cups-filters/fontembed/embed_sfnt.c

#include "embed.h"
#include "embed_pdf_int.h"
#include "embed_sfnt_int.h"
#include "sfnt.h"
#include "sfnt_int.h"
#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

EMB_RIGHT_TYPE emb_otf_get_rights(OTF_FILE *otf) // {{{
{
  EMB_RIGHT_TYPE ret=EMB_RIGHT_FULL;

  int len;
  char *os2=otf_get_table(otf,OTF_TAG('O','S','/','2'),&len);
  if (os2) {
    const unsigned short os2_version=get_USHORT(os2);
    // check len
    assert( (os2_version!=0x0000)||(len==78) );
    assert( (os2_version!=0x0001)||(len==86) );
    assert( (os2_version<0x0002)||(os2_version>0x0004)||(len==96) );
    if (os2_version<=0x0004) {
      // get rights
      unsigned short fsType=get_USHORT(os2+8);
      // from Adobe's Fontpolicies_v9.pdf, pg 13:
      if (fsType==0x0002) {
        ret=EMB_RIGHT_NONE;
      } else {
        ret=fsType&0x0300; // EMB_RIGHT_BITMAPONLY, EMB_RIGHT_NO_SUBSET
        if ((fsType&0x000c)==0x0004) {
          ret|=EMB_RIGHT_READONLY;
        }
      }
    }
    free(os2);
  }
  return ret;
}
// }}}

// NOTE: statically allocated buffer
const char *emb_otf_get_fontname(OTF_FILE *otf) // {{{
{
  static char fontname[64];

  int len;
  const char *fname=otf_get_name(otf,3,1,0x409,6,&len); // microsoft
  if (fname) {
    int iA,iB=0;
    for (iA=0;(iA<63)&&(iA*2<len);iA++) {
      if ( (fname[2*iA]==0)&&
           (fname[2*iA+1]>=33)&&(fname[2*iA+1]<=126)&&
           (!strchr("[](){}<>/%",fname[iA*2+1])) ) {
        fontname[iB++]=fname[iA*2+1];
      }
    }
    fontname[iB]=0;
  } else if ((fname=otf_get_name(otf,1,0,0,6,&len))) { // mac
    int iA,iB=0;
    for (iA=0;(iA<63)&&(iA<len);iA++) {
      if ( (fname[iA]>=33)&&(fname[iA]<=126)&&
           (!strchr("[](){}<>/%",fname[iA])) ) {
        fontname[iB++]=fname[iA];
      }
    }
    fontname[iB]=0;
  } else {
    fontname[0]=0;
  }
  if (!*fontname) {
    // TODO construct a fontname, eg from */*/*/4
    fprintf(stderr,"WARNING: no fontName\n");
  }
  return fontname;
}
// }}}

// TODO? monospaced by actual glyph width?
// TODO? use PCLT table? (esp. CFF, as table dircouraged for glyf fonts)
void emb_otf_get_pdf_fontdescr(OTF_FILE *otf,EMB_PDF_FONTDESCR *ret) // {{{
{
  int len;

//  TODO
//  ... fill in struct
  char *head=otf_get_table(otf,OTF_TAG('h','e','a','d'),&len);
  assert(head); // version is 1.0 from otf_load
  ret->bbxmin=get_SHORT(head+36)*1000/otf->unitsPerEm;
  ret->bbymin=get_SHORT(head+38)*1000/otf->unitsPerEm;
  ret->bbxmax=get_SHORT(head+40)*1000/otf->unitsPerEm;
  ret->bbymax=get_SHORT(head+42)*1000/otf->unitsPerEm;
  const int macStyle=get_USHORT(head+44);
  free(head);

