Sweden-Number/dlls/usp10/opentype.c

2440 lines
87 KiB
C

/*
* Opentype font interfaces for the Uniscribe Script Processor (usp10.dll)
*
* Copyright 2012 CodeWeavers, Aric Stewart
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
*
*/
#include <stdarg.h>
#include <stdlib.h>
#include "windef.h"
#include "winbase.h"
#include "wingdi.h"
#include "winuser.h"
#include "winnls.h"
#include "usp10.h"
#include "winternl.h"
#include "usp10_internal.h"
#include "wine/debug.h"
WINE_DEFAULT_DEBUG_CHANNEL(uniscribe);
#ifdef WORDS_BIGENDIAN
#define GET_BE_WORD(x) (x)
#define GET_BE_DWORD(x) (x)
#else
#define GET_BE_WORD(x) RtlUshortByteSwap(x)
#define GET_BE_DWORD(x) RtlUlongByteSwap(x)
#endif
#define round(x) (((x) < 0) ? (int)((x) - 0.5) : (int)((x) + 0.5))
/* These are all structures needed for the cmap format 12 table */
#define CMAP_TAG MS_MAKE_TAG('c', 'm', 'a', 'p')
typedef struct {
WORD platformID;
WORD encodingID;
DWORD offset;
} CMAP_EncodingRecord;
typedef struct {
WORD version;
WORD numTables;
CMAP_EncodingRecord tables[1];
} CMAP_Header;
typedef struct {
DWORD startCharCode;
DWORD endCharCode;
DWORD startGlyphID;
} CMAP_SegmentedCoverage_group;
typedef struct {
WORD format;
WORD reserved;
DWORD length;
DWORD language;
DWORD nGroups;
CMAP_SegmentedCoverage_group groups[1];
} CMAP_SegmentedCoverage;
/* These are all structures needed for the GDEF table */
enum {BaseGlyph=1, LigatureGlyph, MarkGlyph, ComponentGlyph};
typedef struct {
DWORD Version;
WORD GlyphClassDef;
WORD AttachList;
WORD LigCaretList;
WORD MarkAttachClassDef;
} GDEF_Header;
typedef struct {
WORD ClassFormat;
WORD StartGlyph;
WORD GlyphCount;
WORD ClassValueArray[1];
} OT_ClassDefFormat1;
typedef struct {
WORD Start;
WORD End;
WORD Class;
} OT_ClassRangeRecord;
typedef struct {
WORD ClassFormat;
WORD ClassRangeCount;
OT_ClassRangeRecord ClassRangeRecord[1];
} OT_ClassDefFormat2;
/* These are all structures needed for the GSUB table */
typedef struct {
DWORD version;
WORD ScriptList;
WORD FeatureList;
WORD LookupList;
} GSUB_Header;
typedef struct {
CHAR ScriptTag[4];
WORD Script;
} OT_ScriptRecord;
typedef struct {
WORD ScriptCount;
OT_ScriptRecord ScriptRecord[1];
} OT_ScriptList;
typedef struct {
CHAR LangSysTag[4];
WORD LangSys;
} OT_LangSysRecord;
typedef struct {
WORD DefaultLangSys;
WORD LangSysCount;
OT_LangSysRecord LangSysRecord[1];
} OT_Script;
typedef struct {
WORD LookupOrder; /* Reserved */
WORD ReqFeatureIndex;
WORD FeatureCount;
WORD FeatureIndex[1];
} OT_LangSys;
typedef struct {
CHAR FeatureTag[4];
WORD Feature;
} OT_FeatureRecord;
typedef struct {
WORD FeatureCount;
OT_FeatureRecord FeatureRecord[1];
} OT_FeatureList;
typedef struct {
WORD FeatureParams; /* Reserved */
WORD LookupCount;
WORD LookupListIndex[1];
} OT_Feature;
typedef struct {
WORD LookupCount;
WORD Lookup[1];
} OT_LookupList;
typedef struct {
WORD LookupType;
WORD LookupFlag;
WORD SubTableCount;
WORD SubTable[1];
} OT_LookupTable;
typedef struct {
WORD CoverageFormat;
WORD GlyphCount;
WORD GlyphArray[1];
} OT_CoverageFormat1;
typedef struct {
WORD Start;
WORD End;
WORD StartCoverageIndex;
} OT_RangeRecord;
typedef struct {
WORD CoverageFormat;
WORD RangeCount;
OT_RangeRecord RangeRecord[1];
} OT_CoverageFormat2;
typedef struct {
WORD SubstFormat; /* = 1 */
WORD Coverage;
WORD DeltaGlyphID;
} GSUB_SingleSubstFormat1;
typedef struct {
WORD SubstFormat; /* = 2 */
WORD Coverage;
WORD GlyphCount;
WORD Substitute[1];
}GSUB_SingleSubstFormat2;
typedef struct {
WORD SubstFormat; /* = 1 */
WORD Coverage;
WORD SequenceCount;
WORD Sequence[1];
}GSUB_MultipleSubstFormat1;
typedef struct {
WORD GlyphCount;
WORD Substitute[1];
}GSUB_Sequence;
typedef struct {
WORD SubstFormat; /* = 1 */
WORD Coverage;
WORD LigSetCount;
WORD LigatureSet[1];
}GSUB_LigatureSubstFormat1;
typedef struct {
WORD LigatureCount;
WORD Ligature[1];
}GSUB_LigatureSet;
typedef struct{
WORD LigGlyph;
WORD CompCount;
WORD Component[1];
}GSUB_Ligature;
typedef struct{
WORD SequenceIndex;
WORD LookupListIndex;
}GSUB_SubstLookupRecord;
typedef struct{
WORD SubstFormat; /* = 1 */
WORD Coverage;
WORD ChainSubRuleSetCount;
WORD ChainSubRuleSet[1];
}GSUB_ChainContextSubstFormat1;
typedef struct {
WORD SubstFormat; /* = 3 */
WORD BacktrackGlyphCount;
WORD Coverage[1];
}GSUB_ChainContextSubstFormat3_1;
typedef struct{
WORD InputGlyphCount;
WORD Coverage[1];
}GSUB_ChainContextSubstFormat3_2;
typedef struct{
WORD LookaheadGlyphCount;
WORD Coverage[1];
}GSUB_ChainContextSubstFormat3_3;
typedef struct{
WORD SubstCount;
GSUB_SubstLookupRecord SubstLookupRecord[1];
}GSUB_ChainContextSubstFormat3_4;
typedef struct {
WORD SubstFormat; /* = 1 */
WORD Coverage;
WORD AlternateSetCount;
WORD AlternateSet[1];
} GSUB_AlternateSubstFormat1;
typedef struct{
WORD GlyphCount;
WORD Alternate[1];
} GSUB_AlternateSet;
typedef struct {
WORD SubstFormat;
WORD ExtensionLookupType;
DWORD ExtensionOffset;
} GSUB_ExtensionPosFormat1;
/* These are all structures needed for the GPOS table */
typedef struct {
DWORD version;
WORD ScriptList;
WORD FeatureList;
WORD LookupList;
} GPOS_Header;
typedef struct {
WORD StartSize;
WORD EndSize;
WORD DeltaFormat;
WORD DeltaValue[1];
} OT_DeviceTable;
typedef struct {
WORD AnchorFormat;
WORD XCoordinate;
WORD YCoordinate;
} GPOS_AnchorFormat1;
typedef struct {
WORD AnchorFormat;
WORD XCoordinate;
WORD YCoordinate;
WORD AnchorPoint;
} GPOS_AnchorFormat2;
typedef struct {
WORD AnchorFormat;
WORD XCoordinate;
WORD YCoordinate;
WORD XDeviceTable;
WORD YDeviceTable;
} GPOS_AnchorFormat3;
typedef struct {
WORD XPlacement;
WORD YPlacement;
WORD XAdvance;
WORD YAdvance;
WORD XPlaDevice;
WORD YPlaDevice;
WORD XAdvDevice;
WORD YAdvDevice;
} GPOS_ValueRecord;
typedef struct {
WORD PosFormat;
WORD Coverage;
WORD ValueFormat;
WORD Value[1];
} GPOS_SinglePosFormat1;
typedef struct {
WORD PosFormat;
WORD Coverage;
WORD ValueFormat;
WORD ValueCount;
WORD Value[1];
} GPOS_SinglePosFormat2;
typedef struct {
WORD PosFormat;
WORD Coverage;
WORD ValueFormat1;
WORD ValueFormat2;
WORD PairSetCount;
WORD PairSetOffset[1];
} GPOS_PairPosFormat1;
typedef struct {
WORD PosFormat;
WORD Coverage;
WORD ValueFormat1;
WORD ValueFormat2;
WORD ClassDef1;
WORD ClassDef2;
WORD Class1Count;
WORD Class2Count;
WORD Class1Record[1];
} GPOS_PairPosFormat2;
typedef struct {
WORD SecondGlyph;
WORD Value1[1];
WORD Value2[1];
} GPOS_PairValueRecord;
typedef struct {
WORD PairValueCount;
GPOS_PairValueRecord PairValueRecord[1];
} GPOS_PairSet;
typedef struct {
WORD EntryAnchor;
WORD ExitAnchor;
} GPOS_EntryExitRecord;
typedef struct {
WORD PosFormat;
WORD Coverage;
WORD EntryExitCount;
GPOS_EntryExitRecord EntryExitRecord[1];
} GPOS_CursivePosFormat1;
typedef struct {
WORD PosFormat;
WORD MarkCoverage;
WORD BaseCoverage;
WORD ClassCount;
WORD MarkArray;
WORD BaseArray;
} GPOS_MarkBasePosFormat1;
typedef struct {
WORD BaseAnchor[1];
} GPOS_BaseRecord;
typedef struct {
WORD BaseCount;
GPOS_BaseRecord BaseRecord[1];
} GPOS_BaseArray;
typedef struct {
WORD Class;
WORD MarkAnchor;
} GPOS_MarkRecord;
typedef struct {
WORD MarkCount;
GPOS_MarkRecord MarkRecord[1];
} GPOS_MarkArray;
typedef struct {
WORD PosFormat;
WORD MarkCoverage;
WORD LigatureCoverage;
WORD ClassCount;
WORD MarkArray;
WORD LigatureArray;
} GPOS_MarkLigPosFormat1;
typedef struct {
WORD LigatureCount;
WORD LigatureAttach[1];
} GPOS_LigatureArray;
typedef struct {
WORD LigatureAnchor[1];
} GPOS_ComponentRecord;
typedef struct {
WORD ComponentCount;
GPOS_ComponentRecord ComponentRecord[1];
} GPOS_LigatureAttach;
typedef struct {
WORD PosFormat;
WORD Mark1Coverage;
WORD Mark2Coverage;
WORD ClassCount;
WORD Mark1Array;
WORD Mark2Array;
} GPOS_MarkMarkPosFormat1;
typedef struct {
WORD Mark2Anchor[1];
} GPOS_Mark2Record;
typedef struct {
WORD Mark2Count;
GPOS_Mark2Record Mark2Record[1];
} GPOS_Mark2Array;
typedef struct {
WORD SequenceIndex;
WORD LookupListIndex;
} GPOS_PosLookupRecord;
typedef struct {
WORD PosFormat;
WORD BacktrackGlyphCount;
WORD Coverage[1];
} GPOS_ChainContextPosFormat3_1;
typedef struct {
WORD InputGlyphCount;
WORD Coverage[1];
} GPOS_ChainContextPosFormat3_2;
typedef struct {
WORD LookaheadGlyphCount;
WORD Coverage[1];
} GPOS_ChainContextPosFormat3_3;
typedef struct {
WORD PosCount;
GPOS_PosLookupRecord PosLookupRecord[1];
} GPOS_ChainContextPosFormat3_4;
typedef struct {
WORD PosFormat;
WORD ExtensionLookupType;
DWORD ExtensionOffset;
} GPOS_ExtensionPosFormat1;
/**********
* CMAP
**********/
static VOID *load_CMAP_format12_table(HDC hdc, ScriptCache *psc)
{
CMAP_Header *CMAP_Table = NULL;
int length;
int i;
if (!psc->CMAP_Table)
{
length = GetFontData(hdc, CMAP_TAG , 0, NULL, 0);
if (length != GDI_ERROR)
{
