Closed Bug 1020826 Opened 5 years ago Closed 5 years ago
Thai tones, vowels missing
str - gfx.direct2d.disabled,true - open pangram.html the first bad revision is mozilla-inbound changeset 59b2dd6b5048 Graphics -------- Adapter Description: Mobile Intel(R) 4 Series Express Chipset Family Adapter Drivers: igdumdx32 igd10umd32 Adapter RAM: Unknown ClearType Parameters: Gamma: 2200 Pixel Structure: R ClearType Level: 100 Enhanced Contrast: 100 Device ID: 0x2a42 DirectWrite Enabled: false (6.2.9200.16571) Driver Date: 10-13-2011 Driver Version: 188.8.131.525 GPU #2 Active: false GPU Accelerated Windows: 1/1 Direct3D 9 Vendor ID: 0x8086 WebGL Renderer: Google Inc. -- ANGLE (Mobile Intel(R) 4 Series Express Chipset Family Direct3D9Ex vs_3_0 ps_3_0) windowLayerManagerRemote: false AzureCanvasBackend: skia AzureContentBackend: cairo AzureFallbackCanvasBackend: cairo AzureSkiaAccelerated: 0
I think this only happen with MS Sans Serif font.
Hmm, this doesn't seem to reproduce for me on Win8.1. What font is actually being used in your case? It can't be MS Sans Serif (as specified in the style), as that doesn't support Thai; so fallback must be happening. Please use the Fonts panel in the Inspector, or the Font Info add-on, to determine what font(s) are involved here.
(In reply to Jonathan Kew (:jfkthame) from comment #4) > Hmm, this doesn't seem to reproduce for me on Win8.1. > > What font is actually being used in your case? It can't be MS Sans Serif (as > specified in the style), as that doesn't support Thai; so fallback must be > happening. Please use the Fonts panel in the Inspector, or the Font Info > add-on, to determine what font(s) are involved here. both Inspector and Font Info show MS Sans Serif
That's very strange. Neither my Win8.1 nor XP machines show this. I'll try Win7 later when I get home. Setting needinfo? to remind myself to look again...
Aha, now I understand what's happening. You have the Windows system locale set to Thai, right? In this case, Windows gives us a different MS Sans Serif font -- one that supports the Windows Thai codepage instead of Latin-1. And so it *does* use MS Sans Serif for the Thai characters. I'm still not sure why the diacritics are disappearing altogether, but I suspect it may have something to do with harfbuzz trying (unsuccessfully) to do PUA-based Thai shaping, as this is a bitmap font and doesn't have OpenType tables. Prior to bug 985220, the MS Sans Serif font, being a legacy bitmap font (.fon), would have gone through the Uniscribe or GDI shaper codepaths rather than harfbuzz. If we can't come up with a simple fix here, we should probably back out that patch for now.
(In reply to Jonathan Kew (:jfkthame) from comment #7) > Aha, now I understand what's happening. You have the Windows system locale > set to Thai, right? yes.
How about porting the font substitution (bug 642093) from DWrite? We don't support bitmap fonts with HWA enabled anyway. It will give more consistent behavior between HWA-enabled and disabled environments.
(In reply to Masatoshi Kimura [:emk] from comment #9) > How about porting the font substitution (bug 642093) from DWrite? We don't > support bitmap fonts with HWA enabled anyway. It will give more consistent > behavior between HWA-enabled and disabled environments. I wondered about that, although it would be a significant behavior change for the GDI case. Still, it's worth considering. However, given that merge day is coming up, I think that for Fx32 we should revert bug 985220 (and bugs 1018034 and 1018551, which depend on it), to restore the previous behavior and give us more time to investigate and test this, and determine how to proceed. So I propose to push backouts of those three bugs tonight. Once we have a solution for this issue ready, we can re-land those bugs - presumably for Fx33, as I don't expect to have time for much further investigation this weekend.
The three backouts: https://hg.mozilla.org/integration/mozilla-inbound/rev/0e387a554646 (bug 1018551) https://hg.mozilla.org/integration/mozilla-inbound/rev/6be2448e325c (bug 1018034) https://hg.mozilla.org/integration/mozilla-inbound/rev/5cf33b3e3d5e (bug 985220) (These are also noted in their individual bugs.) Once these merge to m-c, the regression here should be fixed.
verified fixed on nightly 20140608030203.
Thanks for verifying this. In addition, could you please test the Windows build from https://tbpl.mozilla.org/?tree=Try&rev=91b245c25b80, and confirm whether this also works correctly for you? In this build, the patches from bugs 985220, 1018034 and 1018551 have been re-landed, along with a fix that I think should resolve the Thai vowels/tones problem here, but I'd appreciate you checking the behavior.
(In reply to Jonathan Kew (:jfkthame) from comment #14) Compare to attachment 8434765 [details], that build works fine for me except Sara-um. See screenshot.
The build from https://firstname.lastname@example.org/try-win32/ works correctly.
I was just going to ask you to try that one, then I saw you'd already commented. Thanks for your testing!
You need to log in before you can comment on or make changes to this bug.