  char *post=otf_get_table(otf,OTF_TAG('p','o','s','t'),&len);
  assert(post);
  const unsigned int post_version=get_ULONG(post);
  // check length
  assert( (post_version!=0x00010000)||(len==32) );
  assert( (post_version!=0x00020000)||(len>=34+2*otf->numGlyphs) );
  assert( (post_version!=0x00025000)||(len==35+otf->numGlyphs) );
  assert( (post_version!=0x00030000)||(len==32) );
  assert( (post_version!=0x00020000)||(get_USHORT(post+32)==otf->numGlyphs) ); // v4?
//  assert( (post_version==0x00030000)==(!!(otf->flags&OTF_F_FMT_CFF)) ); // ghostscript embedding does this..
  // TODO: v4 (apple) :  uint16 reencoding[numGlyphs]
  if ( (post_version==0x00010000)||
       (post_version==0x00020000)||
       (post_version==0x00025000)||
       (post_version==0x00030000)||
       (post_version==0x00040000) ) {
    ret->italicAngle=get_LONG(post+4)>>16;
    if (get_ULONG(post+12)>0) { // monospaced
      ret->flags|=1;
    }
  } else {
    fprintf(stderr,"WARNING: no italicAngle, no monospaced flag\n");
  }
  free(post);

  char *os2=otf_get_table(otf,OTF_TAG('O','S','/','2'),&len);
  if (os2) {
    const unsigned short os2_version=get_USHORT(os2);
    // check len
    assert( (os2_version!=0x0000)||(len==78) );
    assert( (os2_version!=0x0001)||(len==86) );
    assert( (os2_version<0x0002)||(os2_version>0x0004)||(len==96) );
    if (os2_version<=0x0004) {

      // from PDF14Deltas.pdf, pg 113
      const int weightClass=get_USHORT(os2+4);
      ret->stemV=50+weightClass*weightClass/(65*65); // TODO, really bad
//printf("a %d\n",weightClass);

      if (ret->supplement>=0) { // cid
        ret->panose=ret->data;
        memcpy(ret->panose,os2+30,12); // sFamilyClass + panose
      }
      const unsigned short fsSelection=get_USHORT(os2+62);
      if (fsSelection&0x01) { // italic
        ret->flags|=0x0040;
      }
      if ( (fsSelection&0x10)&&(weightClass>600) ) { // force bold
        ret->flags|=0x0400;
      }
      const unsigned char family_class=get_USHORT(os2+30)>>8;
      if (family_class==10) { // script
        ret->flags|=0x0008;
      }
      if (family_class!=8) { // not sans-serif
        ret->flags|=0x0002;
      }

      ret->avgWidth=get_SHORT(os2+2)*1000/otf->unitsPerEm;
      ret->ascent=get_SHORT(os2+68)*1000/otf->unitsPerEm;
      ret->descent=get_SHORT(os2+70)*1000/otf->unitsPerEm;
      if (os2_version>=0x0002) {
        ret->xHeight=get_SHORT(os2+86)*1000/otf->unitsPerEm;
        ret->capHeight=get_SHORT(os2+88)*1000/otf->unitsPerEm;
      } // else capHeight fixed later
    } else {
      free(os2);
      os2=NULL;
    }
  } else {
    fprintf(stderr,"WARNING: no OS/2 table\n");
    // e.g. subsetted font from ghostscript // e.g. CFF
  }
  if (os2) {
    free(os2);
  } else { // TODO (if(CFF))
    fprintf(stderr,"WARNING: no ascent/descent, capHeight, stemV, flags\n");
    if (macStyle&0x01) { // force bold - just do it on bold
      ret->flags|=0x0400;
    }
    if (macStyle&0x02) { // italic
      ret->flags|=0x0004;
    }
    //  ... flags TODO? (Serif, Script, Italic, AllCap,SmallCap, ForceBold)
  }

