mirror of https://github.com/tongzx/nt5src
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
1374 lines
40 KiB
1374 lines
40 KiB
/***
|
|
**
|
|
** Module: T1Parser
|
|
**
|
|
** Description:
|
|
** This is a module of the T1 to TT font converter. The module
|
|
** will extract information from a T1 font file, by parsing
|
|
** the data/commands found in PFB, PFM and AFM files.
|
|
**
|
|
** Author: Michael Jansson
|
|
**
|
|
** Created: 5/26/93
|
|
**
|
|
***/
|
|
|
|
|
|
/**** INCLUDES */
|
|
/* General types and definitions. */
|
|
#include <string.h>
|
|
#include "types.h"
|
|
|
|
/* Special types and definitions. */
|
|
#include "safemem.h"
|
|
#include "encoding.h"
|
|
#include "metrics.h"
|
|
#include "t1msg.h"
|
|
|
|
/* Module dependent types and prototypes. */
|
|
#include "titott.h"
|
|
#include "t1parser.h"
|
|
#include "charstr.h"
|
|
#include "freader.h"
|
|
#include "mreader.h"
|
|
|
|
|
|
/***** CONSTANTS */
|
|
#define ONE (USHORT)1
|
|
#define BUFLEN (USHORT)512
|
|
|
|
#define PS_ANGLE "/ItalicAngle"
|
|
#define PS_ARRAY "array"
|
|
#define PS_BEGIN "begin"
|
|
#define PS_BLUEFUZZ "/BlueFuzz"
|
|
#define PS_BLUESCALE "/BlueScale"
|
|
#define PS_BLUESHIFT "/BlueShift"
|
|
#define PS_BLUEVALUES "/BlueValues"
|
|
#define PS_CHARSTRINGS "/CharStrings"
|
|
#define PS_COPYRIGHT "/Copyright"
|
|
#define PS_DATE "%%CreationDate:"
|
|
#define PS_DUP "dup"
|
|
#define PS_ENCODING "/Encoding"
|
|
#define PS_END "end"
|
|
#define PS_FAMILY "/FamilyName"
|
|
#define PS_FAMILYBLUES "/FamilyBlues"
|
|
#define PS_FAMILYOTHERBLUES "/FamilyOtherBlues"
|
|
#define PS_FONTMATRIX "/FontMatrix"
|
|
#define PS_FORCEBOLD "/ForceBold"
|
|
#define PS_FULLNAME "/FullName"
|
|
#define PS_HYBRID "hires"
|
|
#define PS_ISFIXED "/isFixedPitch"
|
|
#define PS_LENIV "/lenIV"
|
|
#define PS_NAME "/FontName"
|
|
#define PS_NOACCESS "noaccess"
|
|
#define PS_NOTICE "/Notice"
|
|
#define PS_OTHERBLUES "/OtherBlues"
|
|
#define PS_SNAPH "/StemSnapH"
|
|
#define PS_SNAPV "/StemSnapV"
|
|
#define PS_STDENCODING "StandardEncoding"
|
|
#define PS_STDVW "/StdVW"
|
|
#define PS_STDHW "/StdHW"
|
|
#define PS_SUBRS "/Subrs"
|
|
#define PS_UNDERLINE "/UnderlinePosition"
|
|
#define PS_UTHICK "/UnderlineThickness"
|
|
#define PS_ID "/UniqueID"
|
|
#define PS_VERSION "/version"
|
|
#define PS_WEIGHT "/Weight"
|
|
|
|
|
|
|
|
/***** LOCAL TYPES */
|
|
|
|
struct T1Handle {
|
|
struct FontFile *ff;
|
|
|
|
struct PSState *ps;
|
|
|
|
struct Subrs stdenc[256];
|
|
|
|
USHORT numsubrs;
|
|
struct Subrs *subrs;
|
|
USHORT leniv;
|
|
struct T1Metrics t1m;
|
|
};
|
|
|
|
|
|
/***** MACROS */
|
|
/*-none-*/
|
|
|
|
|
|
/***** PROTOTYPES */
|
|
/*-none-*/
|
|
|
|
/***** STATIC FUNCTIONS */
|
|
|
|
|
|
/***
|
|
** Function: StrToFix
|
|
**
|
|
** Description:
|
|
** This is a "strtod" function, that converts from
|
|
** ascii to fixpoint numbers.
|
|
***/
|
|
static long StrToFix(char *str, char **out, const long base)
|
|
{
|
|
char *fstr;
|
|
long num = 0, frac = 0, exp = 0;
|
|
|
|
if (out)
|
|
(*out) = str;
|
|
|
|
/* Skip white space. */
|
|
while (*str && (*str==' ' || *str=='\t'))
|
|
str++;
|
|
|
|
/* A number? */
|
|
if (*str && ((*str>='0' && *str<='9') || *str=='-') || *str=='.') {
|
|
|
|
num = atoi(str)*base;
|
|
|
|
/* Fraction? */
|
|
fstr = strchr(str, '.');
|
|
if (fstr!=NULL && (strchr(str, ' ')==NULL || fstr<strchr(str, ' '))) {
|
|
|
|
do {
|
|
fstr++;
|
|
} while (*fstr>='0' && *fstr<='9');
|
|
|
|
/* Exponent? */
|
|
if (*fstr=='E')
|
|
exp = atoi(fstr+1);
|
|
else
|
|
exp = 0;
|
|
|
|
fstr--;
|
|
while (*fstr!='.') {
|
|
frac += ((*fstr)-'0')*base;
|
|
frac /= 10;
|
|
fstr--;
|
|
}
|
|
if (num<0)
|
|
num -= frac;
|
|
else
|
|
num += frac;
|
|
|
|
/* Handle exponent. */
|
|
if (exp>0) {
|
|
do {
|
|
num *= 10;
|
|
} while (--exp);
|
|
} else if (exp<0) {
|
|
do {
|
|
num /= 10;
|
|
} while (++exp);
|
|
}
|
|
}
|
|
|
|
/* Skip digits. */
|
|
while (*str && ((*str>='0' && *str<='9') ||
|
|
*str=='.' || *str=='-' || *str=='E'))
|
|
str++;
|
|
|
|
if (out)
|
|
(*out) = str;
|
|
}
|
|
|
|
return num;
