freetype2/src/psaux/afmparse.c

1093 lines
25 KiB
C
Raw Normal View History

/****************************************************************************
*
* afmparse.c
*
* AFM parser (body).
*
2021-01-17 07:18:48 +01:00
* Copyright (C) 2006-2021 by
* David Turner, Robert Wilhelm, and Werner Lemberg.
*
* This file is part of the FreeType project, and may only be used,
* modified, and distributed under the terms of the FreeType project
* license, LICENSE.TXT. By continuing to use, modify, or distribute
* this file you indicate that you have read the license and
* understand and accept it fully.
*
*/
#include <freetype/freetype.h>
#include <freetype/internal/ftdebug.h>
#include <freetype/internal/psaux.h>
#ifndef T1_CONFIG_OPTION_NO_AFM
#include "afmparse.h"
#include "psconv.h"
#include "psauxerr.h"
2006-01-16 23:35:33 +01:00
/**************************************************************************
*
* The macro FT_COMPONENT is used in trace mode. It is an implicit
* parameter of the FT_TRACE() and FT_ERROR() macros, used to print/log
* messages during execution.
*/
#undef FT_COMPONENT
#define FT_COMPONENT afmparse
/**************************************************************************
*
* AFM_Stream
*
* The use of AFM_Stream is largely inspired by parseAFM.[ch] from t1lib.
*
*/
enum
{
AFM_STREAM_STATUS_NORMAL,
AFM_STREAM_STATUS_EOC,
AFM_STREAM_STATUS_EOL,
AFM_STREAM_STATUS_EOF
};
typedef struct AFM_StreamRec_
{
2006-01-16 23:35:33 +01:00
FT_Byte* cursor;
FT_Byte* base;
FT_Byte* limit;
2006-01-16 23:35:33 +01:00
FT_Int status;
} AFM_StreamRec;
#ifndef EOF
#define EOF -1
#endif
2006-01-16 23:35:33 +01:00
/* this works because empty lines are ignored */
#define AFM_IS_NEWLINE( ch ) ( (ch) == '\r' || (ch) == '\n' )
#define AFM_IS_EOF( ch ) ( (ch) == EOF || (ch) == '\x1a' )
#define AFM_IS_SPACE( ch ) ( (ch) == ' ' || (ch) == '\t' )
2006-01-16 23:35:33 +01:00
/* column separator; there is no `column' in the spec actually */
#define AFM_IS_SEP( ch ) ( (ch) == ';' )
2006-01-16 23:35:33 +01:00
#define AFM_GETC() \
( ( (stream)->cursor < (stream)->limit ) ? *(stream)->cursor++ \
: EOF )
2006-01-16 23:35:33 +01:00
#define AFM_STREAM_KEY_BEGIN( stream ) \
(char*)( (stream)->cursor - 1 )
#define AFM_STREAM_KEY_LEN( stream, key ) \
(FT_Offset)( (char*)(stream)->cursor - key - 1 )
#define AFM_STATUS_EOC( stream ) \
2006-01-16 23:35:33 +01:00
( (stream)->status >= AFM_STREAM_STATUS_EOC )
#define AFM_STATUS_EOL( stream ) \
2006-01-16 23:35:33 +01:00
( (stream)->status >= AFM_STREAM_STATUS_EOL )
#define AFM_STATUS_EOF( stream ) \
2006-01-16 23:35:33 +01:00
( (stream)->status >= AFM_STREAM_STATUS_EOF )
static int
afm_stream_skip_spaces( AFM_Stream stream )
{
int ch = 0; /* make stupid compiler happy */
if ( AFM_STATUS_EOC( stream ) )
return ';';
while ( 1 )
{
ch = AFM_GETC();
if ( !AFM_IS_SPACE( ch ) )
break;
}
if ( AFM_IS_NEWLINE( ch ) )