psc->CMAP_Table = HeapAlloc(GetProcessHeap(),0,length);
GetFontData(hdc, CMAP_TAG , 0, psc->CMAP_Table, length);
TRACE("Loaded cmap table of %i bytes\n",length);
}
else
return NULL;
}
CMAP_Table = psc->CMAP_Table;
for (i = 0; i < GET_BE_WORD(CMAP_Table->numTables); i++)
{
if ( (GET_BE_WORD(CMAP_Table->tables[i].platformID) == 3) &&
(GET_BE_WORD(CMAP_Table->tables[i].encodingID) == 10) )
{
CMAP_SegmentedCoverage *format = (CMAP_SegmentedCoverage*)(((BYTE*)CMAP_Table) + GET_BE_DWORD(CMAP_Table->tables[i].offset));
if (GET_BE_WORD(format->format) == 12)
return format;
}
}
return NULL;
}
static int compare_group(const void *a, const void* b)
{
const DWORD *chr = a;
const CMAP_SegmentedCoverage_group *group = b;
if (*chr < GET_BE_DWORD(group->startCharCode))
return -1;
if (*chr > GET_BE_DWORD(group->endCharCode))
return 1;
return 0;
}
DWORD OpenType_CMAP_GetGlyphIndex(HDC hdc, ScriptCache *psc, DWORD utf32c, LPWORD pgi, DWORD flags)
{
/* BMP: use gdi32 for ease */
if (utf32c < 0x10000)
{
WCHAR ch = utf32c;
return GetGlyphIndicesW(hdc,&ch, 1, pgi, flags);
}
if (!psc->CMAP_format12_Table)
psc->CMAP_format12_Table = load_CMAP_format12_table(hdc, psc);
if (flags & GGI_MARK_NONEXISTING_GLYPHS)
*pgi = 0xffff;
else
*pgi = 0;
if (psc->CMAP_format12_Table)
{
CMAP_SegmentedCoverage *format = NULL;
CMAP_SegmentedCoverage_group *group = NULL;
format = (CMAP_SegmentedCoverage *)psc->CMAP_format12_Table;
group = bsearch(&utf32c, format->groups, GET_BE_DWORD(format->nGroups),
sizeof(CMAP_SegmentedCoverage_group), compare_group);
if (group)
{
DWORD offset = utf32c - GET_BE_DWORD(group->startCharCode);
*pgi = GET_BE_DWORD(group->startGlyphID) + offset;
return 0;
}
}
return 0;
}
/**********
* GDEF
**********/
static WORD OT_get_glyph_class(const void *table, WORD glyph)
{
WORD class = 0;
const OT_ClassDefFormat1 *cf1 = table;
if (!table) return 0;
if (GET_BE_WORD(cf1->ClassFormat) == 1)
{
if (glyph >= GET_BE_WORD(cf1->StartGlyph))
{
int index = glyph - GET_BE_WORD(cf1->StartGlyph);
if (index < GET_BE_WORD(cf1->GlyphCount))
class = GET_BE_WORD(cf1->ClassValueArray[index]);
}
}
else if (GET_BE_WORD(cf1->ClassFormat) == 2)
{
const OT_ClassDefFormat2 *cf2 = table;
int i, top;
top = GET_BE_WORD(cf2->ClassRangeCount);
for (i = 0; i < top; i++)
{
if (glyph >= GET_BE_WORD(cf2->ClassRangeRecord[i].Start) &&
glyph <= GET_BE_WORD(cf2->ClassRangeRecord[i].End))
{
class = GET_BE_WORD(cf2->ClassRangeRecord[i].Class);
break;
}
}
}
else
ERR("Unknown Class Format %i\n",GET_BE_WORD(cf1->ClassFormat));
return class;
}
void OpenType_GDEF_UpdateGlyphProps(ScriptCache *psc, const WORD *pwGlyphs, const WORD cGlyphs, WORD* pwLogClust, const WORD cChars, SCRIPT_GLYPHPROP *pGlyphProp)
{
int i;
void *glyph_class_table = NULL;
if (psc->GDEF_Table)
{
const GDEF_Header *header = psc->GDEF_Table;
WORD offset = GET_BE_WORD( header->GlyphClassDef );
if (offset)
glyph_class_table = (BYTE *)psc->GDEF_Table + offset;
}
for (i = 0; i < cGlyphs; i++)
{
WORD class;
int char_count = 0;
int k;
k = USP10_FindGlyphInLogClust(pwLogClust, cChars, i);
if (k >= 0)
{
for (; k < cChars && pwLogClust[k] == i; k++)
char_count++;
}
class = OT_get_glyph_class( glyph_class_table, pwGlyphs[i] );
switch (class)
{
case 0:
case BaseGlyph:
pGlyphProp[i].sva.fClusterStart = 1;
pGlyphProp[i].sva.fDiacritic = 0;
pGlyphProp[i].sva.fZeroWidth = 0;
break;
case LigatureGlyph:
pGlyphProp[i].sva.fClusterStart = 1;
pGlyphProp[i].sva.fDiacritic = 0;
pGlyphProp[i].sva.fZeroWidth = 0;
break;
case MarkGlyph:
pGlyphProp[i].sva.fClusterStart = 0;
pGlyphProp[i].sva.fDiacritic = 1;
pGlyphProp[i].sva.fZeroWidth = 1;
break;
case ComponentGlyph:
pGlyphProp[i].sva.fClusterStart = 0;
pGlyphProp[i].sva.fDiacritic = 0;
pGlyphProp[i].sva.fZeroWidth = 0;
break;
default:
ERR("Unknown glyph class %i\n",class);
pGlyphProp[i].sva.fClusterStart = 1;
pGlyphProp[i].sva.fDiacritic = 0;
pGlyphProp[i].sva.fZeroWidth = 0;
}
if (char_count == 0)
pGlyphProp[i].sva.fClusterStart = 0;
}
}
/**********
* GSUB
**********/
static INT GSUB_apply_lookup(const OT_LookupList* lookup, INT lookup_index, WORD *glyphs, INT glyph_index, INT write_dir, INT *glyph_count);
static INT GSUB_is_glyph_covered(LPCVOID table , UINT glyph)
{
const OT_CoverageFormat1* cf1;
cf1 = table;
if (GET_BE_WORD(cf1->CoverageFormat) == 1)
{
int count = GET_BE_WORD(cf1->GlyphCount);
int i;
TRACE("Coverage Format 1, %i glyphs\n",count);
for (i = 0; i < count; i++)
if (glyph == GET_BE_WORD(cf1->GlyphArray[i]))
return i;
return -1;
}
else if (GET_BE_WORD(cf1->CoverageFormat) == 2)
{
const OT_CoverageFormat2* cf2;
int i;
int count;
cf2 = (const OT_CoverageFormat2*)cf1;
count = GET_BE_WORD(cf2->RangeCount);
TRACE("Coverage Format 2, %i ranges\n",count);
for (i = 0; i < count; i++)
{
if (glyph < GET_BE_WORD(cf2->RangeRecord[i].Start))
return -1;
if ((glyph >= GET_BE_WORD(cf2->RangeRecord[i].Start)) &&
(glyph <= GET_BE_WORD(cf2->RangeRecord[i].End)))
{
return (GET_BE_WORD(cf2->RangeRecord[i].StartCoverageIndex) +
glyph - GET_BE_WORD(cf2->RangeRecord[i].Start));
}
}
return -1;
}
else
ERR("Unknown CoverageFormat %i\n",GET_BE_WORD(cf1->CoverageFormat));
return -1;
}
static const BYTE *GSUB_get_subtable(const OT_LookupTable *look, int index)
{
int offset = GET_BE_WORD(look->SubTable[index]);
if (GET_BE_WORD(look->LookupType) == 7)
{
const GSUB_ExtensionPosFormat1 *ext = (const GSUB_ExtensionPosFormat1 *)((const BYTE *)look + offset);
if (GET_BE_WORD(ext->SubstFormat) == 1)
{
offset += GET_BE_DWORD(ext->ExtensionOffset);
}
else
{
FIXME("Unhandled Extension Substitution Format %i\n",GET_BE_WORD(ext->SubstFormat));
}
}
return (const BYTE *)look + offset;
}
static INT GSUB_apply_SingleSubst(const OT_LookupTable *look, WORD *glyphs, INT glyph_index, INT write_dir, INT *glyph_count)
{
int j;
TRACE("Single Substitution Subtable\n");
for (j = 0; j < GET_BE_WORD(look->SubTableCount); j++)
{
const GSUB_SingleSubstFormat1 *ssf1 = (const GSUB_SingleSubstFormat1*)GSUB_get_subtable(look, j);
if (GET_BE_WORD(ssf1->SubstFormat) == 1)
{
int offset = GET_BE_WORD(ssf1->Coverage);
TRACE(" subtype 1, delta %i\n", GET_BE_WORD(ssf1->DeltaGlyphID));
if (GSUB_is_glyph_covered((const BYTE*)ssf1+offset, glyphs[glyph_index]) != -1)
{
TRACE(" Glyph 0x%x ->",glyphs[glyph_index]);
glyphs[glyph_index] = glyphs[glyph_index] + GET_BE_WORD(ssf1->DeltaGlyphID);
TRACE(" 0x%x\n",glyphs[glyph_index]);
return glyph_index + write_dir;
}
}
else
{
const GSUB_SingleSubstFormat2 *ssf2;
INT index;
INT offset;
ssf2 = (const GSUB_SingleSubstFormat2 *)ssf1;
offset = GET_BE_WORD(ssf1->Coverage);
TRACE(" subtype 2, glyph count %i\n", GET_BE_WORD(ssf2->GlyphCount));
index = GSUB_is_glyph_covered((const BYTE*)ssf2+offset, glyphs[glyph_index]);
TRACE(" Coverage index %i\n",index);
if (index != -1)
{
if (glyphs[glyph_index] == GET_BE_WORD(ssf2->Substitute[index]))
return GSUB_E_NOGLYPH;
TRACE(" Glyph is 0x%x ->",glyphs[glyph_index]);
glyphs[glyph_index] = GET_BE_WORD(ssf2->Substitute[index]);
TRACE("0x%x\n",glyphs[glyph_index]);
return glyph_index + write_dir;
}
}
}
return GSUB_E_NOGLYPH;
}
static INT GSUB_apply_MultipleSubst(const OT_LookupTable *look, WORD *glyphs, INT glyph_index, INT write_dir, INT *glyph_count)
{
int j;
TRACE("Multiple Substitution Subtable\n");
for (j = 0; j < GET_BE_WORD(look->SubTableCount); j++)
{
int offset, index;
const GSUB_MultipleSubstFormat1 *msf1;
msf1 = (const GSUB_MultipleSubstFormat1*)GSUB_get_subtable(look, j);
offset = GET_BE_WORD(msf1->Coverage);
index = GSUB_is_glyph_covered((const BYTE*)msf1+offset, glyphs[glyph_index]);
if (index != -1)
{
const GSUB_Sequence *seq;
int sub_count;
int j;
offset = GET_BE_WORD(msf1->Sequence[index]);
seq = (const GSUB_Sequence*)((const BYTE*)msf1+offset);
sub_count = GET_BE_WORD(seq->GlyphCount);
TRACE(" Glyph 0x%x (+%i)->",glyphs[glyph_index],(sub_count-1));
for (j = (*glyph_count)+(sub_count-1); j > glyph_index; j--)
glyphs[j] =glyphs[j-(sub_count-1)];
for (j = 0; j < sub_count; j++)
if (write_dir < 0)
glyphs[glyph_index + (sub_count-1) - j] = GET_BE_WORD(seq->Substitute[j]);
else
glyphs[glyph_index + j] = GET_BE_WORD(seq->Substitute[j]);
*glyph_count = *glyph_count + (sub_count - 1);
if (TRACE_ON(uniscribe))
{
for (j = 0; j < sub_count; j++)
TRACE(" 0x%x",glyphs[glyph_index+j]);
TRACE("\n");
}
return glyph_index + (sub_count * write_dir);
}
}
return GSUB_E_NOGLYPH;
}
static INT GSUB_apply_AlternateSubst(const OT_LookupTable *look, WORD *glyphs, INT glyph_index, INT write_dir, INT *glyph_count)
{
int j;
TRACE("Alternate Substitution Subtable\n");
for (j = 0; j < GET_BE_WORD(look->SubTableCount); j++)
{
int offset;
const GSUB_AlternateSubstFormat1 *asf1;
INT index;
asf1 = (const GSUB_AlternateSubstFormat1*)GSUB_get_subtable(look, j);
