[ft-devel] change the glyph loading path [Was: rounding bug in MS TT rasterizer]

[b90201047]

Hi,

Quoting David Turner <address@hidden>:
> that's fabulous news. I've investigated the issue, and storing the 
> original font unit
> coordinates is not too trivial at the moment. I hope to hack the fix 
> after the 2.2
> release, which I'm currently preparing by fixing the documentation comments,
> plus other things I'll talk about later.
I am always thinking making driver->load_glyph to only load the raw 
glyph data and hints from the font.  That is, glyphs are loaded as if 
FT_LOAD_NO_SCALE is set and hints are loaded but not parsed.  These raw 
data are then passed to the respective hinting module, be it pshint, 
autofit, or, to be ported, tthint, where scaling and hinting happens.

This is pretty much like what the autofit is doing.  By extending the 
idea to other hinting modules,

* code can be shared.  There are some data structures used by all these 
hinting modules.  Also, driver->load_glyph does not need to handle 
load_flags or do recursive loading anymore.
* glyphs can always be hinted.  Some TrueType CJK fonts does not have 
hints.  Currently, when the TT interpreter is enabled, the glyphs of 
such fonts are not hinted.  That is, it does not fallback to autofit.
* FT_Outline_Get_Orientation works more reliably since it can work on 
the unhinted glyph.  This is especially important for "bad glyph" 
detection, where the ratio of false positive can be largely reduced 
(see the commented-out ft_outline_get_orientation in ftoutln.c).

Under this glyph loading path, tt interpreter has access to the 
unscaled coordinates naturally.

This is a long-term goal, since moving to this path involves huge 
amount of work.  Any comment?

(I am serving the (mandatory) military service here and don't have 
access to the computer until Saturday.  Sorry in advance that I cannot 
respond quickly, if someone would like to discuss this.)