// ? maybe get ascent,descent,capHeight,xHeight,stemV directly from cff
  // Fallbacks
  if ( (!ret->ascent)||(!ret->descent) ) {
    char *hhea=otf_get_table(otf,OTF_TAG('h','h','e','a'),&len);
    if (hhea) {
      ret->ascent=get_SHORT(hhea+4)*1000/otf->unitsPerEm;
      ret->descent=get_SHORT(hhea+6)*1000/otf->unitsPerEm;
    }
    free(hhea);
  }
  if (!ret->stemV) { // TODO? use name
    const unsigned short d_gid=otf_from_unicode(otf,'.');
    if (d_gid) { // stemV=bbox['.'].width;
      len=otf_get_glyph(otf,d_gid);
      assert(len>=10);
      ret->stemV=(get_SHORT(otf->gly+6)-get_SHORT(otf->gly+2))*1000/otf->unitsPerEm;
    } else {
      if (macStyle&1) { // bold
        ret->stemV=165;
      } else {
        ret->stemV=109; // TODO... unserious values...
      }
    }
  }
  if (!ret->capHeight) { // TODO? only reqd. for fonts with latin...
    ret->capHeight=ret->ascent;
    // TODO: OTF spec says:  use metrics of 'H' (0 if not available)
  }
  if (0) { // TODO? uses only adobe latin standard? ?? e.g. Type1
    ret->flags|=0x0020;
  } else {
    ret->flags|=0x0004;
  }
  // TODO SmallCap by font name(?)

// TODO ;   ? cid ?
}
// }}}

// TODO: split generic part and otf part
// TODO: FIXME: gid vs. char   ... NOTE: not called in multi_byte mode...
// Adobe does: char --MacRoman/WinAnsi--> name --AGL--> unicode --cmap(3,1) --> gid   only avoidable by setting 'symbol'+custom(1,0)/(3,0)
// HINT: caller sets len == otf->numGlyphs   (only when not using encoding...)
EMB_PDF_FONTWIDTHS *emb_otf_get_pdf_widths(OTF_FILE *otf,const unsigned short *encoding,int len,const BITSET glyphs) // {{{ glyphs==NULL -> all from 0 to len
{
  assert(otf);

  int first=len,last=0;
  int iA;

  if (glyphs) {
    for (iA=0;iA<len;iA++) { // iA is a "gid" when in multi_byte mode...
      const int gid=(encoding)?encoding[iA]:otf_from_unicode(otf,iA); // TODO
      if (bit_check(glyphs,gid)) {
        if (first>iA) { // first is a character index
          first=iA;
        }
        if (last<iA) {
          last=iA;
        }
      }
    }
  } else {
    first=0;
    last=len;
  }
  if (last<first) {
    // empty
    fprintf(stderr,"WARNING: empty embedding range\n");
    return NULL;
  }

  // ensure hmtx is there
  if (!otf->hmtx) {
    if (otf_load_more(otf)!=0) {
      fprintf(stderr,"Unsupported OTF font / cmap table \n");
      return NULL;
    }
  }

  // now create the array
  EMB_PDF_FONTWIDTHS *ret=emb_pdf_fw_new(last-first+1);
  if (!ret) {
    return NULL;
  }
  ret->first=first;
  ret->last=last;
  ret->widths=ret->data;
  for (iA=0;first<=last;iA++,first++) {
    const int gid=(encoding)?encoding[first]:otf_from_unicode(otf,first); // TODO
    if (gid>=otf->numGlyphs) {
      fprintf(stderr,"Bad glyphid\n");
      assert(0);
      free(ret);
      return NULL;
    }
    if ( (!glyphs)||(bit_check(glyphs,gid)) ) {
      ret->widths[iA]=get_width_fast(otf,gid)*1000/otf->unitsPerEm;
    } // else 0 from calloc
  }

  return ret;
}
// }}}

// otf->hmtx must be there
static int emb_otf_pdf_glyphwidth(void *context,int gid) // {{{
{
  OTF_FILE *otf=(OTF_FILE *)context;
  return get_width_fast(otf,gid)*1000/otf->unitsPerEm;
}
// }}}