offset = GET_BE_WORD(asf1->Coverage);
index = GSUB_is_glyph_covered((const BYTE*)asf1+offset, glyphs[glyph_index]);
if (index != -1)
{
const GSUB_AlternateSet *as;
offset = GET_BE_WORD(asf1->AlternateSet[index]);
as = (const GSUB_AlternateSet*)((const BYTE*)asf1+offset);
FIXME("%i alternates, picking index 0\n",GET_BE_WORD(as->GlyphCount));
if (glyphs[glyph_index] == GET_BE_WORD(as->Alternate[0]))
return GSUB_E_NOGLYPH;
TRACE(" Glyph 0x%x ->",glyphs[glyph_index]);
glyphs[glyph_index] = GET_BE_WORD(as->Alternate[0]);
TRACE(" 0x%x\n",glyphs[glyph_index]);
return glyph_index + write_dir;
}
}
return GSUB_E_NOGLYPH;
}
static INT GSUB_apply_LigatureSubst(const OT_LookupTable *look, WORD *glyphs, INT glyph_index, INT write_dir, INT *glyph_count)
{
int j;
TRACE("Ligature Substitution Subtable\n");
for (j = 0; j < GET_BE_WORD(look->SubTableCount); j++)
{
const GSUB_LigatureSubstFormat1 *lsf1;
int offset,index;
lsf1 = (const GSUB_LigatureSubstFormat1*)GSUB_get_subtable(look, j);
offset = GET_BE_WORD(lsf1->Coverage);
index = GSUB_is_glyph_covered((const BYTE*)lsf1+offset, glyphs[glyph_index]);
TRACE(" Coverage index %i\n",index);
if (index != -1)
{
const GSUB_LigatureSet *ls;
int k, count;
offset = GET_BE_WORD(lsf1->LigatureSet[index]);
ls = (const GSUB_LigatureSet*)((const BYTE*)lsf1+offset);
count = GET_BE_WORD(ls->LigatureCount);
TRACE(" LigatureSet has %i members\n",count);
for (k = 0; k < count; k++)
{
const GSUB_Ligature *lig;
int CompCount,l,CompIndex;
offset = GET_BE_WORD(ls->Ligature[k]);
lig = (const GSUB_Ligature*)((const BYTE*)ls+offset);
CompCount = GET_BE_WORD(lig->CompCount) - 1;
CompIndex = glyph_index+write_dir;
for (l = 0; l < CompCount && CompIndex >= 0 && CompIndex < *glyph_count; l++)
{
int CompGlyph;
CompGlyph = GET_BE_WORD(lig->Component[l]);
if (CompGlyph != glyphs[CompIndex])
break;
CompIndex += write_dir;
}
if (l == CompCount)
{
int replaceIdx = glyph_index;
if (write_dir < 0)
replaceIdx = glyph_index - CompCount;
TRACE(" Glyph is 0x%x (+%i) ->",glyphs[glyph_index],CompCount);
glyphs[replaceIdx] = GET_BE_WORD(lig->LigGlyph);
TRACE("0x%x\n",glyphs[replaceIdx]);
if (CompCount > 0)
{
int j;
for (j = replaceIdx + 1; j < *glyph_count; j++)
glyphs[j] =glyphs[j+CompCount];
*glyph_count = *glyph_count - CompCount;
}
return replaceIdx + write_dir;
}
}
}
}
return GSUB_E_NOGLYPH;
}
static INT GSUB_apply_ChainContextSubst(const OT_LookupList* lookup, const OT_LookupTable *look, WORD *glyphs, INT glyph_index, INT write_dir, INT *glyph_count)
{
int j;
TRACE("Chaining Contextual Substitution Subtable\n");
for (j = 0; j < GET_BE_WORD(look->SubTableCount); j++)
{
const GSUB_ChainContextSubstFormat1 *ccsf1;
int offset;
int dirLookahead = write_dir;
int dirBacktrack = -1 * write_dir;
ccsf1 = (const GSUB_ChainContextSubstFormat1*)GSUB_get_subtable(look, j);
if (GET_BE_WORD(ccsf1->SubstFormat) == 1)
{
static int once;
if (!once++)
FIXME(" TODO: subtype 1 (Simple context glyph substitution)\n");
continue;
}
else if (GET_BE_WORD(ccsf1->SubstFormat) == 2)
{
static int once;
if (!once++)
FIXME(" TODO: subtype 2 (Class-based Chaining Context Glyph Substitution)\n");
continue;
}
else if (GET_BE_WORD(ccsf1->SubstFormat) == 3)
{
int k;
int indexGlyphs;
const GSUB_ChainContextSubstFormat3_1 *ccsf3_1;
const GSUB_ChainContextSubstFormat3_2 *ccsf3_2;
const GSUB_ChainContextSubstFormat3_3 *ccsf3_3;
const GSUB_ChainContextSubstFormat3_4 *ccsf3_4;
int newIndex = glyph_index;
ccsf3_1 = (const GSUB_ChainContextSubstFormat3_1 *)ccsf1;
TRACE(" subtype 3 (Coverage-based Chaining Context Glyph Substitution)\n");
for (k = 0; k < GET_BE_WORD(ccsf3_1->BacktrackGlyphCount); k++)
{
offset = GET_BE_WORD(ccsf3_1->Coverage[k]);
if (GSUB_is_glyph_covered((const BYTE*)ccsf3_1+offset, glyphs[glyph_index + (dirBacktrack * (k+1))]) == -1)
break;
}
if (k != GET_BE_WORD(ccsf3_1->BacktrackGlyphCount))
continue;
TRACE("Matched Backtrack\n");
ccsf3_2 = (const GSUB_ChainContextSubstFormat3_2 *)((BYTE *)ccsf1 +
FIELD_OFFSET(GSUB_ChainContextSubstFormat3_1, Coverage[GET_BE_WORD(ccsf3_1->BacktrackGlyphCount)]));
indexGlyphs = GET_BE_WORD(ccsf3_2->InputGlyphCount);
for (k = 0; k < indexGlyphs; k++)
{
offset = GET_BE_WORD(ccsf3_2->Coverage[k]);
if (GSUB_is_glyph_covered((const BYTE*)ccsf3_1+offset, glyphs[glyph_index + (write_dir * k)]) == -1)
break;
}
if (k != indexGlyphs)
continue;
TRACE("Matched IndexGlyphs\n");
ccsf3_3 = (const GSUB_ChainContextSubstFormat3_3 *)((BYTE *)ccsf3_2 +
FIELD_OFFSET(GSUB_ChainContextSubstFormat3_2, Coverage[GET_BE_WORD(ccsf3_2->InputGlyphCount)]));
for (k = 0; k < GET_BE_WORD(ccsf3_3->LookaheadGlyphCount); k++)
{
offset = GET_BE_WORD(ccsf3_3->Coverage[k]);
if (GSUB_is_glyph_covered((const BYTE*)ccsf3_1+offset, glyphs[glyph_index + (dirLookahead * (indexGlyphs + k))]) == -1)
break;
}
if (k != GET_BE_WORD(ccsf3_3->LookaheadGlyphCount))
continue;
TRACE("Matched LookAhead\n");
ccsf3_4 = (const GSUB_ChainContextSubstFormat3_4 *)((BYTE *)ccsf3_3 +
FIELD_OFFSET(GSUB_ChainContextSubstFormat3_3, Coverage[GET_BE_WORD(ccsf3_3->LookaheadGlyphCount)]));
if (GET_BE_WORD(ccsf3_4->SubstCount))
{
for (k = 0; k < GET_BE_WORD(ccsf3_4->SubstCount); k++)
{
int lookupIndex = GET_BE_WORD(ccsf3_4->SubstLookupRecord[k].LookupListIndex);
int SequenceIndex = GET_BE_WORD(ccsf3_4->SubstLookupRecord[k].SequenceIndex) * write_dir;
TRACE("SUBST: %i -> %i %i\n",k, SequenceIndex, lookupIndex);
newIndex = GSUB_apply_lookup(lookup, lookupIndex, glyphs, glyph_index + SequenceIndex, write_dir, glyph_count);
if (newIndex == -1)
{
ERR("Chain failed to generate a glyph\n");
continue;
}
}
return newIndex;
}
else return GSUB_E_NOGLYPH;
}
}
return -1;
}
static INT GSUB_apply_lookup(const OT_LookupList* lookup, INT lookup_index, WORD *glyphs, INT glyph_index, INT write_dir, INT *glyph_count)
{
int offset;
int type;
const OT_LookupTable *look;
offset = GET_BE_WORD(lookup->Lookup[lookup_index]);
look = (const OT_LookupTable*)((const BYTE*)lookup + offset);
type = GET_BE_WORD(look->LookupType);
TRACE("type %i, flag %x, subtables %i\n",type,GET_BE_WORD(look->LookupFlag),GET_BE_WORD(look->SubTableCount));
if (type == 7)
{
if (GET_BE_WORD(look->SubTableCount))
{
const GSUB_ExtensionPosFormat1 *ext = (const GSUB_ExtensionPosFormat1 *)((const BYTE *)look + GET_BE_WORD(look->SubTable[0]));
if (GET_BE_WORD(ext->SubstFormat) == 1)
{
type = GET_BE_WORD(ext->ExtensionLookupType);
TRACE("extension type %i\n",type);
}
else
{
FIXME("Unhandled Extension Substitution Format %i\n",GET_BE_WORD(ext->SubstFormat));
}
}
else
{
WARN("lookup type is Extension Substitution but no extension subtable exists\n");
}
}
switch(type)
{
case 1:
return GSUB_apply_SingleSubst(look, glyphs, glyph_index, write_dir, glyph_count);
case 2:
return GSUB_apply_MultipleSubst(look, glyphs, glyph_index, write_dir, glyph_count);
case 3:
return GSUB_apply_AlternateSubst(look, glyphs, glyph_index, write_dir, glyph_count);
case 4:
return GSUB_apply_LigatureSubst(look, glyphs, glyph_index, write_dir, glyph_count);
case 6:
return GSUB_apply_ChainContextSubst(lookup, look, glyphs, glyph_index, write_dir, glyph_count);
case 7:
FIXME("Extension Substitution types not valid here\n");
break;
default:
FIXME("We do not handle SubType %i\n",type);
}
return GSUB_E_NOGLYPH;
}
INT OpenType_apply_GSUB_lookup(LPCVOID table, INT lookup_index, WORD *glyphs, INT glyph_index, INT write_dir, INT *glyph_count)
{
const GSUB_Header *header = (const GSUB_Header *)table;
const OT_LookupList *lookup = (const OT_LookupList*)((const BYTE*)header + GET_BE_WORD(header->LookupList));
return GSUB_apply_lookup(lookup, lookup_index, glyphs, glyph_index, write_dir, glyph_count);
}
/**********
* GPOS
**********/
static INT GPOS_apply_lookup(ScriptCache *psc, LPOUTLINETEXTMETRICW lpotm, LPLOGFONTW lplogfont, const SCRIPT_ANALYSIS *analysis, INT* piAdvance,
const OT_LookupList* lookup, INT lookup_index, const WORD *glyphs, INT glyph_index, INT glyph_count, GOFFSET *pGoffset);
static INT GPOS_get_device_table_value(const OT_DeviceTable *DeviceTable, WORD ppem)
{
static const WORD mask[3] = {3,0xf,0xff};
if (DeviceTable && ppem >= GET_BE_WORD(DeviceTable->StartSize) && ppem <= GET_BE_WORD(DeviceTable->EndSize))
{
int format = GET_BE_WORD(DeviceTable->DeltaFormat);
int index = ppem - GET_BE_WORD(DeviceTable->StartSize);
int value;
TRACE("device table, format %i, index %i\n",format, index);
index = index << format;
value = (DeviceTable->DeltaValue[index/sizeof(WORD)] << (index%sizeof(WORD)))&mask[format-1];
TRACE("offset %i, value %i\n",index, value);
if (value > mask[format-1]/2)
value = -1 * ((mask[format-1]+1) - value);
return value;
}
return 0;
}