|
|
}
|
|
|
|
|
|
|
|
/***
|
|
** Function: FreeT1Composite
|
|
**
|
|
** Description:
|
|
** This function frees the memory used to represent
|
|
** a composite acented T1 glyph.
|
|
***/
|
|
static void FreeT1Composite(Composite *comp)
|
|
{
|
|
if (comp) {
|
|
if (comp->cchar)
|
|
Free(comp->cchar);
|
|
Free(comp);
|
|
}
|
|
}
|
|
|
|
|
|
/***
|
|
** Function: UseGlyph
|
|
**
|
|
** Description:
|
|
** This function determines whether a glyph should be
|
|
** converted or not, based on the name of the glyph
|
|
** and a specification of the desired glyphs.
|
|
***/
|
|
static int CDECL compare(const void *arg1, const void *arg2)
|
|
{
|
|
return strcmp( *((const char **)arg1), *((const char **)arg2) );
|
|
}
|
|
static boolean UseGlyph(const struct GlyphFilter *filter,
|
|
Composite *comp,
|
|
const char *name)
|
|
{
|
|
boolean found = FALSE;
|
|
char **result;
|
|
|
|
/* Check if the glyph is explicitly specified. */
|
|
if (filter) {
|
|
|
|
result = (char **)bsearch((char *)&name,
|
|
(char *)filter->name, filter->num,
|
|
sizeof(char *),
|
|
compare);
|
|
|
|
|
|
found = (boolean)(result!=NULL);
|
|
|
|
/* Check if the glyph is specified indirectly through an accented */
|
|
/* composite glyph. */
|
|
if (!found) {
|
|
Composite *c;
|
|
|
|
for (c=comp; c &&
|
|
strcmp(name, c->achar) &&
|
|
strcmp(name, c->bchar); c = c->next);
|
|
found = (boolean)(c!=NULL);
|
|
}
|
|
} else {
|
|
found = TRUE;
|
|
}
|
|
|
|
return found;
|
|
}
|
|
|
|
|
|
|
|
/***
|
|
** Function: ReadFontMatrix
|
|
**
|
|
** Description:
|
|
** Read the command sequence "/FontMatrix[%d %d %d %d]" and
|
|
** record the transformation matrix in the T1 handle.
|
|
***/
|
|
static errcode ReadFontMatrix(struct T1Handle *t1,
|
|
char *str,
|
|
const USHORT len)
|
|
{
|
|
errcode status=SUCCESS;
|
|
f16d16 fmatrix[6];
|
|
USHORT i;
|
|
|
|
if (GetSeq(t1->ff, str, len)) {
|
|
for (i=0; i<6; i++)
|
|
fmatrix[i] = StrToFix(str, &str, F16D16BASE);
|
|
|
|
/* Check if we have the default matrix. */ /*lint -e771 */
|
|
if (fmatrix[2]!=0 ||
|
|
fmatrix[4]!=0 ||
|
|
fmatrix[1]!=0 ||
|
|
fmatrix[5]!=0 ||
|
|
fmatrix[0]!=F16D16PPM ||
|
|
fmatrix[3]!=F16D16PPM ||
|
|
t1->t1m.upem!=2048) { /*lint +e771 */ /* fmatrix[] IS initialized */
|
|
|
|
if ((t1->t1m.fmatrix = Malloc(sizeof(f16d16)*6))==NULL) {
|
|
SetError(status = NOMEM);
|
|
} else {
|
|
t1->t1m.fmatrix[0] = fmatrix[0];
|
|
t1->t1m.fmatrix[1] = fmatrix[1];
|
|
t1->t1m.fmatrix[2] = fmatrix[2];
|
|
t1->t1m.fmatrix[3] = fmatrix[3];
|
|
t1->t1m.fmatrix[4] = fmatrix[4];
|
|
t1->t1m.fmatrix[5] = fmatrix[5];
|
|
}
|
|
} else {
|
|
t1->t1m.fmatrix = NULL;
|
|
}
|
|
} else {
|
|
SetError(status = BADINPUTFILE);
|
|
}
|
|
|
|
return status;
|
|
}
|
|
|
|
|
|
/***
|
|
** Function: ReadEncodingArray
|
|
**
|
|
** Description:
|
|
** Read the command sequence "/Encoding %d array ..." and
|
|
** build an encoding table, or read "/Encoding StdEncoding def"
|
|
** and used the standard encoding table.
|
|
***/
|
|
static errcode ReadEncodingArray(struct T1Handle *t1,
|
|
char *str,
|
|
const USHORT len)
|
|
{
|
|
errcode status = SUCCESS;
|
|
USHORT codes[ENC_MAXCODES];
|
|
char *glyph_name = NULL;
|
|
USHORT i, index;
|
|
|
|
if (Get_Token(t1->ff, str, len)==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else {
|
|
if (strcmp(str, PS_STDENCODING) &&
|
|
((t1->t1m.encSize=(USHORT)atoi(str))!=0)) {
|
|
if ((t1->t1m.encoding = AllocEncodingTable(t1->t1m.encSize))==NULL) {
|
|
SetError(status = NOMEM);
|
|
} else {
|
|
|
|
/* Skip leading proc. */
|
|
while (Get_Token(t1->ff, str, len) && strcmp(str, PS_DUP));
|
|
|
|
/* Read the encoding entries: "<n> <str> put <comment>\n dup" */
|
|
for (i=0; i<t1->t1m.encSize; i++) {
|
|
|
|
/* Get character code. */
|
|
(void)Get_Token(t1->ff, str, len);
|
|
if (str[0]=='8' && str[1]=='#') { /* Octal? */
|
|
index = (USHORT)atoi(&str[2]);
|
|
index = (USHORT)((index/10)*8 + (index%8));
|
|
} else {
|
|
index = (USHORT)atoi(str);
|
|
}
|
|
|
|
/* Get character name. */
|
|
(void)Get_Token(t1->ff, str, len);
|
|
|
|
codes[ENC_MSWINDOWS] = index;
|
|
codes[ENC_UNICODE] = index;
|
|
|
|
if (index<256) {
|
|
codes[ENC_STANDARD] = index;
|
|
codes[ENC_MACCODES] = index;
|
|
} else {
|
|
codes[ENC_STANDARD] = NOTDEFCODE;
|
|
codes[ENC_MACCODES] = NOTDEFCODE;
|
|
}
|
|
if ((glyph_name = Strdup(&str[1]))!=NULL)
|
|
SetEncodingEntry(t1->t1m.encoding, i,
|
|
glyph_name,
|
|
ENC_MAXCODES,
|
|
codes);
|
|
else {
|
|
status = NOMEM;
|
|
break;
|
|
}
|
|
|
|
(void)Get_Token(t1->ff, str, len); /* Pop "dup" */
|
|
(void)Get_Token(t1->ff, str, len); /* Pop "put" or comment. */
|
|
if (str[0]=='%') {
|
|
(void)GetNewLine(t1->ff, str, len);
|
|
(void)Get_Token(t1->ff, str, len); /* Pop "put". */
|
|
}
|
|
|
|
if (strcmp(str, PS_DUP))
|
|
break;
|
|
}
|
|
t1->t1m.encSize = (USHORT)(i+1);
|
|
|
|
/* Rehash the table. */
|
|
RehashEncodingTable(t1->t1m.encoding, t1->t1m.encSize);
|
|
}
|
|
}
|
|
}
|
|
|
|
return status;