stream->status = AFM_STREAM_STATUS_EOL;
else if ( AFM_IS_SEP( ch ) )
stream->status = AFM_STREAM_STATUS_EOC;
else if ( AFM_IS_EOF( ch ) )
stream->status = AFM_STREAM_STATUS_EOF;
return ch;
}
2006-01-16 23:35:33 +01:00
/* read a key or value in current column */
static char*
afm_stream_read_one( AFM_Stream stream )
{
char* str;
afm_stream_skip_spaces( stream );
if ( AFM_STATUS_EOC( stream ) )
return NULL;
str = AFM_STREAM_KEY_BEGIN( stream );
while ( 1 )
{
int ch = AFM_GETC();
if ( AFM_IS_SPACE( ch ) )
break;
else if ( AFM_IS_NEWLINE( ch ) )
{
stream->status = AFM_STREAM_STATUS_EOL;
break;
}
else if ( AFM_IS_SEP( ch ) )
{
stream->status = AFM_STREAM_STATUS_EOC;
break;
}
else if ( AFM_IS_EOF( ch ) )
{
stream->status = AFM_STREAM_STATUS_EOF;
break;
}
}
return str;
}
/* read a string (i.e., read to EOL) */
static char*
afm_stream_read_string( AFM_Stream stream )
{
char* str;
afm_stream_skip_spaces( stream );
if ( AFM_STATUS_EOL( stream ) )
return NULL;
str = AFM_STREAM_KEY_BEGIN( stream );
/* scan to eol */
while ( 1 )
{
int ch = AFM_GETC();
if ( AFM_IS_NEWLINE( ch ) )
{
stream->status = AFM_STREAM_STATUS_EOL;
break;
}
else if ( AFM_IS_EOF( ch ) )
{
stream->status = AFM_STREAM_STATUS_EOF;
break;
}
}
return str;
}
/**************************************************************************
*
* AFM_Parser
*
*/
/* all keys defined in Ch. 7-10 of 5004.AFM_Spec.pdf */
2006-01-16 23:35:33 +01:00
typedef enum AFM_Token_
{
AFM_TOKEN_ASCENDER,
AFM_TOKEN_AXISLABEL,
AFM_TOKEN_AXISTYPE,
AFM_TOKEN_B,
AFM_TOKEN_BLENDAXISTYPES,
AFM_TOKEN_BLENDDESIGNMAP,
AFM_TOKEN_BLENDDESIGNPOSITIONS,
AFM_TOKEN_C,
AFM_TOKEN_CC,
AFM_TOKEN_CH,
AFM_TOKEN_CAPHEIGHT,
AFM_TOKEN_CHARWIDTH,
AFM_TOKEN_CHARACTERSET,
AFM_TOKEN_CHARACTERS,
AFM_TOKEN_DESCENDER,
AFM_TOKEN_ENCODINGSCHEME,
AFM_TOKEN_ENDAXIS,
AFM_TOKEN_ENDCHARMETRICS,
AFM_TOKEN_ENDCOMPOSITES,
AFM_TOKEN_ENDDIRECTION,
AFM_TOKEN_ENDFONTMETRICS,
AFM_TOKEN_ENDKERNDATA,
AFM_TOKEN_ENDKERNPAIRS,
AFM_TOKEN_ENDTRACKKERN,
AFM_TOKEN_ESCCHAR,
AFM_TOKEN_FAMILYNAME,
AFM_TOKEN_FONTBBOX,
AFM_TOKEN_FONTNAME,
AFM_TOKEN_FULLNAME,
AFM_TOKEN_ISBASEFONT,
AFM_TOKEN_ISCIDFONT,
AFM_TOKEN_ISFIXEDPITCH,
AFM_TOKEN_ISFIXEDV,
AFM_TOKEN_ITALICANGLE,
AFM_TOKEN_KP,
AFM_TOKEN_KPH,
AFM_TOKEN_KPX,
AFM_TOKEN_KPY,
AFM_TOKEN_L,
AFM_TOKEN_MAPPINGSCHEME,
AFM_TOKEN_METRICSSETS,
AFM_TOKEN_N,
AFM_TOKEN_NOTICE,
AFM_TOKEN_PCC,
AFM_TOKEN_STARTAXIS,
AFM_TOKEN_STARTCHARMETRICS,
AFM_TOKEN_STARTCOMPOSITES,
AFM_TOKEN_STARTDIRECTION,
AFM_TOKEN_STARTFONTMETRICS,
AFM_TOKEN_STARTKERNDATA,
AFM_TOKEN_STARTKERNPAIRS,
AFM_TOKEN_STARTKERNPAIRS0,
AFM_TOKEN_STARTKERNPAIRS1,
AFM_TOKEN_STARTTRACKKERN,
AFM_TOKEN_STDHW,
AFM_TOKEN_STDVW,
AFM_TOKEN_TRACKKERN,
AFM_TOKEN_UNDERLINEPOSITION,
AFM_TOKEN_UNDERLINETHICKNESS,
AFM_TOKEN_VV,
AFM_TOKEN_VVECTOR,
AFM_TOKEN_VERSION,
AFM_TOKEN_W,
AFM_TOKEN_W0,
AFM_TOKEN_W0X,
AFM_TOKEN_W0Y,
AFM_TOKEN_W1,
AFM_TOKEN_W1X,
AFM_TOKEN_W1Y,
AFM_TOKEN_WX,
