+++ This bug was initially created as a clone of Bug #553981 +++ Changes to gfxPlatform::SetupClusterBoundaries in bug 553981 fixed layout/reftests/first-letter/329069-2.html with CoreText but not with Uniscribe (same issue as Bug 615445, I assume) or DWrite.
Created attachment 496466 [details] WINNT 5.2 reftest failure Windows Server 2003 doesn't have font support, but that's not really an excuse for getting the clusters wrong.
Created attachment 496556 [details] WINNT 6.1 reftest failure The failure with DWrite shaping is just a baseline position difference and so not related to shaping/fonts but first-letter implementation. Add lang="ko" (to improve font-selection) and an additional (not-first-letter) character does not correct the difference.
The baseline difference is not there when font-size:200% is removed.
Summary: first-letter still splits up Hangul Jamo sequences → first-letter/329069-2.html reftest fails due to inconsistent baseline of font-size:200% first-letter
(In reply to comment #1) > Created attachment 496466 [details] > WINNT 5.2 reftest failure > > Windows Server 2003 doesn't have font support, but that's not really an excuse > for getting the clusters wrong. The patch in bug 615445 (attachment 496786 [details] [diff] [review]) should fix the cluster issue on Win2k3/XP. (It doesn't address the baseline discrepancy, though, so the test with font-size will still fail.)
OS: Windows 7 → Windows XP
Created attachment 496886 [details] Crash test I'm getting a ton of crashes from this checkin: bp-2c5545a5-b271-4329-b41c-782f92101210 bp-b3f5708e-a1d4-469e-bf6c-250a42101210 bp-4a21b3a2-cb00-49cf-8794-7e2cf2101210 bp-80ae45d6-85e5-479c-9022-105a02101210 The regression range points right at this bug: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=9ab76fbcb6d7&tochange=11e328a49e0a Linux x86_64 systems only as far as i can see
Gah sorry wrong bug...
(In reply to comment #1) > Windows Server 2003 doesn't have font support, but that's not really an excuse > for getting the clusters wrong. I think CompressedGlyph::SetMissing should not be overwriting FLAG_NOT_CLUSTER_START: http://hg.mozilla.org/mozilla-central/annotate/61361e67bfc4/gfx/thebes/gfxFont.h#l1786
> I think CompressedGlyph::SetMissing should not be overwriting > FLAG_NOT_CLUSTER_START: Spun off into Bug 618870.
Separated the font-size test, so we have a passing Hangul Jamo test on WINNT 6.1. http://hg.mozilla.org/mozilla-central/rev/bac79600ed97
Summary: first-letter/329069-2.html reftest fails due to inconsistent baseline of font-size:200% first-letter → first-letter/617869-1.html reftest fails due to inconsistent baseline of font-size:200% first-letter
Created attachment 497287 [details] WINNT 6.1 reftest failure Failures are similar on WINNT 5.2 and Mac, suggesting this might be a layout bug and perhaps Linux/FreeType is just getting lucky.
Attachment #496886 - Attachment is obsolete: true
Test is: https://dxr.mozilla.org/mozilla-central/source/layout/reftests/first-letter/617869-1.html and then click the Raw link in the righthand side column Reference file is: https://dxr.mozilla.org/mozilla-central/source/layout/reftests/first-letter/617869-1-ref.html and then click the Raw link in the righthand side column When I try such test with Firefox 52.8.0 ESR (under Linux Debian 9.4, 64-bit), I get exactly the reference file. WORKSFORME
You need to log in before you can comment on or make changes to this bug.