static VOID GPOS_get_anchor_values(LPCVOID table, LPPOINT pt, WORD ppem)
{
const GPOS_AnchorFormat1* anchor1 = (const GPOS_AnchorFormat1*)table;
switch (GET_BE_WORD(anchor1->AnchorFormat))
{
case 1:
{
TRACE("Anchor Format 1\n");
pt->x = (short)GET_BE_WORD(anchor1->XCoordinate);
pt->y = (short)GET_BE_WORD(anchor1->YCoordinate);
break;
}
case 2:
{
const GPOS_AnchorFormat2* anchor2 = (const GPOS_AnchorFormat2*)table;
TRACE("Anchor Format 2\n");
pt->x = (short)GET_BE_WORD(anchor2->XCoordinate);
pt->y = (short)GET_BE_WORD(anchor2->YCoordinate);
break;
}
case 3:
{
int offset;
const GPOS_AnchorFormat3* anchor3 = (const GPOS_AnchorFormat3*)table;
TRACE("Anchor Format 3\n");
pt->x = (short)GET_BE_WORD(anchor3->XCoordinate);
pt->y = (short)GET_BE_WORD(anchor3->YCoordinate);
offset = GET_BE_WORD(anchor3->XDeviceTable);
TRACE("ppem %i\n",ppem);
if (offset)
{
const OT_DeviceTable* DeviceTableX = NULL;
DeviceTableX = (const OT_DeviceTable*)((const BYTE*)anchor3 + offset);
pt->x += GPOS_get_device_table_value(DeviceTableX, ppem);
}
offset = GET_BE_WORD(anchor3->YDeviceTable);
if (offset)
{
const OT_DeviceTable* DeviceTableY = NULL;
DeviceTableY = (const OT_DeviceTable*)((const BYTE*)anchor3 + offset);
pt->y += GPOS_get_device_table_value(DeviceTableY, ppem);
}
break;
}
default:
ERR("Unknown Anchor Format %i\n",GET_BE_WORD(anchor1->AnchorFormat));
pt->x = 0;
pt->y = 0;
}
}
static void GPOS_convert_design_units_to_device(LPOUTLINETEXTMETRICW lpotm, LPLOGFONTW lplogfont, int desX, int desY, double *devX, double *devY)
{
int emHeight = lpotm->otmTextMetrics.tmAscent + lpotm->otmTextMetrics.tmDescent - lpotm->otmTextMetrics.tmInternalLeading;
TRACE("emHeight %i lfWidth %i\n",emHeight, lplogfont->lfWidth);
*devX = (desX * emHeight) / (double)lpotm->otmEMSquare;
*devY = (desY * emHeight) / (double)lpotm->otmEMSquare;
if (lplogfont->lfWidth)
FIXME("Font with lfWidth set not handled properly\n");
}
static INT GPOS_get_value_record(WORD ValueFormat, const WORD data[], GPOS_ValueRecord *record)
{
INT offset = 0;
if (ValueFormat & 0x0001) { if (data) record->XPlacement = GET_BE_WORD(data[offset]); offset++; }
if (ValueFormat & 0x0002) { if (data) record->YPlacement = GET_BE_WORD(data[offset]); offset++; }
if (ValueFormat & 0x0004) { if (data) record->XAdvance = GET_BE_WORD(data[offset]); offset++; }
if (ValueFormat & 0x0008) { if (data) record->YAdvance = GET_BE_WORD(data[offset]); offset++; }
if (ValueFormat & 0x0010) { if (data) record->XPlaDevice = GET_BE_WORD(data[offset]); offset++; }
if (ValueFormat & 0x0020) { if (data) record->YPlaDevice = GET_BE_WORD(data[offset]); offset++; }
if (ValueFormat & 0x0040) { if (data) record->XAdvDevice = GET_BE_WORD(data[offset]); offset++; }
if (ValueFormat & 0x0080) { if (data) record->YAdvDevice = GET_BE_WORD(data[offset]); offset++; }
return offset;
}
static VOID GPOS_get_value_record_offsets(const BYTE* head, GPOS_ValueRecord *ValueRecord, WORD ValueFormat, INT ppem, LPPOINT ptPlacement, LPPOINT ptAdvance)
{
if (ValueFormat & 0x0001) ptPlacement->x += (short)ValueRecord->XPlacement;
if (ValueFormat & 0x0002) ptPlacement->y += (short)ValueRecord->YPlacement;
if (ValueFormat & 0x0004) ptAdvance->x += (short)ValueRecord->XAdvance;
if (ValueFormat & 0x0008) ptAdvance->y += (short)ValueRecord->YAdvance;
if (ValueFormat & 0x0010) ptPlacement->x += GPOS_get_device_table_value((const OT_DeviceTable*)(head + ValueRecord->XPlaDevice), ppem);
if (ValueFormat & 0x0020) ptPlacement->y += GPOS_get_device_table_value((const OT_DeviceTable*)(head + ValueRecord->YPlaDevice), ppem);
if (ValueFormat & 0x0040) ptAdvance->x += GPOS_get_device_table_value((const OT_DeviceTable*)(head + ValueRecord->XAdvDevice), ppem);
if (ValueFormat & 0x0080) ptAdvance->y += GPOS_get_device_table_value((const OT_DeviceTable*)(head + ValueRecord->YAdvDevice), ppem);
if (ValueFormat & 0xFF00) FIXME("Unhandled Value Format %x\n",ValueFormat&0xFF00);
}
static const BYTE *GPOS_get_subtable(const OT_LookupTable *look, int index)
{
int offset = GET_BE_WORD(look->SubTable[index]);
if (GET_BE_WORD(look->LookupType) == 9)
{
const GPOS_ExtensionPosFormat1 *ext = (const GPOS_ExtensionPosFormat1 *)((const BYTE *)look + offset);
if (GET_BE_WORD(ext->PosFormat) == 1)
{
offset += GET_BE_DWORD(ext->ExtensionOffset);
}
else
{
FIXME("Unhandled Extension Positioning Format %i\n",GET_BE_WORD(ext->PosFormat));
}
}
return (const BYTE *)look + offset;
}
static VOID GPOS_apply_SingleAdjustment(const OT_LookupTable *look, const SCRIPT_ANALYSIS *analysis, const WORD *glyphs, INT glyph_index,
INT glyph_count, INT ppem, LPPOINT ptAdjust, LPPOINT ptAdvance)
{
int j;
TRACE("Single Adjustment Positioning Subtable\n");
for (j = 0; j < GET_BE_WORD(look->SubTableCount); j++)
{
const GPOS_SinglePosFormat1 *spf1 = (const GPOS_SinglePosFormat1*)GPOS_get_subtable(look, j);
WORD offset;
if (GET_BE_WORD(spf1->PosFormat) == 1)
{
offset = GET_BE_WORD(spf1->Coverage);
if (GSUB_is_glyph_covered((const BYTE*)spf1+offset, glyphs[glyph_index]) != -1)
{
GPOS_ValueRecord ValueRecord = {0,0,0,0,0,0,0,0};
WORD ValueFormat = GET_BE_WORD(spf1->ValueFormat);
GPOS_get_value_record(ValueFormat, spf1->Value, &ValueRecord);
GPOS_get_value_record_offsets((const BYTE*)spf1, &ValueRecord, ValueFormat, ppem, ptAdjust, ptAdvance);
TRACE("Glyph Adjusted by %i,%i\n",ValueRecord.XPlacement,ValueRecord.YPlacement);
}
}
else if (GET_BE_WORD(spf1->PosFormat) == 2)
{
int index;
const GPOS_SinglePosFormat2 *spf2;
spf2 = (const GPOS_SinglePosFormat2*)spf1;
offset = GET_BE_WORD(spf2->Coverage);
index = GSUB_is_glyph_covered((const BYTE*)spf2+offset, glyphs[glyph_index]);
if (index != -1)
{
int size;
GPOS_ValueRecord ValueRecord = {0,0,0,0,0,0,0,0};
WORD ValueFormat = GET_BE_WORD(spf2->ValueFormat);
size = GPOS_get_value_record(ValueFormat, spf2->Value, &ValueRecord);
if (index > 0)
{
offset = size * index;
GPOS_get_value_record(ValueFormat, &spf2->Value[offset], &ValueRecord);
}
GPOS_get_value_record_offsets((const BYTE*)spf2, &ValueRecord, ValueFormat, ppem, ptAdjust, ptAdvance);
TRACE("Glyph Adjusted by %i,%i\n",ValueRecord.XPlacement,ValueRecord.YPlacement);
}
}
else
FIXME("Single Adjustment Positioning: Format %i Unhandled\n",GET_BE_WORD(spf1->PosFormat));
}
}
static void apply_pair_value( const void *pos_table, WORD val_fmt1, WORD val_fmt2, const WORD *pair,
INT ppem, POINT *adjust, POINT *advance )
{
GPOS_ValueRecord val_rec1 = {0,0,0,0,0,0,0,0};
GPOS_ValueRecord val_rec2 = {0,0,0,0,0,0,0,0};
INT size;
size = GPOS_get_value_record( val_fmt1, pair, &val_rec1 );
GPOS_get_value_record( val_fmt2, pair + size, &val_rec2 );
if (val_fmt1)
{
GPOS_get_value_record_offsets( pos_table, &val_rec1, val_fmt1, ppem, adjust, advance );
TRACE( "Glyph 1 resulting cumulative offset is %i,%i design units\n", adjust[0].x, adjust[0].y );
TRACE( "Glyph 1 resulting cumulative advance is %i,%i design units\n", advance[0].x, advance[0].y );
}
if (val_fmt2)
{
GPOS_get_value_record_offsets( pos_table, &val_rec2, val_fmt2, ppem, adjust + 1, advance + 1 );
TRACE( "Glyph 2 resulting cumulative offset is %i,%i design units\n", adjust[1].x, adjust[1].y );
TRACE( "Glyph 2 resulting cumulative advance is %i,%i design units\n", advance[1].x, advance[1].y );
}
}
static INT GPOS_apply_PairAdjustment(const OT_LookupTable *look, const SCRIPT_ANALYSIS *analysis, const WORD *glyphs, INT glyph_index,
INT glyph_count, INT ppem, LPPOINT ptAdjust, LPPOINT ptAdvance)
{
int j;
int write_dir = (analysis->fRTL && !analysis->fLogicalOrder) ? -1 : 1;
if (glyph_index + write_dir < 0 || glyph_index + write_dir >= glyph_count) return glyph_index + 1;
TRACE("Pair Adjustment Positioning Subtable\n");
for (j = 0; j < GET_BE_WORD(look->SubTableCount); j++)
{
const GPOS_PairPosFormat1 *ppf1 = (const GPOS_PairPosFormat1*)GPOS_get_subtable(look, j);
WORD offset;
if (GET_BE_WORD(ppf1->PosFormat) == 1)
{
int index;
WORD ValueFormat1 = GET_BE_WORD(ppf1->ValueFormat1);
WORD ValueFormat2 = GET_BE_WORD(ppf1->ValueFormat2);
INT val_fmt1_size = GPOS_get_value_record( ValueFormat1, NULL, NULL );
INT val_fmt2_size = GPOS_get_value_record( ValueFormat2, NULL, NULL );
offset = GET_BE_WORD(ppf1->Coverage);
index = GSUB_is_glyph_covered((const BYTE*)ppf1+offset, glyphs[glyph_index]);
if (index != -1 && index < GET_BE_WORD(ppf1->PairSetCount))
{
int k;
int pair_count;
const GPOS_PairSet *ps;
const GPOS_PairValueRecord *pair_val_rec;
offset = GET_BE_WORD(ppf1->PairSetOffset[index]);
ps = (const GPOS_PairSet*)((const BYTE*)ppf1+offset);
pair_count = GET_BE_WORD(ps->PairValueCount);
pair_val_rec = ps->PairValueRecord;
for (k = 0; k < pair_count; k++)
{
WORD second_glyph = GET_BE_WORD(pair_val_rec->SecondGlyph);
if (glyphs[glyph_index+write_dir] == second_glyph)