AFM_TOKEN_WY,
AFM_TOKEN_WEIGHT,
AFM_TOKEN_WEIGHTVECTOR,
AFM_TOKEN_XHEIGHT,
N_AFM_TOKENS,
AFM_TOKEN_UNKNOWN
2006-01-16 23:35:33 +01:00
} AFM_Token;
static const char* const afm_key_table[N_AFM_TOKENS] =
{
"Ascender",
"AxisLabel",
"AxisType",
"B",
"BlendAxisTypes",
"BlendDesignMap",
"BlendDesignPositions",
"C",
"CC",
"CH",
"CapHeight",
"CharWidth",
"CharacterSet",
"Characters",
"Descender",
"EncodingScheme",
"EndAxis",
"EndCharMetrics",
"EndComposites",
"EndDirection",
"EndFontMetrics",
"EndKernData",
"EndKernPairs",
"EndTrackKern",
"EscChar",
"FamilyName",
"FontBBox",
"FontName",
"FullName",
"IsBaseFont",
"IsCIDFont",
"IsFixedPitch",
"IsFixedV",
"ItalicAngle",
"KP",
"KPH",
"KPX",
"KPY",
"L",
"MappingScheme",
"MetricsSets",
"N",
"Notice",
"PCC",
"StartAxis",
"StartCharMetrics",
"StartComposites",
"StartDirection",
"StartFontMetrics",
"StartKernData",
"StartKernPairs",
"StartKernPairs0",
"StartKernPairs1",
"StartTrackKern",
"StdHW",
"StdVW",
"TrackKern",
"UnderlinePosition",
"UnderlineThickness",
"VV",
"VVector",
"Version",
"W",
"W0",
"W0X",
"W0Y",
"W1",
"W1X",
"W1Y",
"WX",
"WY",
"Weight",
"WeightVector",
"XHeight"
};
/*
2006-01-16 23:35:33 +01:00
* `afm_parser_read_vals' and `afm_parser_next_key' provide
* high-level operations to an AFM_Stream. The rest of the
2006-01-16 23:35:33 +01:00
* parser functions should use them without accessing the
* AFM_Stream directly.
*/
FT_LOCAL_DEF( FT_Int )
afm_parser_read_vals( AFM_Parser parser,
AFM_Value vals,
FT_Int n )
{
AFM_Stream stream = parser->stream;
char* str;
FT_Int i;
if ( n > AFM_MAX_ARGUMENTS )
return 0;
for ( i = 0; i < n; i++ )
{
FT_Offset len;
AFM_Value val = vals + i;
if ( val->type == AFM_VALUE_TYPE_STRING )
str = afm_stream_read_string( stream );
else
str = afm_stream_read_one( stream );
if ( !str )
break;
len = AFM_STREAM_KEY_LEN( stream, str );
switch ( val->type )
{
2006-01-16 23:35:33 +01:00
case AFM_VALUE_TYPE_STRING:
case AFM_VALUE_TYPE_NAME:
2006-01-27 15:16:16 +01:00
{
FT_Memory memory = parser->memory;
FT_Error error;
if ( !FT_QALLOC( val->u.s, len + 1 ) )
{
ft_memcpy( val->u.s, str, len );
val->u.s[len] = '\0';
}
}
2006-01-16 23:35:33 +01:00
break;
case AFM_VALUE_TYPE_FIXED:
val->u.f = PS_Conv_ToFixed( (FT_Byte**)(void*)&str,
(FT_Byte*)str + len, 0 );
2006-01-16 23:35:33 +01:00
break;
case AFM_VALUE_TYPE_INTEGER:
val->u.i = PS_Conv_ToInt( (FT_Byte**)(void*)&str,
(FT_Byte*)str + len );
2006-01-16 23:35:33 +01:00
break;
case AFM_VALUE_TYPE_BOOL:
val->u.b = FT_BOOL( len == 4 &&
!ft_strncmp( str, "true", 4 ) );
2006-01-16 23:35:33 +01:00
break;
case AFM_VALUE_TYPE_INDEX:
if ( parser->get_index )
val->u.i = parser->get_index( str, len, parser->user_data );
2006-01-16 23:35:33 +01:00
else
val->u.i = 0;
2006-01-16 23:35:33 +01:00
break;
}
}
return i;
}
FT_LOCAL_DEF( char* )
afm_parser_next_key( AFM_Parser parser,
FT_Bool line,
FT_Offset* len )
{
AFM_Stream stream = parser->stream;
