fontconfig: Firefox 31 font quality regression

[dmik]

I noticed that the current SVN build of fontconfig (prior to r1181) gives some strange results when running with Firefox 31 (which still uses the old approach to render fonts not involving pango, see also #75). This needs to be investigated. The screenshots will follow.

[dmik]

ff 31 with old fontconfig

[dmik]

ff 31 with new fontconfig

[dmik]

My guess is that somehow the new font config selects a different pixel size for Workplace Sans for which it doesn't have a bitmap font.

[dmik]

Or, to be more exact, the new fontconfig causes the vector version of Workplace Sans to be matched instead the bitmap version (given the same pattern). As we clearly see that on the first screenshot there are bitmap versions (the menu bar and small letters on the page) while on the second screenshot vector fonts (with not too good hinting) are used. This may also relate to the new algorithm of matching the font weight and slant which makes fontconfig fall back to vector instead of bitmap.

[dmik]

Okay, Silvan quickly found the difference. The old fontconfig forced FC_EMBEDDED_BITMAP=FcTrue for all patterns by default in FcPatternCreate. Doing so gives exactly the same visual results as before. Attaching the new screenshot. A commit from Silvan will follow.

[dmik]

ff 31 with FC_EMBEDDED_BITMAP=FcTrue?

[Silvan Scherrer]

r1184 fixes this regression

[dmik]

Committed r1185 per further discussion with @diver. Note that we should review it all one day because FcDefaultSubstitute is actually supposed to be called by the application as long as FcConfigSubstitute which is a no-op for now (and we need to resolve #75 to make its ll actually work). Once resolved, r1184 and r 1185 should be rolled back I suppose.