// =================================================================================================
// Copyright 2002-2008 Adobe Systems Incorporated
// All Rights Reserved.
//
// NOTICE:  Adobe permits you to use, modify, and distribute this file in accordance with the terms
// of the Adobe license agreement accompanying it.
//
// Adobe patent application tracking #P435, entitled 'Unique markers to simplify embedding data of
// one format in a file with a different format', inventors: Sean Parent, Greg Gilley.
// =================================================================================================

#include "XMP_Environment.h"	// ! This must be the first include!
#include "XMPCore_Impl.hpp"

#include "XMPMeta.hpp"
#include "XMPUtils.hpp"

#include "UnicodeInlines.incl_cpp"
#include "UnicodeConversions.hpp"
#include "ExpatAdapter.hpp"

#if XMP_DebugBuild
	#include <iostream>
#endif

using namespace std;

#if XMP_WinBuild
#ifdef _MSC_VER
	#pragma warning ( disable : 4533 )	// initialization of '...' is skipped by 'goto ...'
	#pragma warning ( disable : 4702 )	// unreachable code
	#pragma warning ( disable : 4800 )	// forcing value to bool 'true' or 'false' (performance warning)
	#pragma warning ( disable : 4996 )	// '...' was declared deprecated
#endif
#endif


// *** Use the XMP_PropIsXyz (Schema, Simple, Struct, Array, ...) macros
// *** Add debug codegen checks, e.g. that typical masking operations really work
// *** Change all uses of strcmp and strncmp to XMP_LitMatch and XMP_LitNMatch


// =================================================================================================
// Local Types and Constants
// =========================


// =================================================================================================
// Static Variables
// ================

#ifndef Trace_ParsingHackery
	#define Trace_ParsingHackery 0
#endif

static const char * kReplaceLatin1[128] =
	{

		// The 0x80..0x9F range is undefined in Latin-1, but is defined in Windows code page 1252.
		// The bytes 0x81, 0x8D, 0x8F, 0x90, and 0x9D are formally undefined by Windows 1252, but
		// their conversion API maps them to U+0081, etc. These are in XML's RestrictedChar set, so
		// we map them to a space.
		
		"\xE2\x82\xAC", " ",            "\xE2\x80\x9A", "\xC6\x92",		// 0x80 .. 0x83
		"\xE2\x80\x9E", "\xE2\x80\xA6", "\xE2\x80\xA0", "\xE2\x80\xA1",	// 0x84 .. 0x87
		"\xCB\x86",     "\xE2\x80\xB0", "\xC5\xA0",     "\xE2\x80\xB9",	// 0x88 .. 0x8B
		"\xC5\x92",     " ",            "\xC5\xBD",     " ",			// 0x8C .. 0x8F

		" ",            "\xE2\x80\x98", "\xE2\x80\x99", "\xE2\x80\x9C",	// 0x90 .. 0x93
		"\xE2\x80\x9D", "\xE2\x80\xA2", "\xE2\x80\x93", "\xE2\x80\x94",	// 0x94 .. 0x97
		"\xCB\x9C",     "\xE2\x84\xA2", "\xC5\xA1",     "\xE2\x80\xBA",	// 0x98 .. 0x9B
		"\xC5\x93",     " ",            "\xC5\xBE",     "\xC5\xB8",		// 0x9C .. 0x9F

		// These are the UTF-8 forms of the official Latin-1 characters in the range 0xA0..0xFF. Not
		// too surprisingly these map to U+00A0, etc. Which is the Unicode Latin Supplement range.
		
		"\xC2\xA0", "\xC2\xA1", "\xC2\xA2", "\xC2\xA3", "\xC2\xA4", "\xC2\xA5", "\xC2\xA6", "\xC2\xA7",	// 0xA0 .. 0xA7
		"\xC2\xA8", "\xC2\xA9", "\xC2\xAA", "\xC2\xAB", "\xC2\xAC", "\xC2\xAD", "\xC2\xAE", "\xC2\xAF",	// 0xA8 .. 0xAF

		"\xC2\xB0", "\xC2\xB1", "\xC2\xB2", "\xC2\xB3", "\xC2\xB4", "\xC2\xB5", "\xC2\xB6", "\xC2\xB7",	// 0xB0 .. 0xB7
		"\xC2\xB8", "\xC2\xB9", "\xC2\xBA", "\xC2\xBB", "\xC2\xBC", "\xC2\xBD", "\xC2\xBE", "\xC2\xBF",	// 0xB8 .. 0xBF

		"\xC3\x80", "\xC3\x81", "\xC3\x82", "\xC3\x83", "\xC3\x84", "\xC3\x85", "\xC3\x86", "\xC3\x87",	// 0xC0 .. 0xC7
		"\xC3\x88", "\xC3\x89", "\xC3\x8A", "\xC3\x8B", "\xC3\x8C", "\xC3\x8D", "\xC3\x8E", "\xC3\x8F",	// 0xC8 .. 0xCF

		"\xC3\x90", "\xC3\x91", "\xC3\x92", "\xC3\x93", "\xC3\x94", "\xC3\x95", "\xC3\x96", "\xC3\x97",	// 0xD0 .. 0xD7
		"\xC3\x98", "\xC3\x99", "\xC3\x9A", "\xC3\x9B", "\xC3\x9C", "\xC3\x9D", "\xC3\x9E", "\xC3\x9F",	// 0xD8 .. 0xDF

		"\xC3\xA0", "\xC3\xA1", "\xC3\xA2", "\xC3\xA3", "\xC3\xA4", "\xC3\xA5", "\xC3\xA6", "\xC3\xA7",	// 0xE0 .. 0xE7
		"\xC3\xA8", "\xC3\xA9", "\xC3\xAA", "\xC3\xAB", "\xC3\xAC", "\xC3\xAD", "\xC3\xAE", "\xC3\xAF",	// 0xE8 .. 0xEF

		"\xC3\xB0", "\xC3\xB1", "\xC3\xB2", "\xC3\xB3", "\xC3\xB4", "\xC3\xB5", "\xC3\xB6", "\xC3\xB7",	// 0xF0 .. 0xF7
		"\xC3\xB8", "\xC3\xB9", "\xC3\xBA", "\xC3\xBB", "\xC3\xBC", "\xC3\xBD", "\xC3\xBE", "\xC3\xBF",	// 0xF8 .. 0xFF

	};


// =================================================================================================
// Local Utilities
// ===============


#define IsHexDigit(ch)		( (('0' <= (ch)) && ((ch) <= '9')) || (('A' <= (ch)) && ((ch) <= 'F')) )
#define HexDigitValue(ch)	( (((ch) - '0') < 10) ? ((ch) - '0') : ((ch) - 'A' + 10) )


// -------------------------------------------------------------------------------------------------
// PickBestRoot
// ------------
static const XML_Node * PickBestRoot ( const XML_Node & xmlParent, XMP_OptionBits options )
{