|
|
}
|
|
|
|
|
|
/***
|
|
** Function: ReadArray
|
|
**
|
|
** Description:
|
|
** Read an array.
|
|
***/
|
|
static errcode ReadArray(struct T1Handle *t1,
|
|
char *str,
|
|
const USHORT len,
|
|
funit *array,
|
|
USHORT maxarr,
|
|
USHORT *cnt)
|
|
{
|
|
errcode status;
|
|
char *nxt;
|
|
|
|
if (GetSeq(t1->ff, str, len)) {
|
|
(*cnt)=0;
|
|
do {
|
|
array[(*cnt)] = (funit)(((StrToFix(str, &nxt, 4L)+8002)>>2) - 2000);
|
|
if (nxt==str)
|
|
break;
|
|
str = nxt;
|
|
} while (++(*cnt)<maxarr);
|
|
status=SUCCESS;
|
|
} else {
|
|
SetError(status = BADINPUTFILE);
|
|
}
|
|
|
|
return status;
|
|
}
|
|
|
|
|
|
|
|
/***
|
|
** Function: ReadFontSubrs
|
|
**
|
|
** Description:
|
|
** Read the command sequence "/Subrs %d array dup %d %d RD %x ND ...",
|
|
** decode and decrypt the subroutines and store them in the T1
|
|
** handle.
|
|
***/
|
|
static errcode ReadFontSubrs(struct T1Handle *t1,
|
|
char *str, const USHORT len)
|
|
{
|
|
errcode status = SUCCESS;
|
|
USHORT index,i,j;
|
|
USHORT count = 0;
|
|
USHORT r;
|
|
short b;
|
|
|
|
/* Get the number of subroutines. */
|
|
if (Get_Token(t1->ff, str, len)==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else {
|
|
count = (USHORT)atoi(str);
|
|
|
|
/* Get the "array" keyword". */
|
|
if ((Get_Token(t1->ff, str, len)==NULL) || strcmp(str, PS_ARRAY)) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else {
|
|
if ((t1->subrs = Malloc((USHORT)sizeof(struct Subrs)*count))==NULL) {
|
|
SetError(status = NOMEM);
|
|
} else {
|
|
memset(t1->subrs, '\0', sizeof(struct Subrs)*count);
|
|
t1->numsubrs = count;
|
|
for (i=0; i<count; i++) {
|
|
|
|
if (Get_Token(t1->ff, str, len)==NULL) { /* Get "dup" */
|
|
SetError(status = BADINPUTFILE);
|
|
break;
|
|
}
|
|
if (strcmp(str, PS_DUP)) {
|
|
SetError(status = BADT1HEADER);
|
|
break;
|
|
}
|
|
|
|
if (Get_Token(t1->ff, str, len)==NULL) { /* Get Subr index. */
|
|
SetError(status=BADINPUTFILE);
|
|
break;
|
|
}
|
|
index = (USHORT)atoi(str);
|
|
if (t1->subrs[index].code) {
|
|
LogError(MSG_WARNING, MSG_DBLIDX, NULL);
|
|
Free(t1->subrs[index].code);
|
|
}
|
|
|
|
if (Get_Token(t1->ff, str, len)==NULL) { /* Get length. */
|
|
SetError(status=BADINPUTFILE);
|
|
break;
|
|
}
|
|
t1->subrs[index].len = (USHORT)(atoi(str) - t1->leniv);
|
|
if ((t1->subrs[index].code
|
|
= Malloc(t1->subrs[index].len))==NULL) {
|
|
SetError(status = NOMEM);
|
|
break;
|
|
}
|
|
|
|
if (Get_Token(t1->ff, str, len)==NULL) { /* Get RD + space */
|
|
SetError(status=BADINPUTFILE);
|
|
break;
|
|
}
|
|
/* Skip space. */
|
|
(void)GetByte(t1->ff);
|
|
|
|
/* Skip lenIV */
|
|
r = 4330;
|
|
for (j=0; j<t1->leniv; j++) {
|
|
b=GetByte(t1->ff);
|
|
(void)Decrypt(&r, (UBYTE)b);
|
|
}
|
|
if (status!=SUCCESS)
|
|
break;
|
|
|
|
/* Get code. */
|
|
for (j=0; j<t1->subrs[index].len; j++) {
|
|
b=GetByte(t1->ff);
|
|
t1->subrs[index].code[j] = Decrypt(&r, (UBYTE)b);
|
|
}
|
|
if (status!=SUCCESS)
|
|
break;
|
|
|
|
if (Get_Token(t1->ff, str, len)==NULL) { /* Get ND */
|
|
SetError(status=BADINPUTFILE);
|
|
break;
|
|
}
|
|
/* Check for non-ATM compatible equivalent to 'ND' */
|
|
if (!strcmp(str, PS_NOACCESS)) {
|
|
(void)Get_Token(t1->ff, str, len);
|
|
}
|
|
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
return status;
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/***** FUNCTIONS */
|
|
|
|
|
|
/***
|
|
** Function: FlushWorkspace
|
|
**
|
|
** Description:
|
|
** Free the resources allocated for the T1 handle.
|
|
***/
|
|
void FlushWorkspace(struct T1Handle *t1)
|
|
{
|
|
USHORT i;
|
|
|
|
/* Free /Subrs */
|
|
if (t1->subrs) {
|
|
for (i=0; i<t1->numsubrs; i++) {
|
|
Free(t1->subrs[i].code);
|
|
}
|
|
Free(t1->subrs);
|
|
}
|
|
t1->subrs = NULL;