2015-04-13 05:16:48 +02:00
char* key = NULL; /* make stupid compiler happy */
if ( line )
{
while ( 1 )
{
/* skip current line */
if ( !AFM_STATUS_EOL( stream ) )
afm_stream_read_string( stream );
stream->status = AFM_STREAM_STATUS_NORMAL;
key = afm_stream_read_one( stream );
/* skip empty line */
if ( !key &&
!AFM_STATUS_EOF( stream ) &&
2006-01-16 23:35:33 +01:00
AFM_STATUS_EOL( stream ) )
continue;
break;
}
}
else
{
while ( 1 )
{
/* skip current column */
while ( !AFM_STATUS_EOC( stream ) )
afm_stream_read_one( stream );
stream->status = AFM_STREAM_STATUS_NORMAL;
key = afm_stream_read_one( stream );
/* skip empty column */
if ( !key &&
!AFM_STATUS_EOF( stream ) &&
2006-01-16 23:35:33 +01:00
AFM_STATUS_EOC( stream ) )
continue;
break;
}
}
if ( len )
*len = ( key ) ? (FT_Offset)AFM_STREAM_KEY_LEN( stream, key )
: 0;
return key;
}
static AFM_Token
afm_tokenize( const char* key,
FT_Offset len )
{
int n;
for ( n = 0; n < N_AFM_TOKENS; n++ )
{
if ( *( afm_key_table[n] ) == *key )
{
for ( ; n < N_AFM_TOKENS; n++ )
{
if ( *( afm_key_table[n] ) != *key )
return AFM_TOKEN_UNKNOWN;
if ( ft_strncmp( afm_key_table[n], key, len ) == 0 )
return (AFM_Token) n;
}
}
}
return AFM_TOKEN_UNKNOWN;
}
FT_LOCAL_DEF( FT_Error )
afm_parser_init( AFM_Parser parser,
FT_Memory memory,
FT_Byte* base,
FT_Byte* limit )
{
AFM_Stream stream = NULL;
FT_Error error;
if ( FT_NEW( stream ) )
return error;
2006-01-16 23:35:33 +01:00
stream->cursor = stream->base = base;
stream->limit = limit;
2006-01-16 23:35:33 +01:00
/* don't skip the first line during the first call */
stream->status = AFM_STREAM_STATUS_EOL;
parser->memory = memory;
parser->stream = stream;
parser->FontInfo = NULL;
parser->get_index = NULL;
return FT_Err_Ok;
}
FT_LOCAL( void )
afm_parser_done( AFM_Parser parser )
{
FT_Memory memory = parser->memory;
FT_FREE( parser->stream );
}
static FT_Error
afm_parser_read_int( AFM_Parser parser,
FT_Int* aint )
{
AFM_ValueRec val;
val.type = AFM_VALUE_TYPE_INTEGER;
if ( afm_parser_read_vals( parser, &val, 1 ) == 1 )
{
*aint = val.u.i;
return FT_Err_Ok;
}
else
return FT_THROW( Syntax_Error );
}
static FT_Error
afm_parse_track_kern( AFM_Parser parser )
{
AFM_FontInfo fi = parser->FontInfo;
AFM_Stream stream = parser->stream;
AFM_TrackKern tk;
char* key;
FT_Offset len;
int n = -1;
FT_Int tmp;
if ( afm_parser_read_int( parser, &tmp ) )
goto Fail;
if ( tmp < 0 )
{
FT_ERROR(( "afm_parse_track_kern: invalid number of track kerns\n" ));
goto Fail;
}
fi->NumTrackKern = (FT_UInt)tmp;
FT_TRACE3(( "afm_parse_track_kern: %u track kern%s expected\n",
fi->NumTrackKern,
fi->NumTrackKern == 1 ? "" : "s" ));
/* Rough sanity check: The minimum line length of the `TrackKern` */
/* command is 20 characters (including the EOL character). */
if ( ( stream->limit - stream->cursor ) / 20 < fi->NumTrackKern )
{
FT_ERROR(( "afm_parse_track_kern:"
" number of track kern entries exceeds stream size\n" ));