	// Look among this parent's content for x:xmpmeta. The recursion for x:xmpmeta is broader than
	// the strictly defined choice, but gives us smaller code.
	for ( size_t childNum = 0, childLim = xmlParent.content.size(); childNum < childLim; ++childNum ) {
		const XML_Node * childNode = xmlParent.content[childNum];
		if ( childNode->kind != kElemNode ) continue;
		if ( (childNode->name == "x:xmpmeta") || (childNode->name == "x:xapmeta") ) return PickBestRoot ( *childNode, 0 );
	}
	// Look among this parent's content for a bare rdf:RDF if that is allowed.
	if ( ! (options & kXMP_RequireXMPMeta) ) {
		for ( size_t childNum = 0, childLim = xmlParent.content.size(); childNum < childLim; ++childNum ) {
			const XML_Node * childNode = xmlParent.content[childNum];
			if ( childNode->kind != kElemNode ) continue;
			if ( childNode->name == "rdf:RDF" ) return childNode;
		}
	}
	
	// Recurse into the content.
	for ( size_t childNum = 0, childLim = xmlParent.content.size(); childNum < childLim; ++childNum ) {
		const XML_Node * foundRoot = PickBestRoot ( *xmlParent.content[childNum], options );
		if ( foundRoot != 0 ) return foundRoot;
	}
	
	return 0;

}	// PickBestRoot

// -------------------------------------------------------------------------------------------------
// FindRootNode
// ------------
//
// Find the XML node that is the root of the XMP data tree. Generally this will be an outer node,
// but it could be anywhere if a general XML document is parsed (e.g. SVG). The XML parser counted
// all possible root nodes, and kept a pointer to the last one. If there is more than one possible
// root use PickBestRoot to choose among them.
//
// If there is a root node, try to extract the version of the previous XMP toolkit.

static const XML_Node * FindRootNode ( XMPMeta * thiz, const XMLParserAdapter & xmlParser, XMP_OptionBits options )
{
	const XML_Node * rootNode = xmlParser.rootNode;
	
	if ( xmlParser.rootCount > 1 ) rootNode = PickBestRoot ( xmlParser.tree, options );
	if ( rootNode == 0 ) return 0;
	
	// We have a root node. Try to extract previous toolkit version number.
	
	XMP_StringPtr verStr = "";
	
		XMP_Assert ( rootNode->name == "rdf:RDF" );
	
		if ( (options & kXMP_RequireXMPMeta) &&
		     ((rootNode->parent == 0) ||
		      ((rootNode->parent->name != "x:xmpmeta") && (rootNode->parent->name != "x:xapmeta"))) ) return 0;

		for ( size_t attrNum = 0, attrLim = rootNode->parent->attrs.size(); attrNum < attrLim; ++attrNum ) {
			const XML_Node * currAttr =rootNode->parent->attrs[attrNum];
			if ( (currAttr->name == "x:xmptk") || (currAttr->name == "x:xaptk") ) {
				verStr = currAttr->value.c_str();
				break;
			}
		}
		
	// Decode the version number into MMmmuubbb digits. If any part is too big, peg it at 99 or 999.
	
	unsigned long part;
	while ( (*verStr != 0) && ((*verStr < '0') || (*verStr > '9')) ) ++verStr;
	
	part = 0;
	while ( (*verStr != 0) && ('0' <= *verStr) && (*verStr <= '9') ) {
		part = (part * 10) + (*verStr - '0');
		++verStr;
	}
	if ( part > 99 ) part = 99;
	thiz->prevTkVer = part * 100*100*1000;
	
	part = 0;
	if ( *verStr == '.' ) ++verStr;
	while ( (*verStr != 0) && ('0' <= *verStr) && (*verStr <= '9') ) {
		part = (part * 10) + (*verStr - '0');
		++verStr;
	}
	if ( part > 99 ) part = 99;
	thiz->prevTkVer += part * 100*1000;
	
	part = 0;
	if ( *verStr == '.' ) ++verStr;
	while ( (*verStr != 0) && ('0' <= *verStr) && (*verStr <= '9') ) {
		part = (part * 10) + (*verStr - '0');
		++verStr;
	}
	if ( part > 99 ) part = 99;
	thiz->prevTkVer += part * 1000;
	
	part = 0;
	if ( *verStr == '-' ) ++verStr;
	while ( (*verStr != 0) && ('0' <= *verStr) && (*verStr <= '9') ) {
		part = (part * 10) + (*verStr - '0');
		++verStr;
	}
	if ( part > 999 ) part = 999;
	thiz->prevTkVer += part;
	
	return rootNode;
	
}	// FindRootNode

// -------------------------------------------------------------------------------------------------
// NormalizeDCArrays
// -----------------
//
// Undo the denormalization performed by the XMP used in Acrobat 5. If a Dublin Core array had only
// one item, it was serialized as a simple property. The xml:lang attribute was dropped from an
// alt-text item if the language was x-default.

// *** This depends on the dc: namespace prefix.

static void
NormalizeDCArrays ( XMP_Node * xmpTree )
{
	XMP_Node * dcSchema = FindSchemaNode ( xmpTree, kXMP_NS_DC, kXMP_ExistingOnly );
	if ( dcSchema == 0 ) return;
	
	for ( size_t propNum = 0, propLimit = dcSchema->children.size(); propNum < propLimit; ++propNum ) {
		XMP_Node *     currProp  = dcSchema->children[propNum];
		XMP_OptionBits arrayForm = 0;
		
		if ( ! XMP_PropIsSimple ( currProp->options ) ) continue;	// Nothing to do if not simple.
		
		if ( (currProp->name == "dc:creator" )     ||	// See if it is supposed to be an array.
		     (currProp->name == "dc:date" ) ) {			// *** Think about an array of char* and a loop.
			arrayForm = kXMP_PropArrayIsOrdered;
		} else if (
		     (currProp->name == "dc:description" ) ||
		     (currProp->name == "dc:rights" )      ||
		     (currProp->name == "dc:title" ) ) {
			arrayForm = kXMP_PropArrayIsAltText;
		} else if (
		     (currProp->name == "dc:contributor" ) ||
		     (currProp->name == "dc:language" )    ||
		     (currProp->name == "dc:publisher" )   ||
		     (currProp->name == "dc:relation" )    ||
		     (currProp->name == "dc:subject" )     ||
		     (currProp->name == "dc:type" ) ) {
			arrayForm = kXMP_PropValueIsArray;
		}
		if ( arrayForm == 0 ) continue;	// Nothing to do if it isn't supposed to be an array.
		
		arrayForm = VerifySetOptions ( arrayForm, 0 );	// Set the implicit array bits.
		XMP_Node * newArray = new XMP_Node ( dcSchema, currProp->name.c_str(), arrayForm );
		dcSchema->children[propNum] = newArray;
		newArray->children.push_back ( currProp );
		currProp->parent = newArray;
		currProp->name = kXMP_ArrayItemName;
		
		if ( XMP_ArrayIsAltText ( arrayForm ) && (! (currProp->options & kXMP_PropHasLang)) ) {
			XMP_Node * newLang = new XMP_Node ( currProp, "xml:lang", "x-default", kXMP_PropIsQualifier );
			currProp->options |= (kXMP_PropHasQualifiers | kXMP_PropHasLang);
			if ( currProp->qualifiers.empty() ) {	// *** Need a util?
				currProp->qualifiers.push_back ( newLang );
			} else {
				currProp->qualifiers.insert ( currProp->qualifiers.begin(), newLang );
			}
		}

