI__FN xpath_node xpath_query::evaluate_node(const xpath_node& n) const
    {
        impl::xpath_ast_node* root = impl::evaluate_node_set_prepare(static_cast<impl::xpath_query_impl*>(_impl));
        if (!root) return xpath_node();
 
        impl::xpath_context c(n, 1, 1);
        impl::xpath_stack_data sd;
 
        impl::xpath_node_set_raw r = root->eval_node_set(c, sd.stack, impl::nodeset_eval_first);
 
        if (sd.oom)
        {
        #ifdef PUGIXML_NO_EXCEPTIONS
            return xpath_node();
        #else
            throw std::bad_alloc();
        #endif
        }
 
        return r.first();
    }
 
    PUGI__FN const xpath_parse_result& xpath_query::result() const
    {
        return _result;
    }
 
    PUGI__FN static void unspecified_bool_xpath_query(xpath_query***)
    {
    }
 
    PUGI__FN xpath_query::operator xpath_query::unspecified_bool_type() const
    {
        return _impl ? unspecified_bool_xpath_query : 0;
    }
 
    PUGI__FN bool xpath_query::operator!() const
    {
        return !_impl;
    }
 
    PUGI__FN xpath_node xml_node::select_node(const char_t* query, xpath_variable_set* variables) const
    {
        xpath_query q(query, variables);
        return q.evaluate_node(*this);
    }
 
    PUGI__FN xpath_node xml_node::select_node(const xpath_query& query) const
    {
        return query.evaluate_node(*this);
    }
 
    PUGI__FN xpath_node_set xml_node::select_nodes(const char_t* query, xpath_variable_set* variables) const
    {
        xpath_query q(query, variables);
        return q.evaluate_node_set(*this);
    }
 
    PUGI__FN xpath_node_set xml_node::select_nodes(const xpath_query& query) const
    {
        return query.evaluate_node_set(*this);
    }
 
    PUGI__FN xpath_node xml_node::select_single_node(const char_t* query, xpath_variable_set* variables) const
    {
        xpath_query q(query, variables);
        return q.evaluate_node(*this);
    }
 
    PUGI__FN xpath_node xml_node::select_single_node(const xpath_query& query) const
    {
        return query.evaluate_node(*this);
    }
}
 
#endif
 
#ifdef __BORLANDC__
#   pragma option pop
#endif
 
// Intel C++ does not properly keep warning state for function templates,
// so popping warning state at the end of translation unit leads to warnings in the middle.
#if defined(_MSC_VER) && !defined(__INTEL_COMPILER)
#   pragma warning(pop)
#endif
 
#if defined(_MSC_VER) && defined(__c2__)
#   pragma clang diagnostic pop
#endif
 
#if defined(__clang__)
#   pragma clang diagnostic pop
#endif
 
// Undefine all local macros (makes sure we're not leaking macros in header-only mode)
#undef PUGI__NO_INLINE
#undef PUGI__UNLIKELY
#undef PUGI__STATIC_ASSERT
#undef PUGI__DMC_VOLATILE
#undef PUGI__UNSIGNED_OVERFLOW
#undef PUGI__MSVC_CRT_VERSION
#undef PUGI__SNPRINTF
#undef PUGI__NS_BEGIN
#undef PUGI__NS_END
#undef PUGI__FN
#undef PUGI__FN_NO_INLINE
#undef PUGI__GETHEADER_IMPL
#undef PUGI__GETPAGE_IMPL
#undef PUGI__GETPAGE
#undef PUGI__NODETYPE
#undef PUGI__IS_CHARTYPE_IMPL
#undef PUGI__IS_CHARTYPE
#undef PUGI__IS_CHARTYPEX
#undef PUGI__ENDSWITH
#undef PUGI__SKIPWS
#undef PUGI__OPTSET
#undef PUGI__PUSHNODE
#undef PUGI__POPNODE
#undef PUGI__SCANFOR
#undef PUGI__SCANWHILE
#undef PUGI__SCANWHILE_UNROLL
#undef PUGI__ENDSEG
#undef PUGI__THROW_ERROR
#undef PUGI__CHECK_ERROR
 
#endif
 
/**
 * Copyright (c) 2006-2018 Arseny Kapoulkine
 *
 * Permission is hereby granted, free of charge, to any person
 * obtaining a copy of this software and associated documentation
 * files (the "Software"), to deal in the Software without
 * restriction, including without limitation the rights to use,
 * copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following
 * conditions:
 *
 * The above copyright notice and this permission notice shall be
 * included in all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
 * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 * OTHER DEALINGS IN THE SOFTWARE.
 */