Re: [ft-devel] CID keyed fonts in sfnt wrappers in 2.3.8

[Michael Toftdal]

Thank you for your comments, mpsuzuki.


On Tue, Mar 10, 2009 at 3:11 PM,  <address@hidden> wrote:

> Hmm, I guess, what you call as "CID keyed fonts in sfnt
> wrappers" is a CFF OpenType including CID-keyed CFF font
> (CFF OpenType for CJK scripts, usually), which is different
> from sfnt-wrapped CID-keyed fonts. If I'm misunderstanding,
> please correct.

You're right, I mean OpenType fonts with a CID-keyed CFFs.

> On Tue, 10 Mar 2009 10:46:22 +0100
> Michael Toftdal <address@hidden> wrote:
>
>>I am working on a piece of software that is (/was) dependent on the
>>FT_FACE_FLAG_CID_KEYED flag and the 2.3.7 fact, that glyph index ==
>>CID for CID keyed fonts (also those in sfnt wrappers). After upgrading
>>to 2.3.8, I obviously got a small problem.
>>I need the CIDs for outputting PostScript, where I unwrap OpenType
>>(CFF) CID keyed fonts and refer to glyphs via the Identity CMaps.
>>After the upgrade to 2.3.8, I could of course look up the CID flag and
>>the Unicode to CID mapping in the font file itself (eg. as part of the
>>unwrapping where I have the information anyway).
>
> I understand your request as following, if I'm wrong, please
> correct.
>
> * When you load a CID-keyed CFF OpenType by FreeType 2.3.7,
>  you can use it as simple CID-keyed font. The "cmap" table
>  in CFF OpenType is ignored, and you can specify a glyph
>  for CID by putting 16-bit CID to FT_Long-typed codepoint in
>  FreeType API.
>
> * When you load a CID-keyed CFF OpenType by FreeType2 2.3.8,
>  you cannot ignore "cmap" table. To specify a glyph for CID,
>  you have to translate a CID to Unicode codepoint, and put
>  it to FT_Long-typed codepoint in FreeType API. Thus, you
>  have to distinguish CID-keyed CFF OpenType and simple CID-
>  keyed font.

Yes.


> * So, now you want a simple method to handle CID-keyed CFF
>  OpenType as simple CID-keyed font.

That would be one solution to my problem, yes. Another would be the
one that I have ended up implementing by adding two methods to the cid
service interface (pardon my FT vocabular -- I haven't spent more that
one or two days in the source by now). One method to tell me whether a
given font really is a CID keyed font, in the sense that we've agreed
upon in the above. And one method to map unicodes to cids for a given
font. The mapping information is already present in the charset.sids
array on the internal cff font class, as far as I can tell. So I just
had to make sure that the sids array wasn't free'd too early and make
export the mapping via my the cid service.

> I suppose the change was introduced on 2008-07-16.
>
> 2008-07-16  Werner Lemberg  <address@hidden>
>
>        Handle CID-keyed fonts wrapped in an SFNT (with cmaps) correctly.
>
>        * src/cff/cffload.c (cff_font_load): Pass `pure_cff'.
>        Invert sids table only if `pure_cff' is set.
>        * src/cff/cffload.h: Udpated.
>
>        * src/cff/cffobjs.c (cff_face_init): Updated.
>        Set FT_FACE_FLAG_CID_KEYED only if pure_cff is set.
>
> Could you check if FreeType2 on CVS at 2008-07-17 shows
> "changed" behaviour?

I'm pretty sure that, you're right. Werner?


>>For my own purpose, I solved the problem by implementing two new
>>functions in the CID service in FreeType. One to get the "internal"
>>CID flag of the font (result is equal to asking for the CID flag in
>>2.3.7 -- can by the way be mimicked for (most?) practical purposes by
>>asking for the ROS and checking the return value). The other function
>>gives me access to the glyph index to CID mapping that is stored in
>>the cff->charset.sids array (which I have to suspend the free'ing
>>of...).
>
> Hmm. If I'm asked to implement the compatibility feature to
> ignore "cmap" in CFF OpenType, I will modify cffobjs.c:cff_face_init(),
> and insert a hook to skip calling sfnt->load_cmap().
> I can pass some parameters to cff_face_init() by num_params
> and params[], so I should discuss the introduction of new
> tag name to ignore "cmap" in CFF OpenType, something like
> FT_PARAM_TAG_IGNORE_SFNT_CMAP.

I don't know the FT source well enough to give any comments on that.

Kind regards,
-Michael