	}
	
}	// NormalizeDCArrays


// -------------------------------------------------------------------------------------------------
// CompareAliasedSubtrees
// ----------------------

// *** Change to do some alias-specific setup, then use CompareSubtrees. One special case for
// *** aliases is a simple to x-default alias, the options and qualifiers obviously differ.

static void
CompareAliasedSubtrees ( XMP_Node * aliasNode, XMP_Node * baseNode, bool outerCall = true )
{
	// ! The outermost call is special. The names almost certainly differ. The qualifiers (and
	// ! hence options) will differ for an alias to the x-default item of a langAlt array.
	if ( (aliasNode->value != baseNode->value) ||
	     (aliasNode->children.size() != baseNode->children.size()) ) {
		XMP_Throw ( "Mismatch between alias and base nodes", kXMPErr_BadXMP );
	}
	if ( ! outerCall ) {
		if ( (aliasNode->name != baseNode->name) ||
		     (aliasNode->options != baseNode->options) ||
		     (aliasNode->qualifiers.size() != baseNode->qualifiers.size()) ) {
			XMP_Throw ( "Mismatch between alias and base nodes", kXMPErr_BadXMP );
		}
	}
	
	for ( size_t childNum = 0, childLim = aliasNode->children.size(); childNum < childLim; ++childNum ) {
		XMP_Node * aliasChild = aliasNode->children[childNum];
		XMP_Node * baseChild  = baseNode->children[childNum];
		CompareAliasedSubtrees ( aliasChild, baseChild, false );
	}
	
	for ( size_t qualNum = 0, qualLim = aliasNode->qualifiers.size(); qualNum < qualLim; ++qualNum ) {
		XMP_Node * aliasQual = aliasNode->qualifiers[qualNum];
		XMP_Node * baseQual  = baseNode->qualifiers[qualNum];
		CompareAliasedSubtrees ( aliasQual, baseQual, false );
	}
	
}	// CompareAliasedSubtrees


// -------------------------------------------------------------------------------------------------
// TransplantArrayItemAlias
// ------------------------

static void
TransplantArrayItemAlias ( XMP_Node * oldParent, size_t oldNum, XMP_Node * newParent )
{
	XMP_Node * childNode = oldParent->children[oldNum];

	if ( newParent->options & kXMP_PropArrayIsAltText ) {
		if ( childNode->options & kXMP_PropHasLang ) {
			XMP_Throw ( "Alias to x-default already has a language qualifier", kXMPErr_BadXMP );	// *** Allow x-default.
		}
		childNode->options |= (kXMP_PropHasQualifiers | kXMP_PropHasLang);
		XMP_Node * langQual = new XMP_Node ( childNode, "xml:lang", "x-default", kXMP_PropIsQualifier );	// *** AddLangQual util?
		if ( childNode->qualifiers.empty() ) {
			childNode->qualifiers.push_back ( langQual );
		} else {
			childNode->qualifiers.insert ( childNode->qualifiers.begin(), langQual );
		}
	}

	oldParent->children.erase ( oldParent->children.begin() + oldNum );
	childNode->name = kXMP_ArrayItemName;
	childNode->parent = newParent;
	if ( newParent->children.empty() ) {
		newParent->children.push_back ( childNode );
	} else {
		newParent->children.insert ( newParent->children.begin(), childNode );
	}

}	// TransplantArrayItemAlias


// -------------------------------------------------------------------------------------------------
// TransplantNamedAlias
// --------------------

static void
TransplantNamedAlias ( XMP_Node * oldParent, size_t oldNum, XMP_Node * newParent, XMP_VarString & newName )
{
	XMP_Node * childNode = oldParent->children[oldNum];

	oldParent->children.erase ( oldParent->children.begin() + oldNum );
	childNode->name = newName;
	childNode->parent = newParent;
	newParent->children.push_back ( childNode );

}	// TransplantNamedAlias


// -------------------------------------------------------------------------------------------------
// MoveExplicitAliases
// -------------------

static void
MoveExplicitAliases ( XMP_Node * tree, XMP_OptionBits parseOptions )
{
	tree->options ^= kXMP_PropHasAliases;
	const bool strictAliasing = ((parseOptions & kXMP_StrictAliasing) != 0);
	
	// Visit all of the top level nodes looking for aliases. If there is no base, transplant the
	// alias subtree. If there is a base and strict aliasing is on, make sure the alias and base 
	// subtrees match.
	
	// ! Use "while" loops not "for" loops since both the schema and property loops can remove the
	// ! current item from the vector being traversed. And don't increment the counter for a delete.
	
	size_t schemaNum = 0;
	while ( schemaNum < tree->children.size() ) {
		XMP_Node * currSchema = tree->children[schemaNum];
		
		size_t propNum = 0;
		while ( propNum < currSchema->children.size() ) {
			XMP_Node * currProp = currSchema->children[propNum];
			if ( ! (currProp->options & kXMP_PropIsAlias) ) {
				++propNum;
				continue;
			}
			currProp->options ^= kXMP_PropIsAlias;

			// Find the base path, look for the base schema and root node.

			XMP_AliasMapPos aliasPos = sRegisteredAliasMap->find ( currProp->name );
			XMP_Assert ( aliasPos != sRegisteredAliasMap->end() );
			XMP_ExpandedXPath & basePath = aliasPos->second;
			XMP_OptionBits arrayOptions = (basePath[kRootPropStep].options & kXMP_PropArrayFormMask);

			XMP_Node * baseSchema = FindSchemaNode ( tree, basePath[kSchemaStep].step.c_str(), kXMP_CreateNodes );
			if ( baseSchema->options & kXMP_NewImplicitNode ) baseSchema->options ^= kXMP_NewImplicitNode;
			XMP_Node * baseNode = FindChildNode ( baseSchema, basePath[kRootPropStep].step.c_str(), kXMP_ExistingOnly );

			if ( baseNode == 0 ) {
			
				if ( basePath.size() == 2 ) {
					// A top-to-top alias, transplant the property.
					TransplantNamedAlias ( currSchema, propNum, baseSchema, basePath[kRootPropStep].step );
				} else {
					// An alias to an array item, create the array and transplant the property.
					baseNode = new XMP_Node ( baseSchema, basePath[kRootPropStep].step.c_str(), arrayOptions );
					baseSchema->children.push_back ( baseNode );
					TransplantArrayItemAlias ( currSchema, propNum, baseNode );
				}
			
			} else if ( basePath.size() == 2 ) {
			
				// The base node does exist and this is a top-to-top alias. Check for conflicts if
				// strict aliasing is on. Remove and delete the alias subtree.
				if ( strictAliasing ) CompareAliasedSubtrees ( currProp, baseNode );
				currSchema->children.erase ( currSchema->children.begin() + propNum );
				delete currProp;
			
			} else {
			
				// This is an alias to an array item and the array exists. Look for the aliased item.
				// Then transplant or check & delete as appropriate.
				
