pugixml.cpp

		{
			// 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;
		}

		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);

		if (is_mutable)
		{
			out_buffer = static_cast<char_t*>(const_cast<void*>(contents));
			out_length = length;
		}
		else
		{
			char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
			if (!buffer) return false;

			if (contents)
				memcpy(buffer, contents, length * sizeof(char_t));
			else
				assert(length == 0);

			buffer[length] = 0;

			out_buffer = buffer;
			out_length = length + 1;
		}

		return true;
	}

#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);
	}

	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);
		size_t length = size / sizeof(char_t);

		if (is_mutable)
		{
			char_t* buffer = const_cast<char_t*>(data);

			convert_wchar_endian_swap(buffer, data, length);

			out_buffer = buffer;
			out_length = length;
		}
		else
		{
			char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
			if (!buffer) return false;

			convert_wchar_endian_swap(buffer, data, length);
			buffer[length] = 0;

			out_buffer = buffer;
			out_length = length + 1;
		}

		return true;
	}

	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)