win dw: Fallback to aliased if ClearType 'succeeds' but returns empty

win dw: Fallback to aliased if ClearType 'succeeds' but returns empty

This appears to happen for some CJK fonts where DW only returns valid
data for bitmaps, not for cleartype data.

This has an artifact (of course) which is that during the run where this
failure case is encountered, cleartype will appear incongrously next to
aliased, which looks poor. So, we'll need to fallback at a higher level.

However, for M37, having something visible is more desirable than
rendering blank space for characters.

For reference, two screenshots. M37 Stable and then with this patch:

http://imgur.com/9pf3rB9,EiTb6Li

See https://code.google.com/p/chromium/issues/detail?id=396624#c10 for
content of test html file.

R=shrikant@chromium.org,bungeman@chromium.org,eae@chromium.org,cpu@chromium.org
BUG=chromium:407945

[scottmg, 2014-08-27 20:47:54]

(Sorry, keep forgetting Skia needs to be patched out of tree. Same patch,
different Base URL.)

[Shrikant Kelkar, 2014-08-27 21:09:13]

https://codereview.chromium.org/510563003/diff/1/src/ports/SkScalerContext_wi...
File src/ports/SkScalerContext_win_dw.cpp (right):

https://codereview.chromium.org/510563003/diff/1/src/ports/SkScalerContext_wi...
src/ports/SkScalerContext_win_dw.cpp:401: void
SkScalerContext_DW::generateMetrics(SkGlyph* glyph) {
nit: optional

generateMetricsInternal(SKGlyph* glyph, textureType, renderingMode) {
  move code from current generateMetrics() here
}
and call return generateMetricsInternal(glyph, fTextureType, fRenderingMode)
from generateMetrics()

[reed1, 2014-08-27 21:11:21]

reed@google.com changed reviewers:
+ reed@google.com

[reed1, 2014-08-27 21:11:21]

also looking at this now

[eae, 2014-08-27 21:17:34]

Great work Scott, getting something to work at all is clearly better than what
we have now. Long term we should probably have skia indicate the failure case
and push the logic up to blink somehow.

[reed1, 2014-08-27 21:44:22]

Does this zap us into BW mode if we encounter an empty glyph (with zero bounds)?
Seems like it might.

Ben is experimenting with a modified version of the patch...

[scottmg, 2014-08-27 21:54:31]

On 2014/08/27 21:44:22, reed1 wrote:
> Does this zap us into BW mode if we encounter an empty glyph (with zero
bounds)?
> Seems like it might.

Yes, I'm not sure there's a way to distinguish to this successful/failure/empty
from an empty glyph. But we could retry per-character rather than modifying the
whole run.

> 
> Ben is experimenting with a modified version of the patch...

Great, thanks. Let me know if you have a different one and I can try on my
repro.

[scottmg, 2014-08-27 22:51:08]

Redone here: https://codereview.chromium.org/504343007/ . Closing this.