				XMP_Node * itemNode = 0;
				if ( arrayOptions & kXMP_PropArrayIsAltText ) {
					XMP_Index xdIndex = LookupLangItem ( baseNode, *xdefaultName );
					if ( xdIndex != -1 ) itemNode = baseNode->children[xdIndex];
				} else if ( ! baseNode->children.empty() ) {
					itemNode = baseNode->children[0];
				}
				
				if ( itemNode == 0 ) {
					TransplantArrayItemAlias ( currSchema, propNum, baseNode );
				} else {
					if ( strictAliasing ) CompareAliasedSubtrees ( currProp, itemNode );
					currSchema->children.erase ( currSchema->children.begin() + propNum );
					delete currProp;
				}

			}

		}	// Property loop
		
		// Increment the counter or remove an empty schema node.
		if ( currSchema->children.size() > 0 ) {
			++schemaNum;
		} else {
			delete tree->children[schemaNum];	// ! Delete the schema node itself.
			tree->children.erase ( tree->children.begin() + schemaNum );
		}
		
	}	// Schema loop
	
}	// MoveExplicitAliases


// -------------------------------------------------------------------------------------------------
// FixGPSTimeStamp
// ---------------

static void
FixGPSTimeStamp ( XMP_Node * exifSchema, XMP_Node * gpsDateTime )
{
	XMP_DateTime binGPSStamp;
	try {
		XMPUtils::ConvertToDate ( gpsDateTime->value.c_str(), &binGPSStamp );
	} catch ( ... ) {
		return;	// Don't let a bad date stop other things.
	}
	if ( (binGPSStamp.year != 0) || (binGPSStamp.month != 0) || (binGPSStamp.day != 0) ) return;
	
	XMP_Node * otherDate = FindChildNode ( exifSchema, "exif:DateTimeOriginal", kXMP_ExistingOnly );
	if ( otherDate == 0 ) otherDate = FindChildNode ( exifSchema, "exif:DateTimeDigitized", kXMP_ExistingOnly );
	if ( otherDate == 0 ) return;

	XMP_DateTime binOtherDate;
	try {
		XMPUtils::ConvertToDate ( otherDate->value.c_str(), &binOtherDate );
	} catch ( ... ) {
		return;	// Don't let a bad date stop other things.
	}
	
	binGPSStamp.year  = binOtherDate.year;
	binGPSStamp.month = binOtherDate.month;
	binGPSStamp.day   = binOtherDate.day;

	XMP_StringPtr goodStr;
	XMP_StringLen goodLen;
	XMPUtils::ConvertFromDate ( binGPSStamp, &goodStr, &goodLen );
	
	gpsDateTime->value.assign ( goodStr, goodLen );

}	// FixGPSTimeStamp


// -------------------------------------------------------------------------------------------------
// MigrateAudioCopyright
// ---------------------
//
// The initial support for WAV files mapped a legacy ID3 audio copyright into a new xmpDM:copyright
// property. This is special case code to migrate that into dc:rights['x-default']. The rules:
//
//   1. If there is no dc:rights array, or an empty array -
//      Create one with dc:rights['x-default'] set from double linefeed and xmpDM:copyright.
//
//   2. If there is a dc:rights array but it has no x-default item -
//      Create an x-default item as a copy of the first item then apply rule #3.
//
//   3. If there is a dc:rights array with an x-default item, look for a double linefeed in the value.
//      A. If no double linefeed, compare the x-default value to the xmpDM:copyright value.
//         A1. If they match then leave the x-default value alone.
//         A2. Otherwise, append a double linefeed and the xmpDM:copyright value to the x-default value.
//      B. If there is a double linefeed, compare the trailing text to the xmpDM:copyright value.
//         B1. If they match then leave the x-default value alone.
//         B2. Otherwise, replace the trailing x-default text with the xmpDM:copyright value.
//
//   4. In all cases, delete the xmpDM:copyright property.

static void
MigrateAudioCopyright ( XMPMeta * xmp, XMP_Node * dmCopyright )
{

	try {
	
		std::string & dmValue = dmCopyright->value;
		static const char * kDoubleLF = "\xA\xA";
		
		XMP_Node * dcSchema = FindSchemaNode ( &xmp->tree, kXMP_NS_DC, kXMP_CreateNodes );
		XMP_Node * dcRightsArray = FindChildNode ( dcSchema, "dc:rights", kXMP_ExistingOnly );
		
		if ( (dcRightsArray == 0) || dcRightsArray->children.empty() ) {
		
			// 1. No dc:rights array, create from double linefeed and xmpDM:copyright.
			dmValue.insert ( 0, kDoubleLF );
			xmp->SetLocalizedText ( kXMP_NS_DC, "rights", "", "x-default",  dmValue.c_str(), 0 );
		
		} else {

			std::string xdefaultStr ( "x-default" );
			
			XMP_Index xdIndex = LookupLangItem ( dcRightsArray, xdefaultStr );
			
			if ( xdIndex < 0 ) {
				// 2. No x-default item, create from the first item.
				XMP_StringPtr firstValue = dcRightsArray->children[0]->value.c_str();
				xmp->SetLocalizedText ( kXMP_NS_DC, "rights", "", "x-default",  firstValue, 0 );
				xdIndex = LookupLangItem ( dcRightsArray, xdefaultStr );
			}
						
			// 3. Look for a double linefeed in the x-default value.
			XMP_Assert ( xdIndex == 0 );
			std::string & defaultValue = dcRightsArray->children[xdIndex]->value;
			XMP_Index lfPos = defaultValue.find ( kDoubleLF );
			
			if ( lfPos < 0 ) {
			
				// 3A. No double LF, compare whole values.
				if ( dmValue != defaultValue ) {
					// 3A2. Append the xmpDM:copyright to the x-default item.
					defaultValue += kDoubleLF;
					defaultValue += dmValue;
				}
			
			} else {
			
				// 3B. Has double LF, compare the tail.
				if ( defaultValue.compare ( lfPos+2, std::string::npos, dmValue ) != 0 ) {
					// 3B2. Replace the x-default tail.
					defaultValue.replace ( lfPos+2, std::string::npos, dmValue );
				}
			
			}

		}
		
		// 4. Get rid of the xmpDM:copyright.
		xmp->DeleteProperty ( kXMP_NS_DM, "copyright" );
	
	} catch ( ... ) {
		// Don't let failures (like a bad dc:rights form) stop other cleanup.
	}

}	// MigrateAudioCopyright


// -------------------------------------------------------------------------------------------------
// RepairAltText
// -------------
//
// Make sure that the array is well-formed AltText. Each item must be simple and have an xml:lang
// qualifier. If repairs are needed, keep simple non-empty items by adding the xml:lang.

static void
RepairAltText ( XMP_Node & tree, XMP_StringPtr schemaNS, XMP_StringPtr arrayName )
{
	XMP_Node * schemaNode = FindSchemaNode ( &tree, schemaNS, kXMP_ExistingOnly );
	if ( schemaNode == 0 ) return;
	
	XMP_Node * arrayNode = FindChildNode ( schemaNode, arrayName, kXMP_ExistingOnly );
	if ( (arrayNode == 0) || XMP_ArrayIsAltText ( arrayNode->options ) ) return;	// Already OK.
	
	if ( ! XMP_PropIsArray ( arrayNode->options ) ) return;	// ! Not even an array, leave it alone.
	// *** Should probably change simple values to LangAlt with 'x-default' item.
	
	arrayNode->options |= (kXMP_PropArrayIsOrdered | kXMP_PropArrayIsAlternate | kXMP_PropArrayIsAltText);
	
	for ( int i = arrayNode->children.size()-1; i >= 0; --i ) {	// ! Need a signed index type.

