The line mAliasForMultiFonts.IndexOfIgnoreCase(fontname) causes all fonts to be matches of an alias because IndexOfIgnoreCase returns -1 on failure.

Oops, wanted to add that this causes crashes (at least on OS/2) because the gfxOS2Font class then unsuccessfully tries to create and match fonts that don't exist on the system.

Karl, I guess this is what you originally wanted to do in gfxFontconfigUtils::ResolveFontName. At least it fixes the alias name detected and the OS/2 crashes for me.

Thanks for diagnosing this, Peter. This should be fixed for 1.9.1. It won't cause crashes on Linux as gfxPangoFontGroup doesn't use ResolveFontName, but will confuse MathML on Linux when STIX fonts are not installed (as the code will think they are installed).

Comment on attachment 350301 [details] [diff] [review] fix typo (In reply to comment #2) > Karl, I guess this is what you originally wanted to do in > gfxFontconfigUtils::ResolveFontName. Yes, that is what I intended. Using kNotFound (defined to be -1 in nsAString.h) would be nice, but I see that nsCStringArray::IndexOfIgnoreCase actually returns a bare -1, and this will be sorted out when nsCStringArray is replaced with nsTArray<nsCString> (bug 461047), so I don't mind what you choose.

I'm wondering how the mAliasForMultiFonts changes in bug 449356 will affect OS/2 when a family from pref("font.alias-list", "sans,sans-serif,serif,monospace"); is explicitly specified in CSS, enclosed in quotes. e.g. body { font-family: "sans-serif" } The previous behavior would have been to resolve "sans-serif" to a sorted list of every family installed, whereas the new behavior would simply leave this as "sans-serif". ISTR that OS/2 has its own brand of fontconfig so may behave differently to the brand that I know. Does the OS/2 fontconfig have a concept of sans,sans-serif,serif,monospace? That question is similar to: Is FcConfigSubstitute anything more than a no-op on OS/2? If the answer to those questions is "No", then I think the best thing to do would be to remove the "font.alias-list" pref on OS/2.

Karl, yes, when the OS/2 version of FC is asked for one of the generic fonts, it matches them to default system fonts (unless the user has overridden that). So leaving sans-serif after the substitution is OK. But you should not worry about specifics of our implementation. We have implemented the FC functions that cairo and Mozilla (the only two users on OS/2) currently use. We usually fix it to work again every time we find a problem.

Pushed fix to trunk: http://hg.mozilla.org/mozilla-central/rev/6156d0a39763

Comment on attachment 350301 [details] [diff] [review] fix typo a191=beltzner

Pushed fix to mozilla-1.9.1: http://hg.mozilla.org/releases/mozilla-1.9.1/rev/1fa38a636427