{
int next = 1;
TRACE("Format 1: Found Pair %x,%x\n",glyphs[glyph_index],glyphs[glyph_index+write_dir]);
apply_pair_value( ppf1, ValueFormat1, ValueFormat2, pair_val_rec->Value1, ppem, ptAdjust, ptAdvance );
if (ValueFormat2) next++;
return glyph_index + next;
}
pair_val_rec = (const GPOS_PairValueRecord *)(pair_val_rec->Value1 + val_fmt1_size + val_fmt2_size);
}
}
}
else if (GET_BE_WORD(ppf1->PosFormat) == 2)
{
const GPOS_PairPosFormat2 *ppf2 = (const GPOS_PairPosFormat2*)ppf1;
int index;
WORD ValueFormat1 = GET_BE_WORD( ppf2->ValueFormat1 );
WORD ValueFormat2 = GET_BE_WORD( ppf2->ValueFormat2 );
INT val_fmt1_size = GPOS_get_value_record( ValueFormat1, NULL, NULL );
INT val_fmt2_size = GPOS_get_value_record( ValueFormat2, NULL, NULL );
WORD class1_count = GET_BE_WORD( ppf2->Class1Count );
WORD class2_count = GET_BE_WORD( ppf2->Class2Count );
offset = GET_BE_WORD( ppf2->Coverage );
index = GSUB_is_glyph_covered( (const BYTE*)ppf2 + offset, glyphs[glyph_index] );
if (index != -1)
{
WORD class1, class2;
class1 = OT_get_glyph_class( (const BYTE *)ppf2 + GET_BE_WORD(ppf2->ClassDef1), glyphs[glyph_index] );
class2 = OT_get_glyph_class( (const BYTE *)ppf2 + GET_BE_WORD(ppf2->ClassDef2), glyphs[glyph_index + write_dir] );
if (class1 < class1_count && class2 < class2_count)
{
const WORD *pair_val = ppf2->Class1Record + (class1 * class2_count + class2) * (val_fmt1_size + val_fmt2_size);
int next = 1;
TRACE( "Format 2: Found Pair %x,%x\n", glyphs[glyph_index], glyphs[glyph_index + write_dir] );
apply_pair_value( ppf2, ValueFormat1, ValueFormat2, pair_val, ppem, ptAdjust, ptAdvance );
if (ValueFormat2) next++;
return glyph_index + next;
}
}
}
else
FIXME("Pair Adjustment Positioning: Format %i Unhandled\n",GET_BE_WORD(ppf1->PosFormat));
}
return glyph_index+1;
}
static VOID GPOS_apply_CursiveAttachment(const OT_LookupTable *look, const SCRIPT_ANALYSIS *analysis, const WORD *glyphs, INT glyph_index,
INT glyph_count, INT ppem, LPPOINT pt)
{
int j;
int write_dir = (analysis->fRTL && !analysis->fLogicalOrder) ? -1 : 1;
if (glyph_index + write_dir < 0 || glyph_index + write_dir >= glyph_count) return;
TRACE("Cursive Attachment Positioning Subtable\n");
for (j = 0; j < GET_BE_WORD(look->SubTableCount); j++)
{
const GPOS_CursivePosFormat1 *cpf1 = (const GPOS_CursivePosFormat1 *)GPOS_get_subtable(look, j);
if (GET_BE_WORD(cpf1->PosFormat) == 1)
{
int index_exit, index_entry;
WORD offset = GET_BE_WORD( cpf1->Coverage );
index_exit = GSUB_is_glyph_covered((const BYTE*)cpf1+offset, glyphs[glyph_index]);
if (index_exit != -1 && cpf1->EntryExitRecord[index_exit].ExitAnchor!= 0)
{
index_entry = GSUB_is_glyph_covered((const BYTE*)cpf1+offset, glyphs[glyph_index+write_dir]);
if (index_entry != -1 && cpf1->EntryExitRecord[index_entry].EntryAnchor != 0)
{
POINT exit_pt, entry_pt;
offset = GET_BE_WORD(cpf1->EntryExitRecord[index_exit].ExitAnchor);
GPOS_get_anchor_values((const BYTE*)cpf1 + offset, &exit_pt, ppem);
offset = GET_BE_WORD(cpf1->EntryExitRecord[index_entry].EntryAnchor);
GPOS_get_anchor_values((const BYTE*)cpf1 + offset, &entry_pt, ppem);
TRACE("Found linkage %x[%i,%i] %x[%i,%i]\n",glyphs[glyph_index], exit_pt.x,exit_pt.y, glyphs[glyph_index+write_dir], entry_pt.x, entry_pt.y);
pt->x = entry_pt.x - exit_pt.x;
pt->y = entry_pt.y - exit_pt.y;
return;
}
}
}
else
FIXME("Cursive Attachment Positioning: Format %i Unhandled\n",GET_BE_WORD(cpf1->PosFormat));
}
return;
}
static int GPOS_apply_MarkToBase(ScriptCache *psc, const OT_LookupTable *look, const SCRIPT_ANALYSIS *analysis, const WORD *glyphs, INT glyph_index, INT glyph_count, INT ppem, LPPOINT pt)
{
int j;
int write_dir = (analysis->fRTL && !analysis->fLogicalOrder) ? -1 : 1;
void *glyph_class_table = NULL;
int rc = -1;
if (psc->GDEF_Table)
{
const GDEF_Header *header = psc->GDEF_Table;
WORD offset = GET_BE_WORD( header->GlyphClassDef );
if (offset)
glyph_class_table = (BYTE *)psc->GDEF_Table + offset;
}
TRACE("MarkToBase Attachment Positioning Subtable\n");
for (j = 0; j < GET_BE_WORD(look->SubTableCount); j++)
{
const GPOS_MarkBasePosFormat1 *mbpf1 = (const GPOS_MarkBasePosFormat1 *)GPOS_get_subtable(look, j);
if (GET_BE_WORD(mbpf1->PosFormat) == 1)
{
int offset = GET_BE_WORD(mbpf1->MarkCoverage);
int mark_index;
mark_index = GSUB_is_glyph_covered((const BYTE*)mbpf1+offset, glyphs[glyph_index]);
if (mark_index != -1)
{
int base_index;
int base_glyph = glyph_index - write_dir;
if (glyph_class_table)
{
while (OT_get_glyph_class(glyph_class_table, glyphs[base_glyph]) == MarkGlyph && base_glyph > 0 && base_glyph < glyph_count)
base_glyph -= write_dir;
}
offset = GET_BE_WORD(mbpf1->BaseCoverage);
base_index = GSUB_is_glyph_covered((const BYTE*)mbpf1+offset, glyphs[base_glyph]);
if (base_index != -1)
{
const GPOS_MarkArray *ma;
const GPOS_MarkRecord *mr;
const GPOS_BaseArray *ba;
const GPOS_BaseRecord *br;
int mark_class;
int class_count = GET_BE_WORD(mbpf1->ClassCount);
int baserecord_size;
POINT base_pt;
POINT mark_pt;
TRACE("Mark %x(%i) and base %x(%i)\n",glyphs[glyph_index], mark_index, glyphs[base_glyph], base_index);
offset = GET_BE_WORD(mbpf1->MarkArray);
ma = (const GPOS_MarkArray*)((const BYTE*)mbpf1 + offset);
if (mark_index > GET_BE_WORD(ma->MarkCount))
{
ERR("Mark index exceeded mark count\n");
return -1;
}
mr = &ma->MarkRecord[mark_index];
mark_class = GET_BE_WORD(mr->Class);
TRACE("Mark Class %i total classes %i\n",mark_class,class_count);
offset = GET_BE_WORD(mbpf1->BaseArray);
ba = (const GPOS_BaseArray*)((const BYTE*)mbpf1 + offset);
baserecord_size = class_count * sizeof(WORD);
br = (const GPOS_BaseRecord*)((const BYTE*)ba + sizeof(WORD) + (baserecord_size * base_index));
offset = GET_BE_WORD(br->BaseAnchor[mark_class]);
GPOS_get_anchor_values((const BYTE*)ba + offset, &base_pt, ppem);
offset = GET_BE_WORD(mr->MarkAnchor);
GPOS_get_anchor_values((const BYTE*)ma + offset, &mark_pt, ppem);
TRACE("Offset on base is %i,%i design units\n",base_pt.x,base_pt.y);
TRACE("Offset on mark is %i,%i design units\n",mark_pt.x, mark_pt.y);
pt->x += base_pt.x - mark_pt.x;
pt->y += base_pt.y - mark_pt.y;
TRACE("Resulting cumulative offset is %i,%i design units\n",pt->x,pt->y);
rc = base_glyph;
}
}
}
else
FIXME("Unhandled Mark To Base Format %i\n",GET_BE_WORD(mbpf1->PosFormat));
}
return rc;
}
static VOID GPOS_apply_MarkToLigature(const OT_LookupTable *look, const SCRIPT_ANALYSIS *analysis, const WORD *glyphs, INT glyph_index,
INT glyph_count, INT ppem, LPPOINT pt)
{
int j;
int write_dir = (analysis->fRTL && !analysis->fLogicalOrder) ? -1 : 1;
TRACE("MarkToLigature Attachment Positioning Subtable\n");
for (j = 0; j < GET_BE_WORD(look->SubTableCount); j++)
{
const GPOS_MarkLigPosFormat1 *mlpf1 = (const GPOS_MarkLigPosFormat1 *)GPOS_get_subtable(look, j);
if (GET_BE_WORD(mlpf1->PosFormat) == 1)
{
int offset = GET_BE_WORD(mlpf1->MarkCoverage);
int mark_index;
mark_index = GSUB_is_glyph_covered((const BYTE*)mlpf1+offset, glyphs[glyph_index]);
if (mark_index != -1)
{
int ligature_index;
offset = GET_BE_WORD(mlpf1->LigatureCoverage);
ligature_index = GSUB_is_glyph_covered((const BYTE*)mlpf1+offset, glyphs[glyph_index - write_dir]);
if (ligature_index != -1)
{
const GPOS_MarkArray *ma;
const GPOS_MarkRecord *mr;
const GPOS_LigatureArray *la;
const GPOS_LigatureAttach *lt;
int mark_class;
int class_count = GET_BE_WORD(mlpf1->ClassCount);
int component_count;
int component_size;
int i;
POINT ligature_pt;
POINT mark_pt;
TRACE("Mark %x(%i) and ligature %x(%i)\n",glyphs[glyph_index], mark_index, glyphs[glyph_index - write_dir], ligature_index);
offset = GET_BE_WORD(mlpf1->MarkArray);
ma = (const GPOS_MarkArray*)((const BYTE*)mlpf1 + offset);
if (mark_index > GET_BE_WORD(ma->MarkCount))
{
ERR("Mark index exceeded mark count\n");
return;
}
mr = &ma->MarkRecord[mark_index];
mark_class = GET_BE_WORD(mr->Class);
TRACE("Mark Class %i total classes %i\n",mark_class,class_count);
offset = GET_BE_WORD(mlpf1->LigatureArray);
la = (const GPOS_LigatureArray*)((const BYTE*)mlpf1 + offset);
if (ligature_index > GET_BE_WORD(la->LigatureCount))
{
ERR("Ligature index exceeded ligature count\n");
return;
}
offset = GET_BE_WORD(la->LigatureAttach[ligature_index]);
lt = (const GPOS_LigatureAttach*)((const BYTE*)la + offset);
component_count = GET_BE_WORD(lt->ComponentCount);
component_size = class_count * sizeof(WORD);
offset = 0;
for (i = 0; i < component_count && !offset; i++)
{
int k;
const GPOS_ComponentRecord *cr = (const GPOS_ComponentRecord*)((const BYTE*)lt->ComponentRecord + (component_size * i));
for (k = 0; k < class_count && !offset; k++)
offset = GET_BE_WORD(cr->LigatureAnchor[k]);
cr = (const GPOS_ComponentRecord*)((const BYTE*)cr + component_size);