|
|
}
|
|
|
|
|
|
/***
|
|
** Function: CleanUpT1
|
|
**
|
|
** Description:
|
|
** Free the resources allocated for the T1 handle.
|
|
***/
|
|
errcode CleanUpT1(struct T1Handle *t1)
|
|
{
|
|
errcode status = SUCCESS;
|
|
AlignmentControl *align;
|
|
Composite *next;
|
|
Blues *blues;
|
|
USHORT i;
|
|
|
|
if (t1) {
|
|
|
|
/* Free the PSState */
|
|
if (t1->ps)
|
|
FreePSState(t1->ps);
|
|
|
|
/* Free /Subrs */
|
|
if (t1->subrs) {
|
|
for (i=0; i<t1->numsubrs; i++) {
|
|
Free(t1->subrs[i].code);
|
|
}
|
|
Free(t1->subrs);
|
|
}
|
|
|
|
/* Clean up font file reader. */
|
|
status = FRCleanUp(t1->ff);
|
|
|
|
/* Clean up font matrix. */
|
|
if (t1->t1m.fmatrix)
|
|
Free(t1->t1m.fmatrix);
|
|
|
|
/* Clean up seac. */
|
|
while (t1->t1m.used_seac) {
|
|
next = t1->t1m.used_seac->next;
|
|
FreeT1Composite(t1->t1m.used_seac);
|
|
t1->t1m.used_seac = next;
|
|
}
|
|
while (t1->t1m.seac) {
|
|
next = t1->t1m.seac->next;
|
|
FreeT1Composite(t1->t1m.seac);
|
|
t1->t1m.seac = next;
|
|
}
|
|
|
|
/* Clean up stdenc. */
|
|
for (i=0; i<256; i++) {
|
|
if (t1->stdenc[i].code) {
|
|
Free(t1->stdenc[i].code);
|
|
t1->stdenc[i].code = NULL;
|
|
t1->stdenc[i].len = 0;
|
|
}
|
|
}
|
|
|
|
/* Clean up encoding table. */
|
|
if (t1->t1m.encoding)
|
|
FreeEncoding(t1->t1m.encoding, t1->t1m.encSize);
|
|
|
|
/* Free strings */
|
|
if (t1->t1m.date)
|
|
Free(t1->t1m.date);
|
|
if (t1->t1m.copyright)
|
|
Free(t1->t1m.copyright);
|
|
if (t1->t1m.name)
|
|
Free(t1->t1m.name);
|
|
if (t1->t1m.id)
|
|
Free(t1->t1m.id);
|
|
if (t1->t1m.notice)
|
|
Free(t1->t1m.notice);
|
|
if (t1->t1m.fullname)
|
|
Free(t1->t1m.fullname);
|
|
if (t1->t1m.weight)
|
|
Free(t1->t1m.weight);
|
|
if (t1->t1m.family)
|
|
Free(t1->t1m.family);
|
|
if (t1->t1m.widths)
|
|
Free(t1->t1m.widths);
|
|
if (t1->t1m.kerns)
|
|
Free(t1->t1m.kerns);
|
|
if (t1->t1m.stems.vwidths)
|
|
Free(t1->t1m.stems.vwidths);
|
|
if (t1->t1m.stems.hwidths)
|
|
Free(t1->t1m.stems.hwidths);
|
|
blues = &(t1->t1m.blues);
|
|
align = &(t1->t1m.blues.align);
|
|
for (i=0; i<blues->blue_cnt/2; i++) {
|
|
Free(align->top[i].pos);
|
|
}
|
|
for (i=0; i<blues->oblue_cnt/2; i++) {
|
|
Free(align->bottom[i].pos);
|
|
}
|
|
|
|
/* Free handle. */
|
|
Free(t1);
|
|
}
|
|
|
|
return status;
|
|
}
|
|
|
|
|
|
|
|
/***
|
|
** Function: InitT1Input
|
|
**
|
|
** Description:
|
|
** Allocate and initiate a handle for a T1 font file, including
|
|
** extracting data from the font prolog that is needed to
|
|
** read the glyphs, such as /FontMatrix, /Subrs and /lenIV.
|
|
***/
|
|
errcode InitT1Input(const struct T1Arg *arg,
|
|
struct T1Handle **t1ref,
|
|
struct T1Metrics **t1mref,
|
|
const short (*check)(const char *,
|
|
const char *,
|
|
const char *))
|
|
{
|
|
errcode status = SUCCESS;
|
|
struct T1Handle *t1;
|
|
struct PSState *ps;
|
|
Blues *blues;
|
|
boolean hybrid = FALSE;
|
|
struct T1Metrics *t1m = NULL;
|
|
char str[BUFLEN];
|
|
USHORT i;
|
|
|
|
/* Allocate the handle. */
|
|
if (((*t1ref)=Malloc((USHORT)sizeof(struct T1Handle)))==NULL ||
|
|
(ps = AllocPSState())==NULL) {
|
|
if ((*t1ref))
|
|
Free((*t1ref));
|
|
SetError(status = NOMEM);
|
|
} else {
|
|
|
|
/* Initiate the T1 record. */
|
|
t1 = (*t1ref);
|
|
t1m = &t1->t1m;
|
|
(*t1mref) = t1m;
|
|
blues = GetBlues(t1m);
|
|
memset(t1, '\0', sizeof(*t1));
|
|
t1->ps = ps;
|
|
t1->leniv = 4;
|
|
t1m->upem = arg->upem;
|
|
t1m->defstdhw = 70;
|
|
t1m->defstdvw = 80;
|
|
|
|
blues->blueScale = 39; /* Should really be 39.625 */
|
|
blues->blueFuzz = 1;
|
|
blues->blueShift = 7 * F8D8;
|
|
blues->align.cvt = 3;
|
|
t1m->stems.storage = 15;
|
|
|
|
/* Initiate font file reader. */
|
|
if ((status=FRInit(arg->name, pfb_file, &t1->ff))==SUCCESS) {
|
|
|
|
/* Read /FontMatrix and /Subrs. */
|
|
while (status==SUCCESS) {
|
|
if (Get_Token(t1->ff, str, BUFLEN)==NULL) {
|
|
SetError(status=BADINPUTFILE);
|
|
|
|
/**** /ForceBold true def ****/
|
|
} else if (!strcmp(str, PS_FORCEBOLD)) {
|
|
if (Get_Token(t1->ff, str, BUFLEN)) {
|
|
if (!strcmp(str, "true") || !strcmp(str, "True"))
|
|
t1m->forcebold = TRUE;
|
|
else
|
|
t1m->forcebold = FALSE;
|
|
status = SUCCESS;
|
|
} else {
|
|
status = BADINPUTFILE;
|
|
}
|
|
|
|
/**** /BlueFuzz 1 def ****/
|
|
} else if (!strcmp(str, PS_BLUEFUZZ)) {
|
|
if (Get_Token(t1->ff, str, BUFLEN)) {
|
|
blues->blueFuzz = (UBYTE)atoi(str);
|
|
status = SUCCESS;
|
|
} else {
|
|
