Re: Why Doesn’t Cairo Use FT_Reference_ Face?

On 15-03-06 08:28 PM, Lawrence D'Oliveiro wrote:
> There is some really peculiar behaviour in the
> cairo_ft_font_face_create_for_ft_face call
> <http://cairographics.org/manual/cairo-FreeType-Fonts.html#cairo-ft-font-face-create-for-ft-face>.
> Cairo wants to keep a reference to the FT_Face, but it seems unable to
> use the FT_Reference_Face call which is provided specifically for that
> purpose
> <http://www.freetype.org/freetype2/docs/reference/ft2-base_interface.html#FT_Reference_Face>!
> So the caller is forced to go through some weird contortions to avoid
> leaking FT_Face objects, or disposing of them prematurely.

Correct.  That's because FT_Reference_Face didn't exist until years after the
cairo API was designed.


> Cairo should call FT_Reference_Face on FT_Face objects that are passed
> to it, and balance these with FT_Done_Face calls when it no longer
> needs these references.
> 
> This would be backward-compatible: the user_data shenanigans documented
> on the Cairo API page would not become problematic, they would simply no
> longer be necessary.

That's in fact not true in all situations.  For example, if an app caches the
cairo_font_face_t in the FT_Face's generic slot, changing the API as you
suggest with create a circular loop causing leaks.  But I'd be happy to ignore
that.  I just hate having to explain this version dependency in the docs...
In HarfBuzz I added a new API called _referenced().  I would prefer that to
changing the semantics of existing API.  The one-line set_user_data doesn't
bother me that much.

Cheers,

-- 
behdad
http://behdad.org/
-- 
cairo mailing list
[email protected]
http://lists.cairographics.org/mailman/listinfo/cairo