freetype2/src/cid/cidload.c

504 lines
13 KiB
C
Raw Normal View History

/*******************************************************************
*
* cidload.h 2.0
*
* CID-keyed foint loader
*
* Copyright 1996-2000 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.
*
*
* This is the new and improved Type 1 data loader for FreeType 2.
* The old loader has several problems: it is slow, complex, difficult
* to maintain, and contains incredible hacks to make it accept some
* ill-formed Type 1 fonts without hiccup-ing. Moreover, about 5%
* of the Type 1 fonts on my machine still aren't loaded correctly
* with it.
*
* This version is much simpler, much faster and also easier to
* read and maintain by a great order of magnitude. The idea behind
* it is to _not_ try to read the Type 1 token stream with a state
* machine (i.e. a Postscript-like interpreter) but rather to perform
* simple pattern-matching.
*
* Indeed, nearly all data definitions follow a simple pattern
* like :
*
* ..... /Field <data> ....
*
* where <data> can be a number, a boolean, a string, or an
* array of numbers. There are a few exceptions, namely the
* encoding, font name, charstrings and subrs and they are
* handled with a special pattern-matching routine.
*
* All other common cases are handled very simply. The matching
* rules are defined in the file "t1tokens.h" through the use
* of several macros calls T1_FIELD_XXX
*
* The function "parse_dict" simply scans *linearly* a given
* dictionary (either the top-level or private one) and calls
* the appropriate callback when it encounters an immediate
* keyword.
*
* This is by far the fastest way one can find to parse and read
* all data :-)
*
* This led to tremendous code size reduction. Note that later,
* the glyph loader will also be _greatly_ simplified, and the
* automatic hinter will replace the clumsy "t1hinter"..
*
******************************************************************/
#include <freetype/internal/ftdebug.h>
#include <freetype/config/ftconfig.h>
#include <freetype/ftmm.h>
#include <freetype/internal/t1types.h>
#include <freetype/internal/t1errors.h>
#include <cidload.h>
#include <stdio.h>
#undef FT_COMPONENT
#define FT_COMPONENT trace_t1load
/* reads a single offset */
LOCAL_FUNC
FT_Long cid_get_offset( FT_Byte** start, FT_Byte offsize )
{
FT_Long result;
FT_Byte* p = *start;
for ( result = 0; offsize > 0; offsize-- )
result = (result << 8) | *p++;
*start = p;
return result;
}
LOCAL_FUNC
void cid_decrypt( FT_Byte* buffer,
FT_Int length,
FT_UShort seed )
{
while ( length > 0 )
{
FT_Byte plain;
plain = (*buffer ^ (seed >> 8));
seed = (*buffer+seed)*52845+22719;
*buffer++ = plain;
length--;
}
}
/***************************************************************************/
/***************************************************************************/
/***** *****/
/***** TYPE 1 SYMBOL PARSING *****/
/***** *****/
/***************************************************************************/
/***************************************************************************/
static FT_Error cid_load_keyword( CID_Face face,
CID_Loader* loader,
const T1_Field_Rec* keyword )
{
FT_Error error;
CID_Parser* parser = &loader->parser;
FT_Byte* object;
CID_Info* cid = &face->cid;
/* if the keyword has a dedicated callback, call it */
if (keyword->type == t1_field_callback)
{
error = keyword->reader( face, parser );