status = BADINPUTFILE;
|
|
}
|
|
|
|
/**** /BlueScale 0.043625 def ****/
|
|
} else if (!strcmp(str, PS_BLUESCALE)) {
|
|
if (Get_Token(t1->ff, str, BUFLEN)) {
|
|
str[5] = '\0';
|
|
blues->blueScale = (UBYTE)atoi(&str[2]);
|
|
status = SUCCESS;
|
|
} else {
|
|
status = BADINPUTFILE;
|
|
}
|
|
|
|
/**** /BlueShift 7 def ****/
|
|
} else if (!strcmp(str, PS_BLUESHIFT)) {
|
|
if (Get_Token(t1->ff, str, BUFLEN)) {
|
|
blues->blueShift = (short)StrToFix(str, NULL, (long)F8D8);
|
|
status = SUCCESS;
|
|
} else {
|
|
status = BADINPUTFILE;
|
|
}
|
|
|
|
/**** /Encoding StandardEncodind def ****/
|
|
} else if (!strcmp(str, PS_ENCODING)) {
|
|
status = ReadEncodingArray(t1, str, BUFLEN);
|
|
|
|
/**** /StdVW [118] def ****/
|
|
} else if (!strcmp(str, PS_STDVW)) {
|
|
USHORT dummy;
|
|
status = ReadArray(t1, str, BUFLEN,
|
|
&t1m->stdvw, ONE, &dummy);
|
|
|
|
|
|
/**** /StdHW [118] def ****/
|
|
} else if (!strcmp(str, PS_STDHW)) {
|
|
USHORT dummy;
|
|
status = ReadArray(t1, str, BUFLEN,
|
|
&t1m->stdhw, ONE, &dummy);
|
|
|
|
/**** /StemSnapV [118 120] def ****/
|
|
} else if (!strcmp(str, PS_SNAPV)) {
|
|
status = ReadArray(t1, str, BUFLEN,
|
|
&t1m->stemsnapv[0],
|
|
MAXSNAP, &t1m->snapv_cnt);
|
|
|
|
/* Add space for the snap enties in the CV table. */
|
|
if (status==SUCCESS)
|
|
blues->align.cvt = (USHORT)(blues->align.cvt +
|
|
t1m->snapv_cnt);
|
|
|
|
/**** /StemSnapH [118 120] def ****/
|
|
} else if (!strcmp(str, PS_SNAPH)) {
|
|
status = ReadArray(t1, str, BUFLEN,
|
|
&t1m->stemsnaph[0],
|
|
MAXSNAP, &t1m->snaph_cnt);
|
|
|
|
/* Add space for the snap enties in the CV table. */
|
|
if (status==SUCCESS)
|
|
blues->align.cvt = (USHORT)(blues->align.cvt +
|
|
t1m->snaph_cnt);
|
|
|
|
/**** /BlueValues [-15 0] def ****/
|
|
} else if (!strcmp(str, PS_BLUEVALUES)) {
|
|
status = ReadArray(t1, str, BUFLEN,
|
|
&(blues->bluevalues[0]),
|
|
MAXBLUE, &(blues->blue_cnt));
|
|
if (blues->blue_cnt%2)
|
|
SetError(status = BADINPUTFILE);
|
|
|
|
/**** /OtherBlues [-15 0] def ****/
|
|
} else if (!strcmp(str, PS_OTHERBLUES)) {
|
|
status = ReadArray(t1, str, BUFLEN,
|
|
&(blues->otherblues[0]),
|
|
MAXBLUE, &(blues->oblue_cnt));
|
|
if (blues->oblue_cnt%2)
|
|
SetError(status = BADINPUTFILE);
|
|
|
|
/**** /FamilyBlues [-15 0] def ****/
|
|
} else if (!strcmp(str, PS_FAMILYBLUES)) {
|
|
status = ReadArray(t1, str, BUFLEN,
|
|
&(blues->familyblues[0]),
|
|
MAXBLUE, &(blues->fblue_cnt));
|
|
|
|
/**** /FamilyOtherBlues [-15 0] def ****/
|
|
} else if (!strcmp(str, PS_FAMILYOTHERBLUES)) {
|
|
status = ReadArray(t1, str, BUFLEN,
|
|
&(blues->familyotherblues[0]),
|
|
MAXBLUE, &(blues->foblue_cnt));
|
|
|
|
/**** /CharString ... */
|
|
} else if (!strcmp(str, PS_CHARSTRINGS)) {
|
|
break;
|
|
|
|
/**** /FontMatrix [0 0.001 0 0.001 0] def ****/
|
|
} else if (GetFontMatrix(t1m)==NULL &&
|
|
!strcmp(str, PS_FONTMATRIX)) {
|
|
status = ReadFontMatrix(t1, str, BUFLEN);
|
|
} else if (!strcmp(str, PS_SUBRS)) {
|
|
/* Discard prior lores /Subrs. */
|
|
FlushWorkspace(t1);
|
|
|
|
/* Read new subrs. */
|
|
status = ReadFontSubrs(t1,str, BUFLEN);
|
|
|
|
/**** /lenIV 4 def ****/
|
|
} else if (!strcmp(str, PS_LENIV)) {
|
|
if (Get_Token(t1->ff, str, BUFLEN)) {
|
|
t1->leniv = (USHORT)atoi(str);
|
|
status = SUCCESS;
|
|
} else {
|
|
status = BADINPUTFILE;
|
|
}
|
|
} else if (t1m->date==NULL && !strcmp(str, PS_DATE)) {
|
|
if ((GetNewLine(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else if ((t1m->date=Strdup(str))==NULL) {
|
|
SetError(status = NOMEM);
|
|
}
|
|
} else if (t1m->copyright==NULL &&
|
|
!strcmp(str, PS_COPYRIGHT)) {
|
|
if ((GetSeq(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else if ((t1m->copyright=Strdup(str))==NULL) {
|
|
SetError(status = NOMEM);
|
|
}
|
|
} else if (t1m->name==NULL && !strcmp(str, PS_NAME)) {
|
|
if ((Get_Token(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else if ((t1m->name=Strdup(&str[1]))==NULL) {
|
|
SetError(status = NOMEM);
|
|
}
|
|
} else if (t1m->id==NULL && !strcmp(str, PS_ID)) {
|
|
if ((Get_Token(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else if ((t1m->id=Strdup(str))==NULL) {
|
|
SetError(status = NOMEM);
|
|
}
|
|
} else if (t1m->version.ver==0 && !strcmp(str, PS_VERSION)) {
|
|
if ((GetSeq(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else {
|
|
t1m->version.ver = (USHORT)atoi(str);
|
|
if (strchr(str, '.'))
