linux: handle basic metric-compatibile font fallback

linux: handle basic metric-compatibile font fallback

When a page CSS asks for:
  font-family: Arial, sans-serif;
and you don't have Arial, fontconfig tries to provide a substitute.

However, as far as I can tell there is no way to distinguish fontconfig picking
Liberation Sans (a metric-equivalent substitute for Arial) or it picking a bad
substitute like Times New Roman.  So historically we didn't allow fontconfig to
do any fallback at all.  (In this example, we would then match attempt to match
sans-serif, which as a last resort we do allow fontconfig to do fallback on, so
it could pick e.g. DejaVu Sans or any other font matching "sans".)

Frustratingly, fontconfig knows whether a font substitute is an equivalent font
or just a last-resort fallback one -- in the XML files the equivalents are
marked with 'binding="same"' -- but I played around with the API for a few
hours and couldn't figure out any way to get this information out of it. 

So I hardcode a list of equivalent fonts.  Ugh.  This fixes some sites where
they specify Arial and then have pixel-width-specified page layout that breaks
when you use any other sans-serif font.

BUG=18159, 32005, 33125
,others I've since forgotten about
TEST=uninstall Arial; visit one of the sites mentioned in the bug list

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=40912

[Evan Martin, 2010-03-05 01:49:24]

http://i.imgur.com/YUdJI.png

Left to right:
1) with msttcorefonts (looks like Windows)
2) current Chrome without msttcorefonts (note metrics are all different)
3) with my patch without msttcorefonts (it doesn't look great, but it looks
better)

[agl, 2010-03-05 16:25:44]

LGTM

http://codereview.chromium.org/668127/diff/5/1003
File skia/ext/SkFontHost_fontconfig_direct.cpp (right):

http://codereview.chromium.org/668127/diff/5/1003#newcode39
skia/ext/SkFontHost_fontconfig_direct.cpp:39: "Arial", "Liberation Sans",
"Albany", "Albany AMT", NULL,
This is pretty much too small to care, so it's still LGTM if you ignore all
this:

This scheme will create an array of pointers, each of which will need a
relocation record. Then you're throwing them into tiny red-black trees where the
branches and pointers dominate the actual data.

Rather one could do:

struct FontEquivClass {
  char name[31];  // static error if too small
  char class;
}

static const FontEquivClass kFontEquivClasses[] = {
  { "Arial", 0 },
  { "Liberation Sans", 0},
  ...
  { "Times New Roman", 1},
  ...
}

Then you just need to lookup each of the two font names and, if they both have
the same class, then they're a match.

You can either scan linearly or block select the records, pipe through sort and
use bsearch(). The latter means that it's slightly more difficult to see the
classes in the source code. (But even a linear scan would be much faster than
the STL set stuff.)

http://codereview.chromium.org/668127/diff/5/1003#newcode50
skia/ext/SkFontHost_fontconfig_direct.cpp:50: for (int i = 0; i <
sizeof(fallbacks)/sizeof(fallbacks[0]); ++i) {
int -> size_t to match sizeof()

[Evan Martin, 2010-03-08 18:26:08]

Ok, rewritten with your idea (minor tweak: class id first so the strings line up
better).  Please double-check my logic.  (I suspect my previous patch was wrong
because it was case sensitive.)

[agl, 2010-03-08 18:31:40]

LGTM

http://codereview.chromium.org/668127/diff/1015/8
File skia/ext/SkFontHost_fontconfig_direct.cpp (right):

http://codereview.chromium.org/668127/diff/1015/8#newcode76
skia/ext/SkFontHost_fontconfig_direct.cpp:76: static const int kClassCount =
s/int/size_t/

[Evan Martin, 2010-03-08 18:37:28]

On 2010/03/08 18:31:40, agl wrote:
> http://codereview.chromium.org/668127/diff/1015/8#newcode76
> skia/ext/SkFontHost_fontconfig_direct.cpp:76: static const int kClassCount =
> s/int/size_t/

Thanks.  I wonder why my compiler isn't complaining about these...