goto Fail;
}
if ( fi->NumTrackKern )
{
FT_Memory memory = parser->memory;
FT_Error error;
if ( FT_QNEW_ARRAY( fi->TrackKerns, fi->NumTrackKern ) )
return error;
}
while ( ( key = afm_parser_next_key( parser, 1, &len ) ) != 0 )
{
AFM_ValueRec shared_vals[5];
switch ( afm_tokenize( key, len ) )
{
case AFM_TOKEN_TRACKKERN:
n++;
if ( n >= (int)fi->NumTrackKern )
{
FT_ERROR(( "afm_parse_track_kern: too many track kern data\n" ));
goto Fail;
}
tk = fi->TrackKerns + n;
shared_vals[0].type = AFM_VALUE_TYPE_INTEGER;
shared_vals[1].type = AFM_VALUE_TYPE_FIXED;
shared_vals[2].type = AFM_VALUE_TYPE_FIXED;
shared_vals[3].type = AFM_VALUE_TYPE_FIXED;
shared_vals[4].type = AFM_VALUE_TYPE_FIXED;
if ( afm_parser_read_vals( parser, shared_vals, 5 ) != 5 )
{
FT_ERROR(( "afm_parse_track_kern:"
" insufficient number of parameters for entry %d\n",
n ));
goto Fail;
}
tk->degree = shared_vals[0].u.i;
tk->min_ptsize = shared_vals[1].u.f;
tk->min_kern = shared_vals[2].u.f;
tk->max_ptsize = shared_vals[3].u.f;
tk->max_kern = shared_vals[4].u.f;
break;
case AFM_TOKEN_ENDTRACKKERN:
case AFM_TOKEN_ENDKERNDATA:
case AFM_TOKEN_ENDFONTMETRICS:
tmp = n + 1;
if ( (FT_UInt)tmp != fi->NumTrackKern )
{
FT_TRACE1(( "afm_parse_track_kern: %s%d track kern entr%s seen\n",
tmp == 0 ? "" : "only ",
tmp,
tmp == 1 ? "y" : "ies" ));
fi->NumTrackKern = (FT_UInt)tmp;
}
else
FT_TRACE3(( "afm_parse_track_kern: %d track kern entr%s seen\n",
tmp,
tmp == 1 ? "y" : "ies" ));
return FT_Err_Ok;
case AFM_TOKEN_UNKNOWN:
break;
default:
goto Fail;
}
}
Fail:
return FT_THROW( Syntax_Error );
}
#undef KERN_INDEX
#define KERN_INDEX( g1, g2 ) ( ( (FT_ULong)g1 << 16 ) | g2 )
2006-01-16 23:35:33 +01:00
/* compare two kerning pairs */
FT_COMPARE_DEF( int )
afm_compare_kern_pairs( const void* a,
const void* b )
{
AFM_KernPair kp1 = (AFM_KernPair)a;
AFM_KernPair kp2 = (AFM_KernPair)b;
FT_ULong index1 = KERN_INDEX( kp1->index1, kp1->index2 );
FT_ULong index2 = KERN_INDEX( kp2->index1, kp2->index2 );
if ( index1 > index2 )
return 1;
else if ( index1 < index2 )
return -1;
else
return 0;
}
static FT_Error
afm_parse_kern_pairs( AFM_Parser parser )
{
AFM_FontInfo fi = parser->FontInfo;
AFM_Stream stream = parser->stream;
AFM_KernPair kp;
char* key;
FT_Offset len;
int n = -1;
FT_Int tmp;
if ( afm_parser_read_int( parser, &tmp ) )
goto Fail;
if ( tmp < 0 )
{
FT_ERROR(( "afm_parse_kern_pairs: invalid number of kern pairs\n" ));
goto Fail;
}
fi->NumKernPair = (FT_UInt)tmp;
FT_TRACE3(( "afm_parse_kern_pairs: %u kern pair%s expected\n",
fi->NumKernPair,
fi->NumKernPair == 1 ? "" : "s" ));
/* Rough sanity check: The minimum line length of the `KP`, */
/* `KPH`,`KPX`, and `KPY` commands is 10 characters (including */
/* the EOL character). */
if ( ( stream->limit - stream->cursor ) / 10 < fi->NumKernPair )
{
FT_ERROR(( "afm_parse_kern_pairs:"
" number of kern pairs exceeds stream size\n" ));