}
if (!offset)
{
ERR("Failed to find avalible ligature connection point\n");
return;
}
GPOS_get_anchor_values((const BYTE*)lt + offset, &ligature_pt, ppem);
offset = GET_BE_WORD(mr->MarkAnchor);
GPOS_get_anchor_values((const BYTE*)ma + offset, &mark_pt, ppem);
TRACE("Offset on ligature is %i,%i design units\n",ligature_pt.x,ligature_pt.y);
TRACE("Offset on mark is %i,%i design units\n",mark_pt.x, mark_pt.y);
pt->x += ligature_pt.x - mark_pt.x;
pt->y += ligature_pt.y - mark_pt.y;
TRACE("Resulting cumulative offset is %i,%i design units\n",pt->x,pt->y);
}
}
}
else
FIXME("Unhandled Mark To Ligature Format %i\n",GET_BE_WORD(mlpf1->PosFormat));
}
}
static BOOL GPOS_apply_MarkToMark(const OT_LookupTable *look, const SCRIPT_ANALYSIS *analysis, const WORD *glyphs, INT glyph_index,
INT glyph_count, INT ppem, LPPOINT pt)
{
int j;
BOOL rc = FALSE;
int write_dir = (analysis->fRTL && !analysis->fLogicalOrder) ? -1 : 1;
TRACE("MarkToMark Attachment Positioning Subtable\n");
for (j = 0; j < GET_BE_WORD(look->SubTableCount); j++)
{
const GPOS_MarkMarkPosFormat1 *mmpf1 = (const GPOS_MarkMarkPosFormat1 *)GPOS_get_subtable(look, j);
if (GET_BE_WORD(mmpf1->PosFormat) == 1)
{
int offset = GET_BE_WORD(mmpf1->Mark1Coverage);
int mark_index;
mark_index = GSUB_is_glyph_covered((const BYTE*)mmpf1+offset, glyphs[glyph_index]);
if (mark_index != -1)
{
int mark2_index;
offset = GET_BE_WORD(mmpf1->Mark2Coverage);
mark2_index = GSUB_is_glyph_covered((const BYTE*)mmpf1+offset, glyphs[glyph_index - write_dir]);
if (mark2_index != -1)
{
const GPOS_MarkArray *ma;
const GPOS_MarkRecord *mr;
const GPOS_Mark2Array *m2a;
const GPOS_Mark2Record *m2r;
int mark_class;
int class_count = GET_BE_WORD(mmpf1->ClassCount);
int mark2record_size;
POINT mark2_pt;
POINT mark_pt;
TRACE("Mark %x(%i) and Mark2 %x(%i)\n",glyphs[glyph_index], mark_index, glyphs[glyph_index - write_dir], mark2_index);
offset = GET_BE_WORD(mmpf1->Mark1Array);
ma = (const GPOS_MarkArray*)((const BYTE*)mmpf1 + offset);
if (mark_index > GET_BE_WORD(ma->MarkCount))
{
ERR("Mark index exceeded mark count\n");
return FALSE;
}
mr = &ma->MarkRecord[mark_index];
mark_class = GET_BE_WORD(mr->Class);
TRACE("Mark Class %i total classes %i\n",mark_class,class_count);
offset = GET_BE_WORD(mmpf1->Mark2Array);
m2a = (const GPOS_Mark2Array*)((const BYTE*)mmpf1 + offset);
mark2record_size = class_count * sizeof(WORD);
m2r = (const GPOS_Mark2Record*)((const BYTE*)m2a + sizeof(WORD) + (mark2record_size * mark2_index));
offset = GET_BE_WORD(m2r->Mark2Anchor[mark_class]);
GPOS_get_anchor_values((const BYTE*)m2a + offset, &mark2_pt, ppem);
offset = GET_BE_WORD(mr->MarkAnchor);
GPOS_get_anchor_values((const BYTE*)ma + offset, &mark_pt, ppem);
TRACE("Offset on mark2 is %i,%i design units\n",mark2_pt.x,mark2_pt.y);
TRACE("Offset on mark is %i,%i design units\n",mark_pt.x, mark_pt.y);
pt->x += mark2_pt.x - mark_pt.x;
pt->y += mark2_pt.y - mark_pt.y;
TRACE("Resulting cumulative offset is %i,%i design units\n",pt->x,pt->y);
rc = TRUE;
}
}
}
else
FIXME("Unhandled Mark To Mark Format %i\n",GET_BE_WORD(mmpf1->PosFormat));
}
return rc;
}
static INT GPOS_apply_ChainContextPos(ScriptCache *psc, LPOUTLINETEXTMETRICW lpotm, LPLOGFONTW lplogfont, const SCRIPT_ANALYSIS *analysis, INT* piAdvance,
const OT_LookupList *lookup, const OT_LookupTable *look, const WORD *glyphs, INT glyph_index,
INT glyph_count, INT ppem, GOFFSET *pGoffset)
{
int j;
int write_dir = (analysis->fRTL && !analysis->fLogicalOrder) ? -1 : 1;
TRACE("Chaining Contextual Positioning Subtable\n");
for (j = 0; j < GET_BE_WORD(look->SubTableCount); j++)
{
int offset;
const GPOS_ChainContextPosFormat3_1 *ccpf3 = (GPOS_ChainContextPosFormat3_1 *)GPOS_get_subtable(look, j);
int dirLookahead = write_dir;
int dirBacktrack = -1 * write_dir;
if (GET_BE_WORD(ccpf3->PosFormat) == 1)
{
static int once;
if (!once++)
FIXME(" TODO: subtype 1 (Simple Chaining Context Glyph Positioning)\n");
continue;
}
else if (GET_BE_WORD(ccpf3->PosFormat) == 2)
{
static int once;
if (!once++)
FIXME(" TODO: subtype 2 (Class-based Chaining Context Glyph Positioning)\n");
continue;
}
else if (GET_BE_WORD(ccpf3->PosFormat) == 3)
{
int k;
int indexGlyphs;
const GPOS_ChainContextPosFormat3_2 *ccpf3_2;
const GPOS_ChainContextPosFormat3_3 *ccpf3_3;
const GPOS_ChainContextPosFormat3_4 *ccpf3_4;
TRACE(" subtype 3 (Coverage-based Chaining Context Glyph Positioning)\n");
for (k = 0; k < GET_BE_WORD(ccpf3->BacktrackGlyphCount); k++)
{
offset = GET_BE_WORD(ccpf3->Coverage[k]);
if (GSUB_is_glyph_covered((const BYTE*)ccpf3+offset, glyphs[glyph_index + (dirBacktrack * (k+1))]) == -1)
break;
}
if (k != GET_BE_WORD(ccpf3->BacktrackGlyphCount))
continue;
TRACE("Matched Backtrack\n");
ccpf3_2 = (const GPOS_ChainContextPosFormat3_2*)((BYTE *)ccpf3 +
FIELD_OFFSET(GPOS_ChainContextPosFormat3_1, Coverage[GET_BE_WORD(ccpf3->BacktrackGlyphCount)]));
indexGlyphs = GET_BE_WORD(ccpf3_2->InputGlyphCount);
for (k = 0; k < indexGlyphs; k++)
{
offset = GET_BE_WORD(ccpf3_2->Coverage[k]);
if (GSUB_is_glyph_covered((const BYTE*)ccpf3+offset, glyphs[glyph_index + (write_dir * k)]) == -1)
break;
}
if (k != indexGlyphs)
continue;
TRACE("Matched IndexGlyphs\n");
ccpf3_3 = (const GPOS_ChainContextPosFormat3_3*)((BYTE *)ccpf3_2 +
FIELD_OFFSET(GPOS_ChainContextPosFormat3_2, Coverage[GET_BE_WORD(ccpf3_2->InputGlyphCount)]));
for (k = 0; k < GET_BE_WORD(ccpf3_3->LookaheadGlyphCount); k++)
{
offset = GET_BE_WORD(ccpf3_3->Coverage[k]);
if (GSUB_is_glyph_covered((const BYTE*)ccpf3+offset, glyphs[glyph_index + (dirLookahead * (indexGlyphs + k))]) == -1)
break;
}
if (k != GET_BE_WORD(ccpf3_3->LookaheadGlyphCount))
continue;
TRACE("Matched LookAhead\n");
ccpf3_4 = (const GPOS_ChainContextPosFormat3_4*)((BYTE *)ccpf3_3 +
FIELD_OFFSET(GPOS_ChainContextPosFormat3_3, Coverage[GET_BE_WORD(ccpf3_3->LookaheadGlyphCount)]));
if (GET_BE_WORD(ccpf3_4->PosCount))
{
for (k = 0; k < GET_BE_WORD(ccpf3_4->PosCount); k++)
{
int lookupIndex = GET_BE_WORD(ccpf3_4->PosLookupRecord[k].LookupListIndex);
int SequenceIndex = GET_BE_WORD(ccpf3_4->PosLookupRecord[k].SequenceIndex) * write_dir;
TRACE("Position: %i -> %i %i\n",k, SequenceIndex, lookupIndex);
GPOS_apply_lookup(psc, lpotm, lplogfont, analysis, piAdvance, lookup, lookupIndex, glyphs, glyph_index + SequenceIndex, glyph_count, pGoffset);
}
return glyph_index + indexGlyphs + GET_BE_WORD(ccpf3_3->LookaheadGlyphCount);
}
else return glyph_index + 1;
}
else
FIXME("Unhandled Chaining Contextual Positioning Format %i\n",GET_BE_WORD(ccpf3->PosFormat));
}
return glyph_index + 1;
}
static INT GPOS_apply_lookup(ScriptCache *psc, LPOUTLINETEXTMETRICW lpotm, LPLOGFONTW lplogfont, const SCRIPT_ANALYSIS *analysis, INT* piAdvance, const OT_LookupList* lookup, INT lookup_index, const WORD *glyphs, INT glyph_index, INT glyph_count, GOFFSET *pGoffset)
{
int offset;
const OT_LookupTable *look;
int ppem = lpotm->otmTextMetrics.tmAscent + lpotm->otmTextMetrics.tmDescent - lpotm->otmTextMetrics.tmInternalLeading;
WORD type;
offset = GET_BE_WORD(lookup->Lookup[lookup_index]);
look = (const OT_LookupTable*)((const BYTE*)lookup + offset);
type = GET_BE_WORD(look->LookupType);
TRACE("type %i, flag %x, subtables %i\n",type,GET_BE_WORD(look->LookupFlag),GET_BE_WORD(look->SubTableCount));
if (type == 9)
{
if (GET_BE_WORD(look->SubTableCount))
{
const GPOS_ExtensionPosFormat1 *ext = (const GPOS_ExtensionPosFormat1 *)((const BYTE *)look + GET_BE_WORD(look->SubTable[0]));
if (GET_BE_WORD(ext->PosFormat) == 1)
{
type = GET_BE_WORD(ext->ExtensionLookupType);
TRACE("extension type %i\n",type);
}
else
{
FIXME("Unhandled Extension Positioning Format %i\n",GET_BE_WORD(ext->PosFormat));
}
}
else
{
WARN("lookup type is Extension Positioning but no extension subtable exists\n");
}
}
switch (type)
{
case 1:
{
double devX, devY;
POINT adjust = {0,0};
POINT advance = {0,0};
GPOS_apply_SingleAdjustment(look, analysis, glyphs, glyph_index, glyph_count, ppem, &adjust, &advance);
if (adjust.x || adjust.y)
{
GPOS_convert_design_units_to_device(lpotm, lplogfont, adjust.x, adjust.y, &devX, &devY);
pGoffset[glyph_index].du += round(devX);
pGoffset[glyph_index].dv += round(devY);
}
if (advance.x || advance.y)
{
GPOS_convert_design_units_to_device(lpotm, lplogfont, advance.x, advance.y, &devX, &devY);
piAdvance[glyph_index] += round(devX);
if (advance.y)
FIXME("Unhandled adjustment to Y advancement\n");
}
break;
}
case 2:
{
POINT advance[2]= {{0,0},{0,0}};
POINT adjust[2]= {{0,0},{0,0}};
double devX, devY;
int index;
int write_dir = (analysis->fRTL && !analysis->fLogicalOrder) ? -1 : 1;
int offset_sign = (analysis->fRTL && analysis->fLogicalOrder) ? -1 : 1;
