[pdftex] CIDFonts in pdfTeX

[Lars Hellström]

A while ago, I started examining new ways of using OpenType fonts with 
TeX, focusing in particular on catering for fonts that have many 
ligatures or alternate glyphs. I expect I'll end up creating some new 
tool for converting fonts to TeX-usable format, but would prefer to 
make do with existing backends (in particular dvips and pdfTeX) as far 
as possible.

One of my early hypotheses, which so far has held beautifly, is that 
OpenType fonts are best served by being treated as CIDFonts. This is 
particularly clear for the combination of TrueType font and PDF, since 
the support in PDF (the standard) for simple TrueType fonts can at best 
be described as a kludge, whereas the composite font (i.e., CIDFont) 
counterpart provides clean control of e.g. the encoding. CIDFonts don't 
seem to have been much used by us TeXies, though (presumably because 
they were introduced to deal with CJK fonts, and their documentation is 
geared towards suggesting such usage).

Getting dvips to map a TeX font onto a (PS type 0 font mapping to) a 
CIDFont is merely a matter of giving it a file with "the font" in a 
suitable PS wrapper, and as of today I have a couple of procedures that 
will generate such a wrapper for an arbitrary TrueType font.[*] 
Unfortunately, pdfTeX doesn't seem so easily instructed. I know what 
data structure should appear in the finished PDF file, but there 
doesn't seem to be a way of making pdfTeX put it there -- at least not

  (a) according to the manual (though that may be incomplete; e.g.
      I've noticed some mechanisms for interpreting encoding
      vector glyph names that exist in the source but aren't
      mentioned in the manual)

and

  (b) from within the map file. Using \pdffontattr I suppose it is
      possible to construct a pretty arbitrary font dictionary from
      within TeX code, but that doesn't seem to be able to get at
      the base fonts of a virtual font, which would really be all
      that the CIDFonts should be. Moreover I'd prefer it if the
      font set-up could be accomplished without pdfTeX-specific
      TeX commands, as that is hard to integrate with LaTeX's
      NFSS.

So therefore I would like to request (I can file a format request if 
you like, but I wouldn't mind getting some reactions first) a new 
pdfTeX feature, to the end of allowing a mapfile to specify that an 
arbitrary PDF data structure should be used as the font dictionary 
corresponding to a specific TeX font.

Now, the most practical way of encoding a PDF data structure for 
embedding is of course to provide it within another PDF file; I 
understand pdfTeX already has the machinery to copy a PDF object (and 
everything it references) from another PDF file, so this shouldn't be 
difficult to implement.[**] Concerning the interface, my suggestion 
would be the following:

  * If the <fontfile> in a mapline has suffix .pdf, then the wanted
    font dictionary can be found in its Catalog/Pages/Resources/Fonts
    dictionary, i.e., among the fonts in the Resources dictionary of
    the root node in the pages tree of that file. The <basename> of
    the mapline is then the key within the Resources/Fonts dictionary
    of the selected font.

For nice appearance, one could structure such "font container" PDF 
files to be demonstrations of the glyphs in the font(s), maybe even 
with instructions for how to configure pdfTeX to use them...


Lars Hellström


[*] The same principles should apply for CFF fonts, but in that case 
one has to cut a bit deeper when performing some necessary surgery, so 
I'm leaving those for later. The PS wrapper is simpler in that case 
than for TrueType, though.

[**] Not difficult, unless one insists on subsetting the fonts. I'm 
perfectly happy with getting the font embedded in full, since that is 
/way/ better than not being able to use it at all. Subsetting can 
always be added later, if anyone cares enough to implement it.

PS: Why are the links in the manual made with /Fit (or maybe it is 
/FitV) destinations??? Constantly having to rezoom to a readable size 
is incredibly annoying! And it's not very difficult to preserve the 
zoom level: just use /XYZ with null for z (and maybe also x and y).

PPS: Please Cc: me in replies; I'm not subscribed to the pdftex mailing 
list.