Re: [ft-devel] error loading Diwani Outline Shaded using the bytecode interpreter

[Werner LEMBERG]

[David, please read!  Perhaps you can provide fixes?  If you need the
particular font for inspection I can send it to you.]


Lars,

sorry for the late reply.

> seems like freetype doesn't like to load the above font (IIRC it's
> part of the arabic fonts for MS Office) when enabling the Bytecode
> interpreter.  FT_Load_Glyph always returns an
> FT_Err_Too_Few_Arguments.

It is a problem in the font: The `prep' table is too short, lacking a
final argument to pop from the stack.  Your solution of catching
FT_Err_Too_Few_Arguments is a good start, but I consider the current
behaviour of FreeType as buggy; more specifically, the font exhibits
the following two bugs.

   . Assuming that Dwnoutsh.ttf loads just fine in Windows, we should
     do the same.  Similar to other places in FreeType I suggest that
     we provide, say, 10 zero-valued arguments on the stack which
     buggy fonts can pop safely while handling the prep table.  Maybe
     we should do the same with other instruction tables too.

   . Regardless of using FT_LOAD_FORCE_AUTOHINT, the `prep' and `fpgm'
     bytecode tables are always executed, both for initializing the
     font and changing the size.  This is a serious problem IMHO; for
     example, the ftview program can't display Dwnoutsh.ttf at all
     exactly for this reason because tt_size_reset always fails.

     I'm not sure how to solve that elegantly; we should probably add
     an `FT_PARAM_TAG_AUTOHINT' flag so that the bytecode interpreter
     stuff can be ignored on a global level for a complete font
     (contrary to FT_LOAD_FORCE_AUTOHINT which acts locally on a
     single glyph).

David?


     Werner