goto Fail;
}
if ( fi->NumKernPair )
{
FT_Memory memory = parser->memory;
FT_Error error;
if ( FT_QNEW_ARRAY( fi->KernPairs, fi->NumKernPair ) )
return error;
}
while ( ( key = afm_parser_next_key( parser, 1, &len ) ) != 0 )
{
AFM_Token token = afm_tokenize( key, len );
switch ( token )
{
case AFM_TOKEN_KP:
case AFM_TOKEN_KPX:
case AFM_TOKEN_KPY:
{
FT_Int r;
AFM_ValueRec shared_vals[4];
n++;
if ( n >= (int)fi->NumKernPair )
{
FT_ERROR(( "afm_parse_kern_pairs: too many kern pairs\n" ));
goto Fail;
}
kp = fi->KernPairs + n;
shared_vals[0].type = AFM_VALUE_TYPE_INDEX;
shared_vals[1].type = AFM_VALUE_TYPE_INDEX;
shared_vals[2].type = AFM_VALUE_TYPE_INTEGER;
shared_vals[3].type = AFM_VALUE_TYPE_INTEGER;
r = afm_parser_read_vals( parser, shared_vals, 4 );
if ( r < 3 )
{
FT_ERROR(( "afm_parse_kern_pairs:"
" insufficient number of parameters for entry %d\n",
n ));
goto Fail;
}
/* index values can't be negative */
kp->index1 = shared_vals[0].u.u;
kp->index2 = shared_vals[1].u.u;
if ( token == AFM_TOKEN_KPY )
{
kp->x = 0;
kp->y = shared_vals[2].u.i;
}
else
{
kp->x = shared_vals[2].u.i;
kp->y = ( token == AFM_TOKEN_KP && r == 4 )
? shared_vals[3].u.i : 0;
}
}
break;
case AFM_TOKEN_ENDKERNPAIRS:
case AFM_TOKEN_ENDKERNDATA:
case AFM_TOKEN_ENDFONTMETRICS:
tmp = n + 1;
if ( (FT_UInt)tmp != fi->NumKernPair )
{
FT_TRACE1(( "afm_parse_kern_pairs: %s%d kern pair%s seen\n",
tmp == 0 ? "" : "only ",
tmp,
tmp == 1 ? "" : "s" ));
fi->NumKernPair = (FT_UInt)tmp;
}
else
FT_TRACE3(( "afm_parse_kern_pairs: %d kern pair%s seen\n",
tmp,
tmp == 1 ? "" : "s" ));
ft_qsort( fi->KernPairs, fi->NumKernPair,
2011-11-30 10:46:53 +01:00
sizeof ( AFM_KernPairRec ),
afm_compare_kern_pairs );
return FT_Err_Ok;
case AFM_TOKEN_UNKNOWN:
break;
default:
goto Fail;
}
}
Fail:
return FT_THROW( Syntax_Error );
}
static FT_Error
afm_parse_kern_data( AFM_Parser parser )
{
FT_Error error;
char* key;
FT_Offset len;
int have_trackkern = 0;
int have_kernpairs = 0;
while ( ( key = afm_parser_next_key( parser, 1, &len ) ) != 0 )
{
switch ( afm_tokenize( key, len ) )
{
case AFM_TOKEN_STARTTRACKKERN:
if ( have_trackkern )
{
FT_ERROR(( "afm_parse_kern_data:"
" invalid second horizontal track kern section\n" ));
goto Fail;
}
error = afm_parse_track_kern( parser );
if ( error )
return error;
have_trackkern = 1;
break;
case AFM_TOKEN_STARTKERNPAIRS:
case AFM_TOKEN_STARTKERNPAIRS0:
if ( have_kernpairs )
{
FT_ERROR(( "afm_parse_kern_data:"
" invalid second horizontal kern pair section\n" ));
goto Fail;
}
error = afm_parse_kern_pairs( parser );
if ( error )
return error;
have_kernpairs = 1;
break;
case AFM_TOKEN_ENDKERNDATA:
case AFM_TOKEN_ENDFONTMETRICS:
return FT_Err_Ok;
case AFM_TOKEN_UNKNOWN:
break;
default:
goto Fail;
}
}
Fail:
return FT_THROW( Syntax_Error );
}
static FT_Error
afm_parser_skip_section( AFM_Parser parser,
FT_Int n,
AFM_Token end_section )
{
char* key;
FT_Offset len;