		XMP_Node * currChild = arrayNode->children[i];

		if ( ! XMP_PropIsSimple ( currChild->options ) ) {

			// Delete non-simple children.
			delete ( currChild );
			arrayNode->children.erase ( arrayNode->children.begin() + i );

		} else if ( ! XMP_PropHasLang ( currChild->options ) ) {
		
			if ( currChild->value.empty() ) {

				// Delete empty valued children that have no xml:lang.
				delete ( currChild );
				arrayNode->children.erase ( arrayNode->children.begin() + i );

			} else {

				// Add an xml:lang qualifier with the value "x-repair".
				XMP_Node * repairLang = new XMP_Node ( currChild, "xml:lang", "x-repair", kXMP_PropIsQualifier );
				if ( currChild->qualifiers.empty() ) {
					currChild->qualifiers.push_back ( repairLang );
				} else {
					currChild->qualifiers.insert ( currChild->qualifiers.begin(), repairLang );
				}
				currChild->options |= (kXMP_PropHasQualifiers | kXMP_PropHasLang);

			}

		}

	}

}	// RepairAltText


// -------------------------------------------------------------------------------------------------
// TouchUpDataModel
// ----------------

static void
TouchUpDataModel ( XMPMeta * xmp )
{
	XMP_Node & tree = xmp->tree;
	
	// Do special case touch ups for certain schema.

	XMP_Node * currSchema = 0;

	currSchema = FindSchemaNode ( &tree, kXMP_NS_EXIF, kXMP_ExistingOnly );
	if ( currSchema != 0 ) {

		// Do a special case fix for exif:GPSTimeStamp.
		XMP_Node * gpsDateTime = FindChildNode ( currSchema, "exif:GPSTimeStamp", kXMP_ExistingOnly );
		if ( gpsDateTime != 0 ) FixGPSTimeStamp ( currSchema, gpsDateTime );
	
		// *** Should probably have RepairAltText change simple values to LangAlt with 'x-default' item.
		// *** For now just do this for exif:UserComment, the one case we know about, late in cycle fix.
		XMP_Node * userComment = FindChildNode ( currSchema, "exif:UserComment", kXMP_ExistingOnly );
		if ( (userComment != 0) && XMP_PropIsSimple ( userComment->options ) ) {
			XMP_Node * newChild = new XMP_Node ( userComment, kXMP_ArrayItemName,
												 userComment->value.c_str(), userComment->options );
			newChild->qualifiers.swap ( userComment->qualifiers );
			if ( ! XMP_PropHasLang ( newChild->options ) ) {
				XMP_Node * langQual = new XMP_Node ( newChild, "xml:lang", "x-default", kXMP_PropIsQualifier );
				newChild->qualifiers.insert ( newChild->qualifiers.begin(), langQual );
				newChild->options |= (kXMP_PropHasQualifiers | kXMP_PropHasLang);
			}
			userComment->value.erase();
			userComment->options = kXMP_PropArrayFormMask;	// ! Happens to have all the right bits.
			userComment->children.push_back ( newChild );
		}

	}

	currSchema = FindSchemaNode ( &tree, kXMP_NS_DM, kXMP_ExistingOnly );
	if ( currSchema != 0 ) {
		// Do a special case migration of xmpDM:copyright to dc:rights['x-default']. Do this before
		// the dc: touch up since it can affect the dc: schema.
		XMP_Node * dmCopyright = FindChildNode ( currSchema, "xmpDM:copyright", kXMP_ExistingOnly );
		if ( dmCopyright != 0 ) MigrateAudioCopyright ( xmp, dmCopyright );
	}

	currSchema = FindSchemaNode ( &tree, kXMP_NS_DC, kXMP_ExistingOnly );
	if ( currSchema != 0 ) {
		// Do a special case fix for dc:subject, make sure it is an unordered array.
		XMP_Node * dcSubject = FindChildNode ( currSchema, "dc:subject", kXMP_ExistingOnly );
		if ( dcSubject != 0 ) {
                        XMP_OptionBits keepMask = static_cast<XMP_OptionBits>(~(kXMP_PropArrayIsOrdered | kXMP_PropArrayIsAlternate | kXMP_PropArrayIsAltText));
			dcSubject->options &= keepMask;	// Make sure any ordered array bits are clear.
		}
	}
	
	// Fix any broken AltText arrays that we know about.
	
	RepairAltText ( tree, kXMP_NS_DC, "dc:description" );	// ! Note inclusion of prefixes for direct node lookup!
	RepairAltText ( tree, kXMP_NS_DC, "dc:rights" );
	RepairAltText ( tree, kXMP_NS_DC, "dc:title" );
	RepairAltText ( tree, kXMP_NS_XMP_Rights, "xmpRights:UsageTerms" );
	RepairAltText ( tree, kXMP_NS_EXIF, "exif:UserComment" );
	
	// Tweak old XMP: Move an instance ID from rdf:about to the xmpMM:InstanceID property. An old
	// instance ID usually looks like "uuid:bac965c4-9d87-11d9-9a30-000d936b79c4", plus InDesign
	// 3.0 wrote them like "bac965c4-9d87-11d9-9a30-000d936b79c4". If the name looks like a UUID
	// simply move it to xmpMM:InstanceID, don't worry about any existing xmpMM:InstanceID. Both
	// will only be present when a newer file with the xmpMM:InstanceID property is updated by an
	// old app that uses rdf:about.
	
	if ( ! tree.name.empty() ) {

		bool nameIsUUID = false;
		XMP_StringPtr nameStr = tree.name.c_str();

		if ( XMP_LitNMatch ( nameStr, "uuid:", 5 ) ) {

			nameIsUUID = true;

		} else if ( tree.name.size() == 36 ) {

			nameIsUUID = true;	// ! Assume true, we'll set it to false below if not.
			for ( int i = 0;  i < 36; ++i ) {
				char ch = nameStr[i];
				if ( ch == '-' ) {
					if ( (i == 8) || (i == 13) || (i == 18) || (i == 23) ) continue;
					nameIsUUID = false;
					break;
				} else {
					if ( (('0' <= ch) && (ch <= '9')) || (('a' <= ch) && (ch <= 'z')) ) continue;
					nameIsUUID = false;
					break;
				}
			}

		}
		
		if ( nameIsUUID ) {

			XMP_ExpandedXPath expPath;
			ExpandXPath ( kXMP_NS_XMP_MM, "InstanceID", &expPath );
			XMP_Node * idNode = FindNode ( &tree, expPath, kXMP_CreateNodes, 0 );
			if ( idNode == 0 ) XMP_Throw ( "Failure creating xmpMM:InstanceID", kXMPErr_InternalFailure );

			idNode->options = 0;	// Clobber any existing xmpMM:InstanceID.
			idNode->value = tree.name;
			idNode->RemoveChildren();
			idNode->RemoveQualifiers();

			tree.name.erase();