goto Exit;
}
/* we must now compute the address of our target object */
switch (keyword->location)
{
case t1_field_cid_info:
object = (FT_Byte*)cid;
break;
case t1_field_font_info:
object = (FT_Byte*)&cid->font_info;
break;
default:
{
CID_FontDict* dict;
if ( parser->num_dict < 0 )
{
FT_ERROR(( "CID.Load_Keyword: invalid use of '%s' !!\n", keyword->ident ));
error = T1_Err_Syntax_Error;
goto Exit;
}
dict = cid->font_dicts + parser->num_dict;
switch (keyword->location)
{
case t1_field_private:
object = (FT_Byte*)&dict->private_dict;
break;
default:
object = (FT_Byte*)dict;
}
}
}
/* now, load the keyword data in the object's field(s) */
if ( keyword->type == t1_field_integer_array ||
keyword->type == t1_field_fixed_array )
error = CID_Load_Field_Table( parser, keyword, object );
else
error = CID_Load_Field( parser, keyword, object );
Exit:
return error;
}
static
FT_Error parse_font_bbox( CID_Face face, CID_Parser* parser )
{
FT_Short temp[4];
T1_BBox* bbox = &face->cid.font_bbox;
(void)CID_ToCoordArray( parser, 4, temp );
bbox->xMin = temp[0];
bbox->yMin = temp[1];
bbox->xMax = temp[2];
bbox->yMax = temp[3];
return 0;
}
static
FT_Error parse_font_matrix( CID_Face face, CID_Parser* parser )
{
FT_Matrix* matrix;
CID_FontDict* dict;
T1_Fixed temp[4];
if (parser->num_dict >= 0)
{
dict = face->cid.font_dicts + parser->num_dict;
matrix = &dict->font_matrix;
(void)CID_ToFixedArray( parser, 4, temp, 3 );
matrix->xx = temp[0];
matrix->yx = temp[1];
matrix->xy = temp[2];
matrix->yy = temp[3];
}
return 0;
}
static
FT_Error parse_fd_array( CID_Face face, CID_Parser* parser )
{
CID_Info* cid = &face->cid;
FT_Memory memory = face->root.memory;
FT_Error error;
FT_Long num_dicts;
num_dicts = CID_ToInt(parser);
if ( !cid->font_dicts )
{
FT_Int n;
if ( ALLOC_ARRAY( cid->font_dicts, num_dicts, CID_FontDict ) )
goto Exit;
cid->num_dicts = (FT_UInt)num_dicts;
/* don't forget to set a few defauts !! */
for ( n = 0; n < cid->num_dicts; n++ )
{
CID_FontDict* dict = cid->font_dicts + n;
/* default value for lenIV !! */
dict->private_dict.lenIV = 4;
}
}
Exit:
return error;
}
static
const T1_Field_Rec t1_field_records[] =
{
#include <cidtokens.h>
{ 0, 0, 0, 0, 0, 0 }
};
static
int is_space( char c )
{
return ( c == ' ' || c == '\t' || c == '\r' || c == '\n' );
}
static
int is_alpha( char c )
{
return ( (c >= 'A' && c <= 'Z') ||
(c >= 'a' && c <= 'z') ||
(c >= '0' && c <= '9') ||
(c == '.') ||
(c == '_') );
}
static
void skip_whitespace( CID_Parser* parser )
{
FT_Byte* cur = parser->cursor;
while ( cur < parser->limit && is_space(*cur) )
cur++;
parser->cursor = cur;
}
static
FT_Error parse_dict( CID_Face face,
CID_Loader* loader,
FT_Byte* base,
FT_Long size )
{
CID_Parser* parser = &loader->parser;
parser->cursor = base;
parser->limit = base + size;
parser->error = 0;
{
FT_Byte* cur = base;
FT_Byte* limit = cur + size;
for ( ;cur < limit; cur++ )
{
/* look for %ADOBegin... */
if ( *cur == '%' && cur + 20 < limit &&
strncmp( (char*)cur, "%ADOBeginFontDict", 17 ) == 0 )
{
cur += 17;
/* if /FDArray was found, then cid->num_dicts is > 0, and */
/* we can start increasing parser->num_dict */
if ( face->cid.num_dicts > 0 )
parser->num_dict++;
}
/* look for immediates */
else if (*cur == '/' && cur+2 < limit)
{
FT_Byte* cur2;
FT_Int len;
cur ++;
cur2 = cur;
while (cur2 < limit && is_alpha(*cur2)) cur2++;
len = cur2-cur;
if (len > 0 && len < 22)
