Re: [Devel] incorrect types for some arguments in the TrueType driver

[Steve Hartwell]

On Apr 17, 2004, at 2:44 PM, Graham Asher wrote:
> This is a detail of TrueType. FreeType uses 32-bit values. For example,
> FT_Get_Char_Index returns FT_UInt, and FT_Load_Glyph takes FT_UInt. For
> consistency, drivers for specific typeface formats should accept this 
> type.
> They can reject any out-of-range values.

You are correct; in particular, since an FT_Slot_LoadFunc is defined as 
a pointer to a function whose glyph_index parameter is an FT_UInt, then 
indeed Load_Glyph() in ttdriver.c should have the same parameter type.  
On a machine which does not promote shorts to longs in function 
parameters, this mismatch will indeed cause problems in Load_Glyph().

But Load_Glyph() is still declared differently from an 
FT_Slot_LoadFunc, since its first parameter is a TT_GlyphSlot instead 
of a FT_GlyphSlot.  They are "compatible" types (the former points to a 
struct which contains the latter), but not in the eyes of the gcc 
compiler.

In fact, the other pointers to functions in tt_driver_class are also 
being cast for the same reason.  Since your friend's problem was due to 
long vs short parameter in Load_Glyph(), I thought it might be useful 
to see if anything else was mismatched, so I changed the declarations 
and definitions in my local copies of ttobjs.h, ttobjs.c, and 
ttdriver.c to match the function pointer types defined in 
FT_Driver_ClassRec, and made the FT_xxx -> TT_xxx casts in the body of 
the functions themselves.

Nothing dramatic :-) I noticed that the definition of Set_Pixel_Sizes() 
in ttdriver.c was missing two parameters (which are unused), but other 
than that, nothing that would cause your friend further problems 
AFAICT.

On the upside, the function pointers no longer need casting in 
tt_driver_class; I can send in the patch if anyone's interested.

> Our test harness now runs perfectly on Palm OS, so these changes seem 
> to
> have had the desired effect.
Great!

Best regards,

Steve Hartwell