		}

	}

}	// TouchUpDataModel


// -------------------------------------------------------------------------------------------------
// DetermineInputEncoding
// ----------------------
//
// Try to determine the character encoding, making a guess if the input is too short. We make some
// simplifying assumtions: the first character must be U+FEFF or ASCII, U+0000 is not allowed. The
// XML 1.1 spec is even more strict, UTF-16 XML documents must begin with U+FEFF, and the first
// "real" character must be '<'. Ignoring the XML declaration, the first XML character could be '<',
// space, tab, CR, or LF.
//
// The possible input sequences are:
//
//   Cases with U+FEFF
//      EF BB BF -- - UTF-8
//      FE FF -- -- - Big endian UTF-16
//      00 00 FE FF - Big endian UTF 32
//      FF FE 00 00 - Little endian UTF-32
//      FF FE -- -- - Little endian UTF-16
//
//   Cases with ASCII
//      nn mm -- -- - UTF-8 -
//      00 00 00 nn - Big endian UTF-32
//      00 nn -- -- - Big endian UTF-16
//      nn 00 00 00 - Little endian UTF-32
//      nn 00 -- -- - Little endian UTF-16
//
// ! We don't check for full patterns, or for errors. We just check enough to determine what the
// ! only possible (or reasonable) case would be.

static XMP_OptionBits
DetermineInputEncoding ( const XMP_Uns8 * buffer, size_t length )
{
	if ( length < 2 ) return kXMP_EncodeUTF8;
	
	XMP_Uns8 * uniChar = (XMP_Uns8*)buffer;	// ! Make sure comparisons are unsigned.
	
	if ( uniChar[0] == 0 ) {
	
		// These cases are:
		//   00 nn -- -- - Big endian UTF-16
		//   00 00 00 nn - Big endian UTF-32
		//   00 00 FE FF - Big endian UTF 32

		if ( (length < 4) || (uniChar[1] != 0) ) return kXMP_EncodeUTF16Big;
		return kXMP_EncodeUTF32Big;
		
	} else if ( uniChar[0] < 0x80 ) {
	
		// These cases are:
		//   nn mm -- -- - UTF-8, includes EF BB BF case
		//   nn 00 00 00 - Little endian UTF-32
		//   nn 00 -- -- - Little endian UTF-16

		if ( uniChar[1] != 0 )  return kXMP_EncodeUTF8;
		if ( (length < 4) || (uniChar[2] != 0) ) return kXMP_EncodeUTF16Little;
		return kXMP_EncodeUTF32Little;

	} else {
	
		// These cases are:
		//   EF BB BF -- - UTF-8
		//   FE FF -- -- - Big endian UTF-16
		//   FF FE 00 00 - Little endian UTF-32
		//   FF FE -- -- - Little endian UTF-16

		if ( uniChar[0] == 0xEF ) return kXMP_EncodeUTF8;
		if ( uniChar[0] == 0xFE ) return kXMP_EncodeUTF16Big;
		if ( (length < 4) || (uniChar[2] != 0) ) return kXMP_EncodeUTF16Little;
		return kXMP_EncodeUTF32Little;

	}
		
}	// DetermineInputEncoding


// -------------------------------------------------------------------------------------------------
// CountUTF8
// ---------
//
// Look for a valid multi-byte UTF-8 sequence and return its length. Returns 0 for an invalid UTF-8
// sequence. Returns a negative value for a partial valid sequence at the end of the buffer.
//
// The checking is not strict. We simply count the number of high order 1 bits in the first byte,
// then look for n-1 following bytes whose high order 2 bits are 1 and 0. We do not check for a
// minimal length representation of the codepoint, or that the codepoint is defined by Unicode.

static int
CountUTF8 ( const XMP_Uns8 * charStart, const XMP_Uns8 * bufEnd )
{
	XMP_Assert ( charStart < bufEnd );		// Catch this in debug builds.
	if ( charStart >= bufEnd ) return 0;	// Don't run-on in release builds.
	if ( (*charStart & 0xC0) != 0xC0 ) return 0;	// Must have at least 2 high bits set.
	
	int byteCount = 2;
	XMP_Uns8 firstByte = *charStart;
	for ( firstByte = firstByte << 2; (firstByte & 0x80) != 0; firstByte = firstByte << 1 ) ++byteCount;
	
	if ( (charStart + byteCount) > bufEnd ) return -byteCount;

	for ( int i = 1; i < byteCount; ++i ) {
		if ( (charStart[i] & 0xC0) != 0x80 ) return 0;
	}
	
	return byteCount;
	
}	// CountUTF8


// -------------------------------------------------------------------------------------------------
// CountControlEscape
// ------------------
//
// Look for a numeric escape sequence for a "prohibited" ASCII control character. These are 0x7F,
// and the range 0x00..0x1F except for tab/LF/CR. Return 0 if this is definitely not a numeric
// escape, the length of the escape if found, or a negative value for a partial escape.

static int
CountControlEscape ( const XMP_Uns8 * escStart, const XMP_Uns8 * bufEnd )
{
	XMP_Assert ( escStart < bufEnd );	// Catch this in debug builds.
	if ( escStart >= bufEnd ) return 0;	// Don't run-on in release builds.
	XMP_Assert ( *escStart == '&' );
	
	size_t tailLen = bufEnd - escStart;
	if ( tailLen < 5 ) return -1;	// Don't need a more thorough check, we'll catch it on the next pass.
	
	if ( strncmp ( (char*)escStart, "&#x", 3 ) != 0 ) return 0;
	
	XMP_Uns8 escValue = 0;
	const XMP_Uns8 * escPos = escStart + 3;
	
	if ( ('0' <= *escPos) && (*escPos <= '9') ) {
		escValue = *escPos - '0';
		++escPos;
	} else if ( ('A' <= *escPos) && (*escPos <= 'F') ) {
		escValue = *escPos - 'A' + 10;
		++escPos;
	} else if ( ('a' <= *escPos) && (*escPos <= 'f') ) {
		escValue = *escPos - 'a' + 10;
		++escPos;
	}
	
	if ( ('0' <= *escPos) && (*escPos <= '9') ) {
		escValue = (escValue << 4) + (*escPos - '0');
		++escPos;
	} else if ( ('A' <= *escPos) && (*escPos <= 'F') ) {
		escValue = (escValue << 4) + (*escPos - 'A' + 10);
		++escPos;
	} else if ( ('a' <= *escPos) && (*escPos <= 'f') ) {
		escValue = (escValue << 4) + (*escPos - 'a' + 10);
		++escPos;
	}
	
	if ( escPos == bufEnd ) return -1;	// Partial escape.
	if ( *escPos != ';' ) return 0;
	
	size_t escLen = escPos - escStart + 1;
	if ( escLen < 5 ) return 0;	// ! Catch "&#x;".
	
	if ( (escValue == kTab) || (escValue == kLF) || (escValue == kCR) ) return 0;	// An allowed escape.
	
	return escLen;	// Found a full "prohibited" numeric escape.
	