|
|
t1m->version.rev = (USHORT)atoi(strchr(str, '.')+1);
|
|
else
|
|
t1m->version.rev = 0;
|
|
}
|
|
} else if (t1m->notice==NULL && !strcmp(str, PS_NOTICE)) {
|
|
if ((GetSeq(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else if ((t1m->notice=Strdup(str))==NULL) {
|
|
SetError(status = NOMEM);
|
|
}
|
|
} else if (t1m->fullname==NULL && !strcmp(str, PS_FULLNAME)) {
|
|
if ((GetSeq(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else if ((t1m->fullname=Strdup(str))==NULL) {
|
|
SetError(status = NOMEM);
|
|
}
|
|
} else if (t1m->family==NULL && !strcmp(str, PS_FAMILY)) {
|
|
if ((GetSeq(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else if ((t1m->family=Strdup(str))==NULL) {
|
|
SetError(status = NOMEM);
|
|
}
|
|
} else if (t1m->weight==NULL && !strcmp(str, PS_WEIGHT)) {
|
|
if ((GetSeq(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else if ((t1m->weight=Strdup(str))==NULL) {
|
|
SetError(status = NOMEM);
|
|
}
|
|
} else if (t1m->angle==0 && !strcmp(str, PS_ANGLE)) {
|
|
if ((Get_Token(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else
|
|
t1m->angle = StrToFix(str, NULL, F16D16BASE);
|
|
} else if (t1m->underline==0 && !strcmp(str, PS_UNDERLINE)) {
|
|
if ((Get_Token(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else
|
|
t1m->underline = (funit)StrToFix(str, NULL, 1L);
|
|
} else if (t1m->uthick==0 && !strcmp(str, PS_UTHICK)) {
|
|
if ((Get_Token(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else
|
|
t1m->uthick = (funit)StrToFix(str, NULL, 1L);
|
|
} else if (!strcmp(str, PS_ISFIXED)) {
|
|
if ((Get_Token(t1->ff, str, BUFLEN))==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else
|
|
if (!strcmp(str, "true") ||
|
|
!strcmp(str, "True") ||
|
|
!strcmp(str, "TRUE"))
|
|
t1m->fixedPitch = TRUE;
|
|
} else if (!strcmp(str, PS_HYBRID)) {
|
|
hybrid = TRUE;
|
|
}
|
|
}
|
|
|
|
/* Change the baseline zone into an OtherBlues[] zone. */
|
|
if (blues->blue_cnt) {
|
|
blues->otherblues[blues->oblue_cnt++] = blues->bluevalues[0];
|
|
blues->otherblues[blues->oblue_cnt++] = blues->bluevalues[1];
|
|
for (i=2; i<blues->blue_cnt; i++)
|
|
blues->bluevalues[i-2] = blues->bluevalues[i];
|
|
blues->blue_cnt -= 2;
|
|
}
|
|
if (blues->fblue_cnt) {
|
|
blues->familyotherblues[blues->foblue_cnt++]
|
|
= blues->familyblues[0];
|
|
blues->familyotherblues[blues->foblue_cnt++]
|
|
= blues->familyblues[1];
|
|
for (i=2; i<blues->fblue_cnt; i++)
|
|
blues->familyblues[i-2] = blues->familyblues[i];
|
|
blues->fblue_cnt -= 2;
|
|
}
|
|
|
|
/* Allocate the space for the blue buckets. */
|
|
for (i=0; i<blues->blue_cnt; i+=2) {
|
|
USHORT size = (USHORT)((ABS(blues->bluevalues[i+1] -
|
|
blues->bluevalues[i]) +
|
|
1 + 2*blues->blueFuzz)*
|
|
(USHORT)sizeof(struct CVTPos));
|
|
if ((blues->align.top[i/2].pos = Malloc(size))==NULL) {
|
|
SetError(status = NOMEM);
|
|
break;
|
|
}
|
|
|
|
/* Make sure that first value is larger than second value. */
|
|
if (blues->bluevalues[i] > blues->bluevalues[i+1]) {
|
|
LogError(MSG_WARNING, MSG_INVBLUES, NULL);
|
|
SWAPINT(blues->bluevalues[i], blues->bluevalues[i+1]);
|
|
}
|
|
}
|
|
for (i=0; i<blues->oblue_cnt; i+=2) {
|
|
USHORT size = (USHORT)((ABS(blues->otherblues[i+1] -
|
|
blues->otherblues[i]) +
|
|
1 + 2*blues->blueFuzz)*
|
|
(USHORT)sizeof(struct CVTPos));
|
|
if ((blues->align.bottom[i/2].pos = Malloc(size))==NULL) {
|
|
SetError(status = NOMEM);
|
|
break;
|
|
}
|
|
|
|
/* Make sure that first value is larger than second value. */
|
|
if (blues->otherblues[i] > blues->otherblues[i+1]) {
|
|
LogError(MSG_WARNING, MSG_INVBLUES, NULL);
|
|
SWAPINT(blues->otherblues[i], blues->otherblues[i+1]);
|
|
}
|
|
}
|
|
|
|
|
|
/* Advance to the first glyph. */
|
|
if (status==SUCCESS) {
|
|
while (Get_Token(t1->ff, str, BUFLEN) &&
|
|
strcmp(str, PS_BEGIN));
|
|
|
|
if (strcmp(str, PS_BEGIN)) {
|
|
SetError(status = BADT1HEADER);
|
|
}
|
|
|
|
/* Skip lores chars if hybrid font. */
|
|
if (status==SUCCESS && hybrid) {
|
|
USHORT count;
|
|
|
|
/* Skip Charstring dictionary. */
|
|
do {
|
|
/* Glyph name, or end. */
|
|
if (Get_Token(t1->ff, str, BUFLEN)==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
break;
|
|
}
|
|
if (!strcmp(str, PS_END))
|
|
break;
|
|
|
|
/* Charstring length. */
|
|
if (Get_Token(t1->ff, str, BUFLEN)==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
break;
|
|
}
|
|
count = (USHORT)(atoi(str)+1);
|
|
|
|
/* Delimiter. */
|
|
if (Get_Token(t1->ff, str, BUFLEN)==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
break;
|
|
}
|
|
|
|
/* Charstring */
|
|
for (i=0; i<count; i++)
|
|
(void)GetByte(t1->ff);
|
|
|
|
/* Delimiter */
|
|
if (Get_Token(t1->ff, str, BUFLEN)==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
break;
|
|
}
|
|
} while (status==SUCCESS);
|
|
|
|
/* Skip to the beginning of next charstring. */
|
|
while (Get_Token(t1->ff, str, BUFLEN) &&
|
|
strcmp(str, PS_BEGIN));
|
|
|
|
if (strcmp(str, PS_BEGIN)) {
|
|
SetError(status = BADT1HYBRID);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if ((status==SUCCESS) && t1m && check(t1m->name,
|
|
t1m->copyright,
|
|
t1m->notice)!=SUCCESS)
|
|
status = NOCOPYRIGHT;
|
|
|
|
return status;