EMB_PDF_FONTWIDTHS *emb_otf_get_pdf_cidwidths(OTF_FILE *otf,const BITSET glyphs) // {{{ // glyphs==NULL -> output all
{
  assert(otf);

  // ensure hmtx is there
  if (!otf->hmtx) {
    if (otf_load_more(otf)!=0) {
      fprintf(stderr,"Unsupported OTF font / cmap table \n");
      return NULL;
    }
  }
//  int dw=emb_otf_pdf_glyphwidth(otf,0); // e.g.
  int dw=-1; // let them estimate

  return emb_pdf_fw_cidwidths(glyphs,otf->numGlyphs,dw,emb_otf_pdf_glyphwidth,otf);
}
// }}}

/*** PS stuff ***/

#include "dynstring.h"

const char *aglfn13(unsigned short uni); // aglfn13.c
#include "macroman.h"

// TODO? optimize pascal string skipping? (create index)
// NOTE: might return a statically allocated string
static const char *emb_otf_get_post_name(const char *post,unsigned short gid) // {{{
{
  if (!post) {
    return NULL;
  }
  const unsigned int post_version=get_ULONG(post);
  if (post_version==0x00010000) { // font has only 258 chars... font cannot be used on windows
    if (gid<sizeof(macRoman)/sizeof(macRoman[0])) {
      return macRoman[gid];
    }
  } else if (post_version==0x00020000) {
    const unsigned short num_glyphs=get_USHORT(post+32);
    // assert(num_glyphs==otf->numGlyphs);
    if (gid<num_glyphs) {
      unsigned short idx=get_USHORT(post+34+2*gid);
      if (idx<258) {
        if (idx<sizeof(macRoman)/sizeof(macRoman[0])) {
          return macRoman[idx];
        }
      } else if (idx<32768) {
        const unsigned char *pos=(unsigned char *)post+34+2*num_glyphs;
        for (idx-=258;idx>0;idx--) { // this sucks...
          pos+=*pos+1; // skip this string
        }
        // convert pascal string to asciiz
        static char ret[256];
        const unsigned char len=*pos;
        memcpy(ret,(const char *)pos+1,len);
        ret[len]=0;
        return ret;
      }
    }
  } else if (post_version==0x00025000) { // similiar to 0x00010000, deprecated
    const unsigned short num_glyphs=get_USHORT(post+32);
    if (gid<num_glyphs) {
      const unsigned short idx=post[34+gid]+gid; // post is signed char *
      if (idx<sizeof(macRoman)/sizeof(macRoman[0])) {
        return macRoman[idx];
      }
    }
  } else if (post_version==0x00030000) {
    // no glyph names, sorry
//  } else if (post_version==0x00040000) { // apple AAT ?!
  }
  return NULL;
}
// }}}

// TODO!? to_unicode should be able to represent more than one unicode character?
// NOTE: statically allocated string
static const char *get_glyphname(const char *post,unsigned short *to_unicode,int charcode,unsigned short gid) // {{{ if charcode==0 -> force gid to be used
{
  if (gid==0) {
    return ".notdef";
  }
  const char *postName=emb_otf_get_post_name(post,gid);
  if (postName) {
    return postName;
  }
  static char ret[255];
  if (charcode) {
    if (to_unicode) { // i.e. encoding was there
      charcode=to_unicode[charcode];
      // TODO!? to_unicode should be able to represent more than one unicode character?
      // TODO for additional credit: for ligatures, etc  create /f_f /uni12341234  or the like
    }
    const char *aglname=aglfn13(charcode); // TODO? special case ZapfDingbats?
    if (aglname) {
      return aglname;
    }
    snprintf(ret,250,"uni%04X",charcode); // allows extraction
  } else {
    snprintf(ret,250,"c%d",gid);  // last resort: only by gid
  }
  return ret;
}
// }}}

struct OUTFILTER_PS {
  OUTPUT_FN out;
  void *ctx;
  int len;
};

