Re: [Devel] Hinter changes

[David Turner]

Hello,

Rayiner Hashem wrote:

> Whoa. All the hinter changes have gotten really confusing. I would appreciate 
> it if someone would disabuse me of any incorrect notions I have about the 
> 2.1.5 codebase:
>
> 1) The auto-hinter is used for most fonts.
Nope, only for those formats that do not provide their own hinters, i.e. 
TrueType
(when the bytecode interpreter isn't used) as well as PFR

The Type1, Type42, CID and CFF formats use the Postscript hinter

> 2) The unpatented hinter is used for a few fonts that require it.

In 2.1.5, the unpatented hinter is disabled. This is due to internal 
limitations of the engine,
(the unpatented hinter works very well otherwise). This should change 
shortly

> 3) The native TrueType hinter is never used.
>  
Except if you manually activate it.

Additionnally, it's possible to force the use of the auto-hinter with 
the FT_LOAD_FORCE_AUTOHINT
flag when loading glyphs. I don't remember precisely what configuration 
option in fontconfig or Xft
can be used..

> Is that correct? Also, what happened to the PSHinter? IIRC, a lot of the 
> improvements in rendering in the 2.1.3-2.1.4 seemed to happen in the 
> PSHinter, and now word is that the PSHinter is broken?
>  
It's still used, but it has certain problems, though I hope it's not too 
dramatic

> I don't know what kind of results everyone else is getting, but my fonts are 
> getting worse, not better. I think a lot of the effort spent on more precise 
> hinting has really affected glyph shape. Part of this has to do with my 
> setup: Postscript fonts on a 133dpi screen. Standard TrueType hinting looks 
> absolutely terrible, because 1-pixel wide strokes are too thin to be legible. 
> Also, what is the status of sub-pixel rendering? Last I recall, Ichanges in 
> Freetype broke Xft's sub-pixel AA somewhere between 2.1.2 and 2.1.3. From my 
> tests on the FreeType from CVS, this still seems to be the case.
Did you update to the latests fontconfig/libXft2. If not, the 
explanation could come from the fact that
the first LCD-optimized algorithm in Xft used some intrisic properties 
of the auto and postscript hinter
(i.e. that stem widths were always integer pixels) that changed in later 
released (because allowing
fractional widths also increases shape consistency significantly).

Some enhancements have been made in FreeType to still support LCD 
rendering, but I don't know
if libXft has been updated to use them. I suppose so though.. Keith, any 
info on the topic ?

> Lastly, my fonts keep changing drastically with every release of 
Freetype. Should this

> really be happening with minor version number bumps, or am I the only one 
> seeing such changes?
>  
We try to improve the various hinters on each new release, so it's 
normal to see that not two
distinct minor release generate the same results. Finding algorithms 
that work well on all kinds
of fonts is extremely difficult, and we don't necessarily have the time 
to test each increment on
a very large collection.

If you encounter something that qualifies as a regression, please inform 
us. Be aware though that
many distinct opinions about what constitutes "good rendering" exist 
when it comes to fonts, so
the question isn't necessarily easy..

Hope this helps,

- David Turner
- The FreeType Project  (www.freetype.org)