|
|
}
|
|
|
|
|
|
|
|
/***
|
|
** Function: GetT1Glyph
|
|
**
|
|
** Description:
|
|
** The current file position of the T1 font file must be
|
|
** at the begining of an entry in the /CharStrings dictionary.
|
|
** The function will decode the font commands, parse them, and
|
|
** finally build a representation of the glyph.
|
|
***/
|
|
errcode GetT1Glyph(struct T1Handle *t1,
|
|
struct T1Glyph *glyph,
|
|
const struct GlyphFilter *filter)
|
|
{
|
|
errcode status = SUCCESS;
|
|
/* struct encoding *enc; */
|
|
char str[BUFLEN];
|
|
UBYTE *code;
|
|
USHORT len;
|
|
USHORT i;
|
|
USHORT r = 4330;
|
|
short b;
|
|
|
|
/* Get glyph name or end. */
|
|
if (Get_Token(t1->ff, str, BUFLEN)==NULL) {
|
|
SetError(status = BADINPUTFILE);
|
|
} else if (!strcmp(str, PS_END)) {
|
|
status = DONE;
|
|
} else if (str[0]!='/') {
|
|
SetError(status = BADCHARSTRING);
|
|
} else {
|
|
if ((glyph->name = Strdup(&str[1]))==NULL) {
|
|
SetError(status = NOMEM);
|
|
} else {
|
|
|
|
/* Get length of charstring. */
|
|
(void)Get_Token(t1->ff, str, BUFLEN);
|
|
len = (USHORT)atoi(str);
|
|
|
|
/* Get RD + space */
|
|
(void)Get_Token(t1->ff, str, BUFLEN);
|
|
(void)GetByte(t1->ff);
|
|
|
|
/* Get commands. */
|
|
if (len<BUFLEN)
|
|
code = (UBYTE *)str;
|
|
else
|
|
if ((code = Malloc(len*sizeof(UBYTE)))==NULL) {
|
|
SetError(status = NOMEM);
|
|
}
|
|
|
|
if (code) {
|
|
for (i=0; i<len; i++) {
|
|
b = GetByte(t1->ff);
|
|
code[i] = (UBYTE)Decrypt(&r, (UBYTE)b);
|
|
}
|
|
|
|
/* Parse commands. */
|
|
if (status==SUCCESS) {
|
|
if (t1->t1m.encoding!=NULL ||
|
|
UseGlyph(filter, t1->t1m.seac, glyph->name)) {
|
|
InitPS(t1->ps);
|
|
status = ParseCharString(glyph,
|
|
&t1->t1m.seac,
|
|
t1->ps,
|
|
t1->subrs,
|
|
&code[t1->leniv],
|
|
(USHORT)(len-t1->leniv));
|
|
|
|
/* Skip normal conversion for the ".notdef" glyph. */
|
|
if (!strcmp(glyph->name, ".notdef"))
|
|
status = SKIP;
|
|
|
|
} else {
|
|
status = SKIP;
|
|
|
|
/***
|
|
|
|
Two approaches are implemented for the management of
|
|
composite glyphs:
|
|
|
|
1) It is up to the client to specify a GlyphFilter such
|
|
that all 'seac' characters has their dependent base and
|
|
accent character in the filter as well.
|
|
|
|
2) The converter manages a list of the dependent characters,
|
|
which are converted when found.
|
|
|
|
Approach 2) will typically cause the converter to use more
|
|
memory than what is available in the small memory model,
|
|
which is why the default is to disabled it.
|
|
|
|
***/
|
|
|
|
|
|
#if 0
|
|
|
|
/* Record StandardEncoding glyphs, for 'seac' */
|
|
if ((enc = LookupPSName(t1->t1m.encoding,
|
|
t1->t1m.encSize,
|
|
glyph->name)) &&
|
|
(i = LookupCharCode(enc, ENC_STANDARD))!=0) {
|
|
if ((t1->stdenc[i].code
|
|
= Malloc(len-t1->leniv))==NULL) {
|
|
SetError(status = NOMEM);
|
|
} else {
|
|
memcpy(t1->stdenc[i].code,
|
|
&code[t1->leniv],
|
|
sizeof(UBYTE) * (len - t1->leniv));
|
|
t1->stdenc[i].len = len - t1->leniv;
|
|
}
|
|
}
|
|
#endif
|
|
}
|
|
|
|
if (code!=(UBYTE *)str)
|
|
Free(code);
|
|
|
|
/* Get ND */
|
|
(void)Get_Token(t1->ff, str, BUFLEN);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
return status;