}	// CountControlEscape


// -------------------------------------------------------------------------------------------------
// ProcessUTF8Portion
// ------------------
//
// Early versions of the XMP spec mentioned allowing ISO Latin-1 input. There are also problems with
// some clients placing ASCII control characters within XMP values. This is an XML problem, the XML
// spec only allows tab (0x09), LF (0x0A), and CR (0x0D) from the 0x00..0x1F range. As a concession
// to this we scan 8-bit input for byte sequences that are not valid UTF-8 or in the 0x00..0x1F
// range and replace each byte as follows:
//   0x00..0x1F - Replace with a space, except for tab, CR, and LF.
//   0x7F       - Replace with a space. This is ASCII Delete, not allowed by ISO Latin-1.
//   0x80..0x9F - Replace with the UTF-8 for a corresponding Unicode character.
//   0xA0..0XFF - Replace with the UTF-8 for a corresponding Unicode character.
//
// The 0x80..0x9F range is not defined by Latin-1. But the Windows 1252 code page defines these and
// is otherwise the same as Latin-1.
//
// For at least historical compatibility reasons we also find and replace singly escaped ASCII
// control characters. The Expat parser we're using does not allow numeric escapes like "&#x10;".
// The XML spec is clear that raw controls are not allowed (in the RestrictedChar set), but it isn't
// as clear about numeric escapes for them. At any rate, Expat complains, so we treat the numeric
// escapes like raw characters and replace them with a space.
//
// We check for 1 or 2 hex digits ("&#x9;" or "&#x09;") and upper or lower case ("&#xA;" or "&#xa;").
// The full escape sequence is 5 or 6 bytes.

static size_t
ProcessUTF8Portion ( XMLParserAdapter * xmlParser,
					 const XMP_Uns8 *   buffer,
					 size_t				length,
					 bool				last )
{
	const XMP_Uns8 * bufEnd = buffer + length;
	
	const XMP_Uns8 * spanStart = buffer;
	const XMP_Uns8 * spanEnd;
		
	for ( spanEnd = spanStart; spanEnd < bufEnd; ++spanEnd ) {

		if ( (0x20 <= *spanEnd) && (*spanEnd <= 0x7E) && (*spanEnd != '&') ) continue;	// A regular ASCII character.

		if ( *spanEnd >= 0x80 ) {
		
			// See if this is a multi-byte UTF-8 sequence, or a Latin-1 character to replace.

			int uniLen = CountUTF8 ( spanEnd, bufEnd );

			if ( uniLen > 0 ) {

				// A valid UTF-8 character, keep it as-is.
				spanEnd += uniLen - 1;	// ! The loop increment will put back the +1.

			} else if ( (uniLen < 0) && (! last) ) {

				// Have a partial UTF-8 character at the end of the buffer and more input coming.
				xmlParser->ParseBuffer ( spanStart, (spanEnd - spanStart), false );
				return (spanEnd - buffer);

			} else {

				// Not a valid UTF-8 sequence. Replace the first byte with the Latin-1 equivalent.
				xmlParser->ParseBuffer ( spanStart, (spanEnd - spanStart), false );
				const char * replacement = kReplaceLatin1 [ *spanEnd - 0x80 ];
				xmlParser->ParseBuffer ( replacement, strlen ( replacement ), false );
				spanStart = spanEnd + 1;	// ! The loop increment will do "spanEnd = spanStart".

			}
		
		} else if ( (*spanEnd < 0x20) || (*spanEnd == 0x7F) ) {

			// Replace ASCII controls other than tab, LF, and CR with a space.

			if ( (*spanEnd == kTab) || (*spanEnd == kLF) || (*spanEnd == kCR) ) continue;

			xmlParser->ParseBuffer ( spanStart, (spanEnd - spanStart), false );
			xmlParser->ParseBuffer ( " ", 1, false );
			spanStart = spanEnd + 1;	// ! The loop increment will do "spanEnd = spanStart".
		
		} else {
		
			// See if this is a numeric escape sequence for a prohibited ASCII control.
			
			XMP_Assert ( *spanEnd == '&' );
			int escLen = CountControlEscape ( spanEnd, bufEnd );
			
			if ( escLen < 0 ) {

				// Have a partial numeric escape in this buffer, wait for more input.
				if ( last ) continue;	// No more buffers, not an escape, absorb as normal input.
				xmlParser->ParseBuffer ( spanStart, (spanEnd - spanStart), false );
				return (spanEnd - buffer);

			} else if ( escLen > 0 ) {

				// Have a complete numeric escape to replace.
				xmlParser->ParseBuffer ( spanStart, (spanEnd - spanStart), false );
				xmlParser->ParseBuffer ( " ", 1, false );
				spanStart = spanEnd + escLen;
				spanEnd = spanStart - 1;	// ! The loop continuation will increment spanEnd!

			}

		}
		
	}
	
	XMP_Assert ( spanEnd == bufEnd );

	if ( spanStart < bufEnd ) xmlParser->ParseBuffer ( spanStart, (spanEnd - spanStart), false );
	if ( last ) xmlParser->ParseBuffer ( " ", 1, true );
	
	return length;

}	// ProcessUTF8Portion


// -------------------------------------------------------------------------------------------------
// ParseFromBuffer
// ---------------
//
// Although most clients will probably parse everything in one call, we have a buffered API model
// and need to support even the extreme case of 1 byte at a time parsing. This is considerably
// complicated by some special cases for 8-bit input. Because of this, the first thing we do is
// determine whether the input is 8-bit, UTF-16, or UTF-32.
//
// Both the 8-bit special cases and the encoding determination are easier to do with 8 bytes or more
// of input. The XMLParserAdapter class has a pending-input buffer for this. At the start of parsing
// we (moght) try to fill this buffer before determining the input character encoding. After that,
// we (might) use this buffer with the current input to simplify the logic in Process8BitInput. The
// "(might)" part means that we don't actually use the pending-input buffer unless we have to. In
// particular, the common case of single-buffer parsing won't use it.

void
XMPMeta::ParseFromBuffer ( XMP_StringPtr  buffer,
						   XMP_StringLen  xmpSize,
						   XMP_OptionBits options )
{
	if ( (buffer == 0) && (xmpSize != 0) ) XMP_Throw ( "Null parse buffer", kXMPErr_BadParam );
	if ( xmpSize == kXMP_UseNullTermination ) xmpSize = strlen ( buffer );
	
	const bool lastClientCall = ((options & kXMP_ParseMoreBuffers) == 0);	// *** Could use FlagIsSet & FlagIsClear macros.
	
	this->tree.ClearNode();	// Make sure the target XMP object is totally empty.

	if ( this->xmlParser == 0 ) {
		if ( (xmpSize == 0) && lastClientCall ) return;	// Tolerate empty parse. Expat complains if there are no XML elements.
		this->xmlParser = XMP_NewExpatAdapter();
	}
	
	XMLParserAdapter& parser = *this->xmlParser;
	
	#if 0	// XMP_DebugBuild
		if ( parser.parseLog != 0 ) {
			char message [200];	// AUDIT: Using sizeof(message) below for snprintf length is safe.
			snprintf ( message, sizeof(message), "<!-- ParseFromBuffer, length = %d, options = %X%s -->",	// AUDIT: See above.
					   xmpSize, options, (lastClientCall ? " (last)" : "") );
			fwrite ( message, 1, strlen(message), parser.parseLog );
			fflush ( parser.parseLog );
		}
	#endif
		
	try {	// Cleanup the tree and xmlParser if anything fails.
	