index = GPOS_apply_PairAdjustment(look, analysis, glyphs, glyph_index, glyph_count, ppem, adjust, advance);
if (adjust[0].x || adjust[0].y)
{
GPOS_convert_design_units_to_device(lpotm, lplogfont, adjust[0].x, adjust[0].y, &devX, &devY);
pGoffset[glyph_index].du += round(devX) * offset_sign;
pGoffset[glyph_index].dv += round(devY);
}
if (advance[0].x || advance[0].y)
{
GPOS_convert_design_units_to_device(lpotm, lplogfont, advance[0].x, advance[0].y, &devX, &devY);
piAdvance[glyph_index] += round(devX);
}
if (adjust[1].x || adjust[1].y)
{
GPOS_convert_design_units_to_device(lpotm, lplogfont, adjust[1].x, adjust[1].y, &devX, &devY);
pGoffset[glyph_index + write_dir].du += round(devX) * offset_sign;
pGoffset[glyph_index + write_dir].dv += round(devY);
}
if (advance[1].x || advance[1].y)
{
GPOS_convert_design_units_to_device(lpotm, lplogfont, advance[1].x, advance[1].y, &devX, &devY);
piAdvance[glyph_index + write_dir] += round(devX);
}
return index;
}
case 3:
{
POINT desU = {0,0};
double devX, devY;
int write_dir = (analysis->fRTL && !analysis->fLogicalOrder) ? -1 : 1;
GPOS_apply_CursiveAttachment(look, analysis, glyphs, glyph_index, glyph_count, ppem, &desU);
if (desU.x || desU.y)
{
GPOS_convert_design_units_to_device(lpotm, lplogfont, desU.x, desU.y, &devX, &devY);
/* Windows does not appear to apply X offsets here */
pGoffset[glyph_index].dv = round(devY) + pGoffset[glyph_index+write_dir].dv;
}
break;
}
case 4:
{
double devX, devY;
POINT desU = {0,0};
int base_index = GPOS_apply_MarkToBase(psc, look, analysis, glyphs, glyph_index, glyph_count, ppem, &desU);
if (base_index != -1)
{
GPOS_convert_design_units_to_device(lpotm, lplogfont, desU.x, desU.y, &devX, &devY);
if (!analysis->fRTL) pGoffset[glyph_index].du = round(devX) - piAdvance[base_index];
else
{
if (analysis->fLogicalOrder) devX *= -1;
pGoffset[glyph_index].du = round(devX);
}
pGoffset[glyph_index].dv = round(devY);
}
break;
}
case 5:
{
double devX, devY;
POINT desU = {0,0};
GPOS_apply_MarkToLigature(look, analysis, glyphs, glyph_index, glyph_count, ppem, &desU);
if (desU.x || desU.y)
{
GPOS_convert_design_units_to_device(lpotm, lplogfont, desU.x, desU.y, &devX, &devY);
pGoffset[glyph_index].du = (round(devX) - piAdvance[glyph_index-1]);
pGoffset[glyph_index].dv = round(devY);
}
break;
}
case 6:
{
double devX, devY;
POINT desU = {0,0};
int write_dir = (analysis->fRTL && !analysis->fLogicalOrder) ? -1 : 1;
if (GPOS_apply_MarkToMark(look, analysis, glyphs, glyph_index, glyph_count, ppem, &desU))
{
GPOS_convert_design_units_to_device(lpotm, lplogfont, desU.x, desU.y, &devX, &devY);
if (analysis->fRTL && analysis->fLogicalOrder) devX *= -1;
pGoffset[glyph_index].du = round(devX) + pGoffset[glyph_index - write_dir].du;
pGoffset[glyph_index].dv = round(devY) + pGoffset[glyph_index - write_dir].dv;
}
break;
}
case 8:
{
return GPOS_apply_ChainContextPos(psc, lpotm, lplogfont, analysis, piAdvance, lookup, look, glyphs, glyph_index, glyph_count, ppem, pGoffset);
}
default:
FIXME("We do not handle SubType %i\n",type);
}
return glyph_index+1;
}
INT OpenType_apply_GPOS_lookup(ScriptCache *psc, LPOUTLINETEXTMETRICW lpotm, LPLOGFONTW lplogfont, const SCRIPT_ANALYSIS *analysis, INT* piAdvance, INT lookup_index, const WORD *glyphs, INT glyph_index, INT glyph_count, GOFFSET *pGoffset)
{
const GPOS_Header *header = (const GPOS_Header *)psc->GPOS_Table;
const OT_LookupList *lookup = (const OT_LookupList*)((const BYTE*)header + GET_BE_WORD(header->LookupList));
return GPOS_apply_lookup(psc, lpotm, lplogfont, analysis, piAdvance, lookup, lookup_index, glyphs, glyph_index, glyph_count, pGoffset);
}
static void GSUB_initialize_script_cache(ScriptCache *psc)
{
int i;
if (psc->GSUB_Table)
{
const OT_ScriptList *script;
const GSUB_Header* header = (const GSUB_Header*)psc->GSUB_Table;
script = (const OT_ScriptList*)((const BYTE*)header + GET_BE_WORD(header->ScriptList));
psc->script_count = GET_BE_WORD(script->ScriptCount);
TRACE("initializing %i scripts in this font\n",psc->script_count);
if (psc->script_count)
{
psc->scripts = HeapAlloc(GetProcessHeap(),HEAP_ZERO_MEMORY,sizeof(LoadedScript) * psc->script_count);
for (i = 0; i < psc->script_count; i++)
{
int offset = GET_BE_WORD(script->ScriptRecord[i].Script);
psc->scripts[i].tag = MS_MAKE_TAG(script->ScriptRecord[i].ScriptTag[0], script->ScriptRecord[i].ScriptTag[1], script->ScriptRecord[i].ScriptTag[2], script->ScriptRecord[i].ScriptTag[3]);
psc->scripts[i].gsub_table = ((const BYTE*)script + offset);
}
}
}
}
static void GPOS_expand_script_cache(ScriptCache *psc)
{
int i, count;
const OT_ScriptList *script;
const GPOS_Header* header = (const GPOS_Header*)psc->GPOS_Table;
if (!header)
return;
script = (const OT_ScriptList*)((const BYTE*)header + GET_BE_WORD(header->ScriptList));
count = GET_BE_WORD(script->ScriptCount);
if (!count)
return;
if (!psc->script_count)
{
psc->script_count = count;
TRACE("initializing %i scripts in this font\n",psc->script_count);
if (psc->script_count)
{
psc->scripts = HeapAlloc(GetProcessHeap(),HEAP_ZERO_MEMORY,sizeof(LoadedScript) * psc->script_count);
for (i = 0; i < psc->script_count; i++)
{
int offset = GET_BE_WORD(script->ScriptRecord[i].Script);
psc->scripts[i].tag = MS_MAKE_TAG(script->ScriptRecord[i].ScriptTag[0], script->ScriptRecord[i].ScriptTag[1], script->ScriptRecord[i].ScriptTag[2], script->ScriptRecord[i].ScriptTag[3]);
psc->scripts[i].gpos_table = ((const BYTE*)script + offset);
}
}
}
else
{
for (i = 0; i < count; i++)
{
int j;
int offset = GET_BE_WORD(script->ScriptRecord[i].Script);
OPENTYPE_TAG tag = MS_MAKE_TAG(script->ScriptRecord[i].ScriptTag[0], script->ScriptRecord[i].ScriptTag[1], script->ScriptRecord[i].ScriptTag[2], script->ScriptRecord[i].ScriptTag[3]);
for (j = 0; j < psc->script_count; j++)
{
if (psc->scripts[j].tag == tag)
{
psc->scripts[j].gpos_table = ((const BYTE*)script + offset);
break;
}
}
if (j == psc->script_count)
{
psc->script_count++;
psc->scripts = HeapReAlloc(GetProcessHeap(),HEAP_ZERO_MEMORY,psc->scripts, sizeof(LoadedScript) * psc->script_count);
psc->scripts[j].tag = tag;
psc->scripts[j].gpos_table = ((const BYTE*)script + offset);
}
}
}
}
static void _initialize_script_cache(ScriptCache *psc)
{
if (!psc->scripts_initialized)
{
GSUB_initialize_script_cache(psc);
GPOS_expand_script_cache(psc);
psc->scripts_initialized = TRUE;
}
}
HRESULT OpenType_GetFontScriptTags(ScriptCache *psc, OPENTYPE_TAG searchingFor, int cMaxTags, OPENTYPE_TAG *pScriptTags, int *pcTags)
{
int i;
HRESULT rc = S_OK;
_initialize_script_cache(psc);
*pcTags = psc->script_count;
if (!searchingFor && cMaxTags < *pcTags)
rc = E_OUTOFMEMORY;
else if (searchingFor)
rc = USP_E_SCRIPT_NOT_IN_FONT;
for (i = 0; i < psc->script_count; i++)
{
if (i < cMaxTags)
pScriptTags[i] = psc->scripts[i].tag;
if (searchingFor)
{
if (searchingFor == psc->scripts[i].tag)
{
pScriptTags[0] = psc->scripts[i].tag;
*pcTags = 1;
rc = S_OK;
break;
}
}
}
return rc;
}
static void GSUB_initialize_language_cache(LoadedScript *script)
{
int i;
if (script->gsub_table)
{
DWORD offset;
const OT_Script* table = script->gsub_table;
script->language_count = GET_BE_WORD(table->LangSysCount);
offset = GET_BE_WORD(table->DefaultLangSys);
if (offset)
{
script->default_language.tag = MS_MAKE_TAG('d','f','l','t');
script->default_language.gsub_table = (const BYTE*)table + offset;
}
if (script->language_count)
{
TRACE("Deflang %p, LangCount %i\n",script->default_language.gsub_table, script->language_count);
script->languages = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(LoadedLanguage) * script->language_count);
for (i = 0; i < script->language_count; i++)
{
int offset = GET_BE_WORD(table->LangSysRecord[i].LangSys);
script->languages[i].tag = MS_MAKE_TAG(table->LangSysRecord[i].LangSysTag[0], table->LangSysRecord[i].LangSysTag[1], table->LangSysRecord[i].LangSysTag[2], table->LangSysRecord[i].LangSysTag[3]);
script->languages[i].gsub_table = ((const BYTE*)table + offset);
}
}
}
}
static void GPOS_expand_language_cache(LoadedScript *script)
{
int count;
const OT_Script* table = script->gpos_table;
DWORD offset;
if (!table)
return;
offset = GET_BE_WORD(table->DefaultLangSys);
if (offset)
script->default_language.gpos_table = (const BYTE*)table + offset;
count = GET_BE_WORD(table->LangSysCount);
TRACE("Deflang %p, LangCount %i\n",script->default_language.gpos_table, count);
if (!count)
return;
if (!script->language_count)
{
int i;
script->language_count = count;
script->languages = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, sizeof(LoadedLanguage) * script->language_count);
for (i = 0; i < script->language_count; i++)
{
int offset = GET_BE_WORD(table->LangSysRecord[i].LangSys);