{
/* now, compare the immediate name to the keyword table */
const T1_Field_Rec* keyword = t1_field_records;
for (;;)
{
FT_Byte* name;
name = (FT_Byte*)keyword->ident;
if (!name) break;
if ( cur[0] == name[0] &&
len == (FT_Int)strlen((const char*)name) )
{
FT_Int n;
for ( n = 1; n < len; n++ )
if (cur[n] != name[n])
break;
if (n >= len)
{
/* we found it - run the parsing callback !! */
parser->cursor = cur2;
skip_whitespace( parser );
parser->error = cid_load_keyword( face, loader, keyword );
if (parser->error)
return parser->error;
cur = parser->cursor;
break;
}
}
keyword++;
}
}
}
}
}
return parser->error;
}
/* read the subrmap and the subrs of each font dict */
static
FT_Error cid_read_subrs( CID_Face face )
{
CID_Info* cid = &face->cid;
FT_Memory memory = face->root.memory;
FT_Stream stream = face->root.stream;
FT_Error error;
FT_UInt n;
CID_Subrs* subr;
FT_UInt max_offsets = 0;
FT_ULong* offsets = 0;
if ( ALLOC_ARRAY( face->subrs, cid->num_dicts, CID_Subrs ) )
goto Exit;
subr = face->subrs;
for ( n = 0; n < cid->num_dicts; n++, subr++ )
{
CID_FontDict* dict = cid->font_dicts + n;
FT_UInt count, num_subrs = dict->num_subrs;
FT_ULong data_len;
FT_Byte* p;
/* reallocate offsets array if needed */
if ( num_subrs+1 > max_offsets )
{
FT_UInt new_max = (num_subrs+1+3) & -4;
if ( REALLOC_ARRAY( offsets, max_offsets, new_max, FT_ULong ) )
goto Fail;
max_offsets = new_max;
}
/* read the subrmap's offsets */
if ( FILE_Seek( cid->data_offset + dict->subrmap_offset ) ||
ACCESS_Frame( (num_subrs+1) * dict->sd_bytes ) )
goto Fail;
p = (FT_Byte*)stream->cursor;
for ( count = 0; count <= num_subrs; count++ )
offsets[count] = cid_get_offset( &p, dict->sd_bytes );
FORGET_Frame();
/* now, compute the size of subrs charstrings, allocate and read them */
data_len = offsets[num_subrs] - offsets[0];
if ( ALLOC_ARRAY( subr->code, num_subrs+1, FT_Byte* ) ||
ALLOC( subr->code[0], data_len ) )
goto Fail;
if ( FILE_Seek( cid->data_offset + offsets[0] ) ||
FILE_Read( subr->code[0], data_len ) )
goto Exit;
/* set up pointers */
for ( count = 1; count <= num_subrs; count++ )
{
FT_UInt len;
len = offsets[count] - offsets[count-1];
subr->code[count] = subr->code[count-1] + len;
}
/* decrypt subroutines */
for ( count = 0; count < num_subrs; count++ )
{
FT_UInt len;
len = offsets[count+1] - offsets[count];
cid_decrypt( subr->code[count], len, 4330 );
}
subr->num_subrs = num_subrs;
}
Exit:
FREE( offsets );
return error;
Fail:
if (face->subrs)
{
for ( n = 0; n < cid->num_dicts; n++ )
{
if (face->subrs[n].code)
FREE( face->subrs[n].code[0] );
FREE( face->subrs[n].code );
}
FREE( face->subrs );
}
goto Exit;
}
static
void t1_init_loader( CID_Loader* loader, CID_Face face )
{
UNUSED(face);
MEM_Set( loader, 0, sizeof(*loader) );
}
static
void t1_done_loader( CID_Loader* loader )
{
CID_Parser* parser = &loader->parser;
/* finalize parser */
CID_Done_Parser( parser );
}
LOCAL_FUNC
FT_Error CID_Open_Face( CID_Face face )
{
CID_Loader loader;
CID_Parser* parser;
FT_Error error;
t1_init_loader( &loader, face );
parser = &loader.parser;
error = CID_New_Parser( parser, face->root.stream, face->root.memory );
if (error) goto Exit;
error = parse_dict( face, &loader,
parser->postscript,
parser->postscript_len );
if (error) goto Exit;
face->cid.data_offset = loader.parser.data_offset;
error = cid_read_subrs( face );
Exit:
t1_done_loader( &loader );
return error;
}