		// Determine the character encoding before doing any real parsing. This is needed to do the
		// 8-bit special processing.
		
		if ( parser.charEncoding == XMP_OptionBits(-1) ) {

			if ( (parser.pendingCount == 0) && (xmpSize >= kXMLPendingInputMax) ) {

				// This ought to be the common case, the first buffer is big enough.
				parser.charEncoding = DetermineInputEncoding ( (XMP_Uns8*)buffer, xmpSize );

			} else {
			
				// Try to fill the pendingInput buffer before calling DetermineInputEncoding.

				size_t pendingOverlap = kXMLPendingInputMax - parser.pendingCount;
				if ( pendingOverlap > xmpSize ) pendingOverlap = xmpSize;

				memcpy ( &parser.pendingInput[parser.pendingCount], buffer, pendingOverlap );	// AUDIT: Count is safe.
				buffer += pendingOverlap;
				xmpSize -= pendingOverlap;
				parser.pendingCount += pendingOverlap;

				if ( (! lastClientCall) && (parser.pendingCount < kXMLPendingInputMax) ) return;
				parser.charEncoding = DetermineInputEncoding ( parser.pendingInput, parser.pendingCount );
				
				#if Trace_ParsingHackery
					fprintf ( stderr, "XMP Character encoding is %d\n", parser.charEncoding );
				#endif
			
			}

		}
		
		// We have the character encoding. Process UTF-16 and UTF-32 as is. UTF-8 needs special
		// handling to take care of things like ISO Latin-1 or unescaped ASCII controls.

		XMP_Assert ( parser.charEncoding != XMP_OptionBits(-1) );

		if ( parser.charEncoding != kXMP_EncodeUTF8 ) {
		
			if ( parser.pendingCount > 0 ) {
				// Might have pendingInput from the above portion to determine the character encoding.
				parser.ParseBuffer ( parser.pendingInput, parser.pendingCount, false );
			}
			parser.ParseBuffer ( buffer, xmpSize, lastClientCall );
			
		} else {

			#if Trace_ParsingHackery
				fprintf ( stderr, "Parsing %d bytes @ %.8X, %s, %d pending, context: %.8s\n",
						  xmpSize, buffer, (lastClientCall ? "last" : "not last"), parser.pendingCount, buffer );
			#endif

			// The UTF-8 processing is a bit complex due to the need to tolerate ISO Latin-1 input.
			// This is done by scanning the input for byte sequences that are not valid UTF-8,
			// assuming they are Latin-1 characters in the range 0x80..0xFF. This requires saving a
			// pending input buffer to handle partial UTF-8 sequences at the end of a buffer.
			
			while ( parser.pendingCount > 0 ) {
			
				// We've got some leftover input, process it first then continue with the current
				// buffer. Try to fill the pendingInput buffer before parsing further. We use a loop
				// for weird edge cases like a 2 byte input buffer, using 1 byte for pendingInput,
				// then having a partial UTF-8 end and need to absorb more.
				
				size_t pendingOverlap = kXMLPendingInputMax - parser.pendingCount;
				if ( pendingOverlap > xmpSize ) pendingOverlap = xmpSize;
				
				memcpy ( &parser.pendingInput[parser.pendingCount], buffer, pendingOverlap );	// AUDIT: Count is safe.
				parser.pendingCount += pendingOverlap;
				buffer += pendingOverlap;
				xmpSize -= pendingOverlap;

				if ( (! lastClientCall) && (parser.pendingCount < kXMLPendingInputMax) ) return;
				size_t bytesDone = ProcessUTF8Portion ( &parser, parser.pendingInput, parser.pendingCount, lastClientCall );
				size_t bytesLeft = parser.pendingCount - bytesDone;

				#if Trace_ParsingHackery
					fprintf ( stderr, "   ProcessUTF8Portion handled %d pending bytes\n", bytesDone );
				#endif
				
				if ( bytesDone == parser.pendingCount ) {

					// Done with all of the pending input, move on to the current buffer.
					parser.pendingCount = 0;

				} else if ( bytesLeft <= pendingOverlap ) {

					// The leftover pending input all came from the current buffer. Exit this loop.
					buffer -= bytesLeft;
					xmpSize += bytesLeft;
					parser.pendingCount = 0;

				} else if ( xmpSize > 0 ) {

					// Pull more of the current buffer into the pending input and try again.
					// Backup by this pass's overlap so the loop entry code runs OK.
					parser.pendingCount -= pendingOverlap;
					buffer -= pendingOverlap;
					xmpSize += pendingOverlap;

				} else {

					// There is no more of the current buffer. Wait for more. Partial sequences at
					// the end of the last buffer should be treated as Latin-1 by ProcessUTF8Portion.
					XMP_Assert ( ! lastClientCall );
					parser.pendingCount = bytesLeft;
					memcpy ( &parser.pendingInput[0], &parser.pendingInput[bytesDone], bytesLeft );	// AUDIT: Count is safe.
					return;

				}
			
			}
			
			// Done with the pending input, process the current buffer.

			size_t bytesDone = ProcessUTF8Portion ( &parser, (XMP_Uns8*)buffer, xmpSize, lastClientCall );

			#if Trace_ParsingHackery
				fprintf ( stderr, "   ProcessUTF8Portion handled %d additional bytes\n", bytesDone );
			#endif
			
			if ( bytesDone < xmpSize ) {

				XMP_Assert ( ! lastClientCall );
				size_t bytesLeft = xmpSize - bytesDone;
				if ( bytesLeft > kXMLPendingInputMax ) XMP_Throw ( "Parser bytesLeft too large", kXMPErr_InternalFailure );

				memcpy ( parser.pendingInput, &buffer[bytesDone], bytesLeft );	// AUDIT: Count is safe.
				parser.pendingCount = bytesLeft;
				return;	// Wait for the next buffer.

			}

		}
		
		if ( lastClientCall ) {
		
			#if XMP_DebugBuild && DumpXMLParseTree
				if ( parser.parseLog == 0 ) parser.parseLog = stdout;
				DumpXMLTree ( parser.parseLog, parser.tree, 0 );
			#endif

			const XML_Node * xmlRoot = FindRootNode ( this, *this->xmlParser, options );

			if ( xmlRoot != 0 ) {

				ProcessRDF ( &this->tree, *xmlRoot, options );
				NormalizeDCArrays ( &this->tree );
				if ( this->tree.options & kXMP_PropHasAliases ) MoveExplicitAliases ( &this->tree, options );
				TouchUpDataModel ( this );
				
				// Delete empty schema nodes. Do this last, other cleanup can make empty schema.
				size_t schemaNum = 0;
				while ( schemaNum < this->tree.children.size() ) {
					XMP_Node * currSchema = this->tree.children[schemaNum];
					if ( currSchema->children.size() > 0 ) {
						++schemaNum;
					} else {
						delete this->tree.children[schemaNum];	// ! Delete the schema node itself.
						this->tree.children.erase ( this->tree.children.begin() + schemaNum );
					}
				}
				
			}

			delete this->xmlParser;
			this->xmlParser = 0;

		}
		
	} catch ( ... ) {

		delete this->xmlParser;
		this->xmlParser = 0;
		prevTkVer = 0;
		this->tree.ClearNode();
		throw;

	}
	
}	// ParseFromBuffer

// =================================================================================================