while ( n-- > 0 )
{
key = afm_parser_next_key( parser, 1, NULL );
if ( !key )
goto Fail;
}
while ( ( key = afm_parser_next_key( parser, 1, &len ) ) != 0 )
{
AFM_Token token = afm_tokenize( key, len );
if ( token == end_section || token == AFM_TOKEN_ENDFONTMETRICS )
return FT_Err_Ok;
}
Fail:
return FT_THROW( Syntax_Error );
}
FT_LOCAL_DEF( FT_Error )
afm_parser_parse( AFM_Parser parser )
{
FT_Memory memory = parser->memory;
AFM_FontInfo fi = parser->FontInfo;
FT_Error error = FT_ERR( Syntax_Error );
char* key;
FT_Offset len;
FT_Int metrics_sets = 0;
if ( !fi )
return FT_THROW( Invalid_Argument );
key = afm_parser_next_key( parser, 1, &len );
if ( !key || len != 16 ||
ft_strncmp( key, "StartFontMetrics", 16 ) != 0 )
return FT_THROW( Unknown_File_Format );
while ( ( key = afm_parser_next_key( parser, 1, &len ) ) != 0 )
{
AFM_ValueRec shared_vals[4];
switch ( afm_tokenize( key, len ) )
{
case AFM_TOKEN_METRICSSETS:
if ( afm_parser_read_int( parser, &metrics_sets ) )
goto Fail;
if ( metrics_sets != 0 && metrics_sets != 2 )
{
error = FT_THROW( Unimplemented_Feature );
goto Fail;
}
break;
case AFM_TOKEN_ISCIDFONT:
shared_vals[0].type = AFM_VALUE_TYPE_BOOL;
if ( afm_parser_read_vals( parser, shared_vals, 1 ) != 1 )
goto Fail;
2006-01-27 15:16:16 +01:00
fi->IsCIDFont = shared_vals[0].u.b;
break;
case AFM_TOKEN_FONTBBOX:
shared_vals[0].type = AFM_VALUE_TYPE_FIXED;
shared_vals[1].type = AFM_VALUE_TYPE_FIXED;
shared_vals[2].type = AFM_VALUE_TYPE_FIXED;
shared_vals[3].type = AFM_VALUE_TYPE_FIXED;
if ( afm_parser_read_vals( parser, shared_vals, 4 ) != 4 )
goto Fail;
fi->FontBBox.xMin = shared_vals[0].u.f;
fi->FontBBox.yMin = shared_vals[1].u.f;
fi->FontBBox.xMax = shared_vals[2].u.f;
fi->FontBBox.yMax = shared_vals[3].u.f;
break;
case AFM_TOKEN_ASCENDER:
shared_vals[0].type = AFM_VALUE_TYPE_FIXED;
if ( afm_parser_read_vals( parser, shared_vals, 1 ) != 1 )
goto Fail;
fi->Ascender = shared_vals[0].u.f;
break;
case AFM_TOKEN_DESCENDER:
shared_vals[0].type = AFM_VALUE_TYPE_FIXED;
if ( afm_parser_read_vals( parser, shared_vals, 1 ) != 1 )
goto Fail;
fi->Descender = shared_vals[0].u.f;
break;
case AFM_TOKEN_STARTCHARMETRICS:
{
FT_Int n = 0;
if ( afm_parser_read_int( parser, &n ) )
goto Fail;
error = afm_parser_skip_section( parser, n,
AFM_TOKEN_ENDCHARMETRICS );
if ( error )
return error;
}
break;
case AFM_TOKEN_STARTKERNDATA:
error = afm_parse_kern_data( parser );
if ( error )
goto Fail;
/* we only support kern data, so ... */
/* fall through */
case AFM_TOKEN_ENDFONTMETRICS:
return FT_Err_Ok;
default:
break;
}
}
Fail:
FT_FREE( fi->TrackKerns );
fi->NumTrackKern = 0;
FT_FREE( fi->KernPairs );
fi->NumKernPair = 0;
fi->IsCIDFont = 0;
return error;
}
2006-01-16 23:35:33 +01:00
#else /* T1_CONFIG_OPTION_NO_AFM */
/* ANSI C doesn't like empty source files */
typedef int _afm_parse_dummy;
#endif /* T1_CONFIG_OPTION_NO_AFM */
2006-01-16 23:35:33 +01:00
/* END */