// TODO: for maximum compatiblity (PS<2013 interpreter)  split only on table or glyph boundary (needs lookup in loca table!)
// Note: table boundaries are at each call!
static void outfilter_ascii_ps(const char *buf,int len,void *context)  // {{{
{
  struct OUTFILTER_PS *of=context;
  OUTPUT_FN out=of->out;
  int iA;

  (*out)("<",1,of->ctx);
  of->len++;

  const char *last=buf;
  char tmp[256];
  while (len>0) {
    for (iA=0;(iA<76)&&(len>0);iA+=2,len--) {
      const unsigned char ch=buf[iA>>1];
      tmp[iA]="0123456789abcdef"[ch>>4];
      tmp[iA+1]="0123456789abcdef"[ch&0x0f];
    }
    buf+=iA>>1;
    if (buf<last+64000) {
      if (len>0) {
        tmp[iA++]='\n';
      }
      (*out)(tmp,iA,of->ctx);
    } else {
      last=buf;
      strcpy(tmp+iA,"00>\n<");
      iA+=5;
      (*out)(tmp,iA,of->ctx);
    }
    of->len+=iA;
  }

  (*out)("00>\n",4,of->ctx);
  of->len+=4;
}
// }}}

static void outfilter_binary_ps(const char *buf,int len,void *context)  // {{{
{
  struct OUTFILTER_PS *of=context;
  OUTPUT_FN out=of->out;

  char tmp[100];
  while (len>0) {
    const int maxlen=(len>64000)?64000:len;
    const int l=sprintf(tmp,"%d RD ",maxlen);
    (*out)(tmp,l,of->ctx);
    of->len+=l;

    (*out)(buf,maxlen,of->ctx);
    (*out)("\n",1,of->ctx);
    of->len+=maxlen+1;
    len-=maxlen;
    buf+=maxlen;
  }
}
// }}}

/*
  encoding:  character-code -> glyph id  ["required", NULL: identity, i.e. from_unicode()] // TODO: respect subsetting
  to_unicode:  character-code -> unicode  [NULL: no char names]  // kind-of "reverse" of encoding (to_unicode does not make sense without >encoding)

Status:
  - we need a 0..255 encoding to be used in the PS file
  - we want to allow the use of encoding[];  this should map from your desired PS-stream output character (0..255) directly to the gid
  - if encoding[] is not used, MacRoman/WinAnsi/latin1 is expected (easiest: latin1, as it is a subset of unicode)
    i.e. your want to output latin1 to the PS-stream
  - len is the length of >encoding, or the "last used latin1 character"
  - oh. in multibyte-mode no >encoding probably should mean identity(gid->gid) not (latin1->gid)
  - non-multibyte PDF -> only 255 chars  ... not recommended (we can't just map to gids, but only to names, which acro will then cmap(3,1) to gids)

  => problem with subsetting BITSET (keyed by gid); we want BITSET keyed by 0..255 (via encoding)

  // TODO: a) multi font encoding
  // TODO: b) cid/big font encoding (PS>=2015) [/CIDFontType 2]     : CMap does Charcode->CID, /CIDMap does CID->GID [e.g. Identity/delta value]
  //          (also needed [or a)] for loca>64000 if split, etc)      e.g. /CIDMap 0  [requires PS>=3011?]
  //          [Danger: do not split composites]
  // TODO? incremental download [/GlyphDirectory array or dict]     : /GlyphDirectory does GID-><glyf entry> mapping
  //       need 'fake' gdir table (size,offset=0) in sfnt; loca, glyf can be ommited; hmtx can be omitted for PS>=3011 [/MetricsCount 2]
  //       idea is to fill initial null entries in the array/dict   [Beware of save/restore!]
  // NOTE: even when subsetting the font has to come first in the PS file