script->languages[i].tag = MS_MAKE_TAG(table->LangSysRecord[i].LangSysTag[0], table->LangSysRecord[i].LangSysTag[1], table->LangSysRecord[i].LangSysTag[2], table->LangSysRecord[i].LangSysTag[3]);
script->languages[i].gpos_table = ((const BYTE*)table + offset);
}
}
else if (count)
{
int i,j;
for (i = 0; i < count; i++)
{
int offset = GET_BE_WORD(table->LangSysRecord[i].LangSys);
OPENTYPE_TAG tag = MS_MAKE_TAG(table->LangSysRecord[i].LangSysTag[0], table->LangSysRecord[i].LangSysTag[1], table->LangSysRecord[i].LangSysTag[2], table->LangSysRecord[i].LangSysTag[3]);
for (j = 0; j < script->language_count; j++)
{
if (script->languages[j].tag == tag)
{
script->languages[j].gpos_table = ((const BYTE*)table + offset);
break;
}
}
if (j == script->language_count)
{
script->language_count++;
script->languages = HeapReAlloc(GetProcessHeap(),HEAP_ZERO_MEMORY,script->languages, sizeof(LoadedLanguage) * script->language_count);
script->languages[j].tag = tag;
script->languages[j].gpos_table = ((const BYTE*)table + offset);
}
}
}
}
static void _initialize_language_cache(LoadedScript *script)
{
if (!script->languages_initialized)
{
GSUB_initialize_language_cache(script);
GPOS_expand_language_cache(script);
script->languages_initialized = TRUE;
}
}
HRESULT OpenType_GetFontLanguageTags(ScriptCache *psc, OPENTYPE_TAG script_tag, OPENTYPE_TAG searchingFor, int cMaxTags, OPENTYPE_TAG *pLanguageTags, int *pcTags)
{
int i;
HRESULT rc = S_OK;
LoadedScript *script = NULL;
_initialize_script_cache(psc);
for (i = 0; i < psc->script_count; i++)
{
if (psc->scripts[i].tag == script_tag)
{
script = &psc->scripts[i];
break;
}
}
if (!script)
return E_INVALIDARG;
_initialize_language_cache(script);
if (!searchingFor && cMaxTags < script->language_count)
rc = E_OUTOFMEMORY;
else if (searchingFor)
rc = E_INVALIDARG;
*pcTags = script->language_count;
for (i = 0; i < script->language_count; i++)
{
if (i < cMaxTags)
pLanguageTags[i] = script->languages[i].tag;
if (searchingFor)
{
if (searchingFor == script->languages[i].tag)
{
pLanguageTags[0] = script->languages[i].tag;
*pcTags = 1;
rc = S_OK;
break;
}
}
}
if (script->default_language.gsub_table)
{
if (i < cMaxTags)
pLanguageTags[i] = script->default_language.tag;
if (searchingFor && FAILED(rc))
{
pLanguageTags[0] = script->default_language.tag;
}
i++;
*pcTags = (*pcTags) + 1;
}
return rc;
}
static void GSUB_initialize_feature_cache(LPCVOID table, LoadedLanguage *language)
{
int i;
if (language->gsub_table)
{
const OT_LangSys *lang = language->gsub_table;
const GSUB_Header *header = (const GSUB_Header *)table;
const OT_FeatureList *feature_list;
language->feature_count = GET_BE_WORD(lang->FeatureCount);
TRACE("%i features\n",language->feature_count);
if (language->feature_count)
{
language->features = HeapAlloc(GetProcessHeap(),0,sizeof(LoadedFeature)*language->feature_count);
feature_list = (const OT_FeatureList*)((const BYTE*)header + GET_BE_WORD(header->FeatureList));
for (i = 0; i < language->feature_count; i++)
{
const OT_Feature *feature;
int j;
int index = GET_BE_WORD(lang->FeatureIndex[i]);
language->features[i].tag = MS_MAKE_TAG(feature_list->FeatureRecord[index].FeatureTag[0], feature_list->FeatureRecord[index].FeatureTag[1], feature_list->FeatureRecord[index].FeatureTag[2], feature_list->FeatureRecord[index].FeatureTag[3]);
language->features[i].feature = ((const BYTE*)feature_list + GET_BE_WORD(feature_list->FeatureRecord[index].Feature));
feature = (const OT_Feature*)language->features[i].feature;
language->features[i].lookup_count = GET_BE_WORD(feature->LookupCount);
language->features[i].lookups = HeapAlloc(GetProcessHeap(),0,sizeof(WORD) * language->features[i].lookup_count);
for (j = 0; j < language->features[i].lookup_count; j++)
language->features[i].lookups[j] = GET_BE_WORD(feature->LookupListIndex[j]);
language->features[i].tableType = FEATURE_GSUB_TABLE;
}
}
}
}
static void GPOS_expand_feature_cache(LPCVOID table, LoadedLanguage *language)
{
int i, count;
const OT_LangSys *lang = language->gpos_table;
const GPOS_Header *header = (const GPOS_Header *)table;
const OT_FeatureList *feature_list;
if (!lang)
return;
count = GET_BE_WORD(lang->FeatureCount);
feature_list = (const OT_FeatureList*)((const BYTE*)header + GET_BE_WORD(header->FeatureList));
TRACE("%i features\n",count);
if (!count)
return;
if (!language->feature_count)
{
language->feature_count = count;
if (language->feature_count)
{
language->features = HeapAlloc(GetProcessHeap(),0,sizeof(LoadedFeature)*language->feature_count);
for (i = 0; i < language->feature_count; i++)
{
const OT_Feature *feature;
int j;
int index = GET_BE_WORD(lang->FeatureIndex[i]);
language->features[i].tag = MS_MAKE_TAG(feature_list->FeatureRecord[index].FeatureTag[0], feature_list->FeatureRecord[index].FeatureTag[1], feature_list->FeatureRecord[index].FeatureTag[2], feature_list->FeatureRecord[index].FeatureTag[3]);
language->features[i].feature = ((const BYTE*)feature_list + GET_BE_WORD(feature_list->FeatureRecord[index].Feature));
feature = (const OT_Feature*)language->features[i].feature;
language->features[i].lookup_count = GET_BE_WORD(feature->LookupCount);
language->features[i].lookups = HeapAlloc(GetProcessHeap(),0,sizeof(WORD) * language->features[i].lookup_count);
for (j = 0; j < language->features[i].lookup_count; j++)
language->features[i].lookups[j] = GET_BE_WORD(feature->LookupListIndex[j]);
language->features[i].tableType = FEATURE_GPOS_TABLE;
}
}
}
else
{
language->features = HeapReAlloc(GetProcessHeap(),0,language->features, sizeof(LoadedFeature)*(language->feature_count + count));
for (i = 0; i < count; i++)
{
const OT_Feature *feature;
int j;
int index = GET_BE_WORD(lang->FeatureIndex[i]);
int idx = language->feature_count + i;
language->features[idx].tag = MS_MAKE_TAG(feature_list->FeatureRecord[index].FeatureTag[0], feature_list->FeatureRecord[index].FeatureTag[1], feature_list->FeatureRecord[index].FeatureTag[2], feature_list->FeatureRecord[index].FeatureTag[3]);
language->features[idx].feature = ((const BYTE*)feature_list + GET_BE_WORD(feature_list->FeatureRecord[index].Feature));
feature = (const OT_Feature*)language->features[idx].feature;
language->features[idx].lookup_count = GET_BE_WORD(feature->LookupCount);
language->features[idx].lookups = HeapAlloc(GetProcessHeap(),0,sizeof(WORD) * language->features[idx].lookup_count);
for (j = 0; j < language->features[idx].lookup_count; j++)
language->features[idx].lookups[j] = GET_BE_WORD(feature->LookupListIndex[j]);
language->features[idx].tableType = FEATURE_GPOS_TABLE;
}
language->feature_count += count;
}
}
static void _initialize_feature_cache(ScriptCache *psc, LoadedLanguage *language)
{
if (!language->features_initialized)
{
GSUB_initialize_feature_cache(psc->GSUB_Table, language);
GPOS_expand_feature_cache(psc->GPOS_Table, language);
language->features_initialized = TRUE;
}
}
HRESULT OpenType_GetFontFeatureTags(ScriptCache *psc, OPENTYPE_TAG script_tag, OPENTYPE_TAG language_tag, BOOL filtered, OPENTYPE_TAG searchingFor, char tableType, int cMaxTags, OPENTYPE_TAG *pFeatureTags, int *pcTags, LoadedFeature** feature)
{
int i;
HRESULT rc = S_OK;
LoadedScript *script = NULL;
LoadedLanguage *language = NULL;
_initialize_script_cache(psc);
for (i = 0; i < psc->script_count; i++)
{
if (psc->scripts[i].tag == script_tag)
{
script = &psc->scripts[i];
break;
}
}
if (!script)
{
*pcTags = 0;
if (!filtered)
return S_OK;
else
return E_INVALIDARG;
}
_initialize_language_cache(script);
if ((script->default_language.gsub_table || script->default_language.gpos_table) && script->default_language.tag == language_tag)
language = &script->default_language;
else
{
for (i = 0; i < script->language_count; i++)
{
if (script->languages[i].tag == language_tag)
{
language = &script->languages[i];
break;
}
}
}
if (!language)
{
*pcTags = 0;
return S_OK;
}
_initialize_feature_cache(psc, language);
if (tableType)
{
*pcTags = 0;
for (i = 0; i < language->feature_count; i++)
if (language->features[i].tableType == tableType)
*pcTags = (*pcTags)+1;
}
else
*pcTags = language->feature_count;
if (!searchingFor && cMaxTags < *pcTags)
rc = E_OUTOFMEMORY;
else if (searchingFor)
rc = E_INVALIDARG;
for (i = 0; i < language->feature_count; i++)
{
if (i < cMaxTags)
{
if (!tableType || language->features[i].tableType == tableType)
pFeatureTags[i] = language->features[i].tag;
}
if (searchingFor)
{
if ((searchingFor == language->features[i].tag) &&
(!tableType || language->features[i].tableType == tableType))
{
pFeatureTags[0] = language->features[i].tag;
*pcTags = 1;
if (feature)
*feature = &language->features[i];
rc = S_OK;
break;
}
}
}
return rc;
}