pugixml.cpp

			{
				*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; // should not get here
		}
	}

	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; // should not get here
		}
	}

	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(const 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 += 4;
			}
			else PUGI__THROW_ERROR(status_bad_doctype, s);

			return s;
		}

		char_t* parse_doctype_ignore(char_t* s)
		{
			assert(s[0] == '<' && s[1] == '!' && s[2] == '[');
			s++;

			while (*s)
			{
				if (s[0] == '<' && s[1] == '!' && s[2] == '[')
				{
					// nested ignore section
					s = parse_doctype_ignore(s);
					if (!s) return s;
				}
				else if (s[0] == ']' && s[1] == ']' && s[2] == '>')
				{
					// ignore section end
					s += 3;

					return s;
				}
				else s++;
			}

			PUGI__THROW_ERROR(status_bad_doctype, s);
		}

		char_t* parse_doctype_group(char_t* s, char_t endch, bool toplevel)
		{
			assert((s[0] == '<' || s[0] == 0) && s[1] == '!');
			s++;

			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 = parse_doctype_group(s, endch, false);
						if (!s) return s;

						// skip >
						assert(*s == '>');
						s++;
					}
				}
				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 == '>')
				{
					return s;
				}
				else s++;
			}

			if (!toplevel || 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, true);
				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;

						char_t* name = cursor->name;
						if (!name) PUGI__THROW_ERROR(status_end_element_mismatch, s);
						
						while (PUGI__IS_CHARTYPE(*s, ct_symbol))
						{
							if (*s++ != *name++) PUGI__THROW_ERROR(status_end_element_mismatch, s);
						}

						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, s);
						}
							
						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))
					{
						PUGI__PUSHNODE(node_pcdata); // Append a new node on the tree.
						cursor->value = s; // Save the offset.

						s = strconv_pcdata(s);
								
						PUGI__POPNODE(); // Pop since this is a standalone.
						
						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)
		{
			// allocator object is a part of document object
			xml_allocator& alloc_ = *static_cast<xml_allocator*>(xmldoc);

			// 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;
	
			// create parser on stack
			xml_parser parser(alloc_);

			// 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);

			// update allocator state
			alloc_ = parser.alloc;

			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 : root->first_child;

				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;
	}

#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)
		{
			uint8_t* dest = r_u8;
			uint8_t* end = utf_decoder<utf8_writer>::decode_wchar_block(data, length, dest);

			return static_cast<size_t>(end - dest);
		}

		// convert to utf16
		if (encoding == encoding_utf16_be || encoding == encoding_utf16_le)
		{
			uint16_t* dest = r_u16;

			// convert to native utf16
			uint16_t* end = utf_decoder<utf16_writer>::decode_wchar_block(data, length, dest);

			// swap if necessary
			xml_encoding native_encoding = is_little_endian() ? encoding_utf16_le : encoding_utf16_be;