... special information: when multi-byte PDF encoding is used <gid> is output.
    therefore /Encoding /Identity-H + /CIDSystemInfo Adobe-Identity-0 will yield 1-1 mapping for font.
    problem is that text is not selectable. therefore there is the /ToUnicode CMap option
*/
int emb_otf_ps(OTF_FILE *otf,unsigned short *encoding,int len,unsigned short *to_unicode,OUTPUT_FN output,void *context) // {{{
{
  const int binary=0; // binary format? // TODO
  if (len>256) {
    fprintf(stderr,"Encoding too big(%d) for Type42\n",len);
    return -1;
  }
  if (len<1) {
    fprintf(stderr,"At least .notdef required in Type42\n");
    return -1;
  }
  if (!encoding) {
    to_unicode=NULL; // does not make sense
  }
  int iA,ret=0;

  DYN_STRING ds;
  if (dyn_init(&ds,1024)==-1) {
    return -1;
  }

  int rlen=0;
  char *head=otf_get_table(otf,OTF_TAG('h','e','a','d'),&rlen);
  if (!head) {
    free(ds.buf);
    return -1;
  }
  dyn_printf(&ds,"%%!PS-TrueTypeFont-%d-%d\n",
                 otf->version,get_ULONG(head+4));
  const int bbxmin=get_SHORT(head+36)*1000/otf->unitsPerEm,
            bbymin=get_SHORT(head+38)*1000/otf->unitsPerEm,
            bbxmax=get_SHORT(head+40)*1000/otf->unitsPerEm,
            bbymax=get_SHORT(head+42)*1000/otf->unitsPerEm;
  free(head);

  char *post=otf_get_table(otf,OTF_TAG('p','o','s','t'),&rlen);
  if ( (!post)&&(rlen!=-1) ) { // other error than "not found"
    free(ds.buf);
    return -1;
  }
  if (post) {
    const unsigned int minMem=get_ULONG(post+16),maxMem=get_ULONG(post+20);
    if (minMem) {
      dyn_printf(&ds,"%%VMusage: %d %d\n",minMem,maxMem);
    }
  }

  // don't forget the coordinate scaling...
  dyn_printf(&ds,"11 dict begin\n"
                 "/FontName /%s def\n"
                 "/FontType 42 def\n"
                 "/FontMatrix [1 0 0 1 0 0] def\n"
                 "/FontBBox [%f %f %f %f] def\n"
                 "/PaintType 0 def\n",
//                 "/XUID [42 16#%X 16#%X 16#%X 16#%X] def\n"  // TODO?!? (md5 of font data)  (16# means base16)
                 emb_otf_get_fontname(otf),
                 bbxmin/1000.0,bbymin/1000.0,bbxmax/1000.0,bbymax/1000.0);
  if (post) {
    dyn_printf(&ds,"/FontInfo 4 dict dup begin\n"
// TODO? [even non-post]: /version|/Notice|/Copyright|/FullName|/FamilyName|/Weight  () readonly def\n   from name table: 5 7 0 4 1 2
// using: otf_get_name(otf,3,1,0x409,?,&len) / otf_get_name(otf,1,0,0,?,&len)   + encoding
                   "  /ItalicAngle %d def\n"
                   "  /isFixedPitch %s def\n"
                   "  /UnderlinePosition %f def\n"
                   "  /UnderlineThickness %f def\n"
                   "end readonly def\n",
                   get_LONG(post+4)>>16,
                   (get_ULONG(post+12)?"true":"false"),
                   (get_SHORT(post+8)-get_SHORT(post+10)/2)/(float)otf->unitsPerEm,
                   get_SHORT(post+10)/(float)otf->unitsPerEm);
  }
  dyn_printf(&ds,"/Encoding 256 array\n"
                 "0 1 255 { 1 index exch /.notdef put } for\n");
  for (iA=0;iA<len;iA++) { // encoding data: 0...255 -> /glyphname
    const int gid=(encoding)?encoding[iA]:otf_from_unicode(otf,iA);
    if (gid!=0) {
      dyn_printf(&ds,"dup %d /%s put\n",
                     iA,get_glyphname(post,to_unicode,iA,gid));
    }
  }
  dyn_printf(&ds,"readonly def\n");

  if (binary) {
    dyn_printf(&ds,"/RD { string currentfile exch readstring pop } executeonly def\n");
  }
  dyn_printf(&ds,"/sfnts[\n");

  if (ds.len<0) {
    free(post);
    free(ds.buf);
    return -1;
  }
  (*output)(ds.buf,ds.len,context);
  ret+=ds.len;
  ds.len=0;