|
|
}
|
|
|
|
|
|
|
|
/***
|
|
** Function: FreeT1Glyph
|
|
**
|
|
** Description:
|
|
** This function frees the memory used to represent
|
|
** a glyph that has been translated.
|
|
***/
|
|
void FreeT1Glyph(T1Glyph *glyph)
|
|
{
|
|
Flex *flex;
|
|
Stem *stem;
|
|
Stem3 *stem3;
|
|
|
|
|
|
if (glyph->name)
|
|
Free(glyph->name);
|
|
while (glyph->hints.vstems) {
|
|
stem = glyph->hints.vstems->next;
|
|
Free(glyph->hints.vstems);
|
|
glyph->hints.vstems = stem;
|
|
}
|
|
while (glyph->hints.hstems) {
|
|
stem = glyph->hints.hstems->next;
|
|
Free(glyph->hints.hstems);
|
|
glyph->hints.hstems = stem;
|
|
}
|
|
while (glyph->hints.vstems3) {
|
|
stem3 = glyph->hints.vstems3->next;
|
|
Free(glyph->hints.vstems3);
|
|
glyph->hints.vstems3 = stem3;
|
|
}
|
|
while (glyph->hints.hstems3) {
|
|
stem3 = glyph->hints.hstems3->next;
|
|
Free(glyph->hints.hstems3);
|
|
glyph->hints.hstems3 = stem3;
|
|
}
|
|
while (glyph->hints.flex) {
|
|
flex = glyph->hints.flex->next;
|
|
Free(glyph->hints.flex);
|
|
glyph->hints.flex = flex;
|
|
}
|
|
while (glyph->paths) {
|
|
Outline *path = glyph->paths;
|
|
glyph->paths = path->next;
|
|
if (path->count) {
|
|
Free(path->onoff);
|
|
Free(path->pts);
|
|
}
|
|
Free(path);
|
|
}
|
|
memset((void *)glyph, '\0', sizeof(T1Glyph));
|
|
}
|
|
|
|
|
|
|
|
/***
|
|
** Function: GetT1Composite
|
|
**
|
|
** Description:
|
|
** This function unlinks the first composite glyph
|
|
** from the list of recorded composite glyphs, which
|
|
** is returned to the caller.
|
|
***/
|
|
struct Composite *GetT1Composite(struct T1Handle *t1)
|
|
{
|
|
struct Composite *comp;
|
|
|
|
comp = t1->t1m.seac;
|
|
if (comp) {
|
|
t1->t1m.seac = comp->next;
|
|
comp->next = t1->t1m.used_seac;
|
|
t1->t1m.used_seac = comp;
|
|
}
|
|
|
|
return comp;
|
|
}
|
|
|
|
|
|
|
|
/***
|
|
** Function: GetT1BaseGlyph
|
|
**
|
|
** Description:
|
|
** This function parses the charstring code associated to the
|
|
** base character of a composite character, if that glyph
|
|
** is not already converted.
|
|
***/
|
|
errcode GetT1BaseGlyph(struct T1Handle *t1,
|
|
const struct Composite *comp,
|
|
struct T1Glyph *glyph)
|
|
{
|
|
struct encoding *enc;
|
|
struct Subrs *subr;
|
|
errcode status = SUCCESS;
|
|
|
|
if ((enc = LookupPSName(t1->t1m.encoding,
|
|
t1->t1m.encSize,
|
|
comp->bchar))==NULL) {
|
|
LogError(MSG_WARNING, MSG_BADENC, comp->bchar);
|
|
return SKIP;
|
|
}
|
|
|
|
subr = &t1->stdenc[LookupCharCode(enc, ENC_STANDARD)];
|
|
|
|
if (subr->len==0) {
|
|
status = SKIP; /* Missing or already done. */
|
|
} else {
|
|
InitPS(t1->ps);
|
|
if ((glyph->name = Strdup((char*)comp->achar))==NULL) {
|
|
SetError(status = NOMEM);
|
|
} else {
|
|
status = ParseCharString(glyph,
|
|
&t1->t1m.seac,
|
|
t1->ps,
|
|
t1->subrs,
|
|
subr->code,
|
|
subr->len);
|
|
}
|
|
Free(subr->code);
|
|
subr->code = NULL;
|
|
subr->len = 0;
|
|
}
|
|
return status;
|
|
}
|
|
|
|
|
|
|
|
/***
|
|
** Function: GetT1AccentGlyph
|
|
**
|
|
** Description:
|
|
** This function parses the charstring code associated to the
|
|
** accent character of a composite character, if that glyph
|
|
** is not already converted.
|
|
***/
|
|
errcode GetT1AccentGlyph(struct T1Handle *t1,
|
|
const struct Composite *comp,
|
|
struct T1Glyph *glyph)
|
|
{
|
|
struct encoding *enc;
|
|
struct Subrs *subr;
|
|
errcode status = SUCCESS;
|
|
|
|
if ((enc = LookupPSName(t1->t1m.encoding,
|
|
t1->t1m.encSize,
|
|
comp->achar))==NULL) {
|
|
LogError(MSG_WARNING, MSG_BADENC, comp->achar);
|
|
return SKIP;
|
|
}
|
|
|
|
subr = &t1->stdenc[LookupCharCode(enc, ENC_STANDARD)];
|
|
|
|
if (subr->len==0) {
|
|
status = SKIP; /* Missing or already done. */
|
|
} else {
|
|
InitPS(t1->ps);
|
|
if ((glyph->name = Strdup((char *)comp->achar))==NULL) {
|
|
SetError(status = NOMEM);
|
|
} else {
|
|
status = ParseCharString(glyph,
|
|
&t1->t1m.seac,
|
|
t1->ps,
|
|
t1->subrs,
|
|
subr->code,
|
|
subr->len);
|
|
}
|
|
Free(subr->code);
|
|
subr->code = NULL;
|
|
subr->len = 0;
|
|
}
|
|
return status;
|
|
}
|
|
|
|
|
|
|
|
/***
|
|
** Function: ReadOtherMetrics
|
|
**
|
|
** Description:
|
|
** Return font level information about the T1 font (mostly
|
|
** metrics).
|
|
***/
|
|
errcode ReadOtherMetrics(struct T1Metrics *t1m,
|
|
const char *metrics)
|
|
{
|
|
errcode status = SUCCESS;
|
|
|
|
if ((status = ReadFontMetrics(metrics, t1m))==NOMETRICS) {
|
|
t1m->flags = DEFAULTMETRICS;
|
|
status = SUCCESS;
|
|
} else {
|
|
t1m->flags = USEMETRICS;
|
|
}
|
|
|
|
return status;
|
|
}
|