freetype hack committed

[Egbert Eich]

Keith Packard writes:
 > 
 > Around 22 o'clock on Mar 2, Egbert Eich wrote:
 > 
 > > The code in question:
 > > 
 > >               TT_Size tt_size = (TT_Size)instance->size;
 > >                 if( !( !(instance->load_flags & FT_LOAD_NO_BITMAP) 
 > >                        && tt_size->strike_index != 0xFFFFU ) ) correct=1;
 > >  
 > > was added with 1.36.
 > 
 > Ah, ok.  This is part of the code that Juliusz was upset about; the XTT
 > integration which was not well tested or understood by him.  I didn't take
 > that code on the strength of his objections.  But, if we want to be XFree86
 > 4.4-ish, we may have to.
 > 
 > This code detects when the size requested by the application precisely 
 > matches an available bitmap (but only for TrueType faces).  When this 
 > happens, it forces the exact computation of the bounding box so that
 > metrics will be correct.  
 > 
 > It's just an optimization to speed computation of metrics for outlines, so
 > we can safely disable it an expect a slight performance penalty for 
 > large outline fonts loaded a sizes other than those for which bitmaps are 
 > supplied.  A fix which would generate correct (if slow) results would be:
 > 
 > 	if ((instance->load_flags & FT_LOAD_NO_BITMAP) ||
 > 	    (face->face->face_flags & FT_FACE_FLAG_FIXED_SIZES) == 0) correct = 1;

Thank you. This looks much better. (also the logic is much less obfuscated).
 > 
 > I'd prefer this approach; otherwise, we're depending on the layout of an
 > internal data structure which has an #ifdef that affects the presence of
 > the strike_index member.
 > 

Right. Also we have plenty of other structures above this - not all
of these are external.

I'll add this change and commit it later today.

Thanks,

Egbert.