// TODO: only tables as in otf_subset
// TODO:  somehow communicate table boundaries:
  //   otf_action_copy  does exactly one output call (per table)
  //   only otf_action_replace might do two (padding)
  // {{{ copy tables verbatim (does not affect ds .len)
  struct _OTF_WRITE *otfree=NULL;
#if 0
  struct _OTF_WRITE *otw;
  otwfree=otw=malloc(sizeof(struct _OTF_WRITE)*otf->numTables);
  if (!otw) {
    fprintf(stderr,"Bad alloc: %m\n");
    free(post);
    free(ds.buf);
    return -1;
  }
  // just copy everything
  for (iA=0;iA<otf->numTables;iA++) {
    otw[iA].tag=otf->tables[iA].tag;
    otw[iA].action=otf_action_copy;
    otw[iA].param=otf;
    otw[iA].length=iA;
  }
  int numTables=otf->numTables;
#else
  struct _OTF_WRITE otw[]={ // sorted
      {OTF_TAG('c','m','a','p'),otf_action_copy,otf,},
      {OTF_TAG('c','v','t',' '),otf_action_copy,otf,},
      {OTF_TAG('f','p','g','m'),otf_action_copy,otf,},
      {OTF_TAG('g','l','y','f'),otf_action_copy,otf,},
      {OTF_TAG('h','e','a','d'),otf_action_copy,otf,},
      {OTF_TAG('h','h','e','a'),otf_action_copy,otf,},
      {OTF_TAG('h','m','t','x'),otf_action_copy,otf,},
      {OTF_TAG('l','o','c','a'),otf_action_copy,otf,},
      {OTF_TAG('m','a','x','p'),otf_action_copy,otf,},
      {OTF_TAG('n','a','m','e'),otf_action_copy,otf,},
      {OTF_TAG('p','r','e','p'),otf_action_copy,otf,},
      // vhea vmtx (never used in PDF, but possible in PS>=3011)
      {0,0,0,0}};
  int numTables=otf_intersect_tables(otf,otw);
#endif

  struct OUTFILTER_PS of;
  of.out=output;
  of.ctx=context;
  of.len=0;
  if (binary) {
    iA=otf_write_sfnt(otw,otf->version,numTables,outfilter_binary_ps,&of);
  } else {
    iA=otf_write_sfnt(otw,otf->version,numTables,outfilter_ascii_ps,&of);
  }
  free(otfree);
  if (iA==-1) {
    free(post);
    free(ds.buf);
    return -1;
  }
  ret+=of.len;
  // }}} done copying

  dyn_printf(&ds,"] def\n");

  dyn_printf(&ds,"/CharStrings %d dict dup begin\n"
                 "/.notdef 0 def\n",len);
  for (iA=0;iA<len;iA++) { // charstrings data: /glyphname -> gid
    const int gid=(encoding)?encoding[iA]:otf_from_unicode(otf,iA);
    if (gid) {
      dyn_printf(&ds,"/%s %d def\n",get_glyphname(post,to_unicode,iA,gid),gid);
    }
    // (respecting subsetting...)
  }
  dyn_printf(&ds,"end readonly def\n");
  dyn_printf(&ds,"FontName currentdict end definefont pop\n");
  free(post);

  if (ds.len<0) {
    free(ds.buf);
    return -1;
  }
  (*output)(ds.buf,ds.len,context);
  ret+=ds.len;
  ds.len=0;

  free(ds.buf);
  return ret;
}
// }}}