| Summary: | Issue with mark advance zeroing in generic shaper | ||
|---|---|---|---|
| Product: | HarfBuzz | Reporter: | Elie Roux <elie.roux> |
| Component: | src | Assignee: | Behdad Esfahbod <freedesktop> |
| Status: | RESOLVED FIXED | QA Contact: | |
| Severity: | normal | ||
| Priority: | medium | CC: | chris.sherlock79, dr.khaled.hosny, elie.roux, freedesktop, jmadero.dev |
| Version: | unspecified | ||
| Hardware: | Other | ||
| OS: | All | ||
| URL: | http://lists.freedesktop.org/archives/harfbuzz/2013-April/003101.html | ||
| Whiteboard: | |||
| i915 platform: | i915 features: | ||
| Attachments: |
Font showing the bug
buggy result with LibreOffice Good result (with Harfbuzz's hb-view) docx document with the issue. |
||
Created attachment 72364 [details]
buggy result with LibreOffice
Created attachment 72365 [details]
Good result (with Harfbuzz's hb-view)
The bug seems to have appeared in Harfbuzz (version from two days ago), and is still present with latest LibreOffice version. There is another bug with the text ཨཱཿཀ , also present in Harfbuzz. I will bugreport to ICU too. Oops. Sorry for the noise. I thought this is filed against HarfBuzz. Reverting title back to what it was. Can you give additional steps on how you are using this font? What operating system, what language is it? I might have to run this through Ibus but I want to make sure I know what language it is before I test it. Also, what key combination do I put in to get the same result that you show in your two images? Marking as NEEDINFO, once you provide the information mark as UNCONFIRMED and we'll investigate. Thanks! Hello, I'm using the font under Linux, this is tibetan. You cannot reproduce the unicode characters I have pasted, unless you have m17n-bo-ewts (under Linux) or TISE (under Windows), the best it to copy/paste them (གཚོའི་ཁིའུ་ཨཱཿཀ). If you can input tibetan unicode with ewts transliteration, you can write "mtsho'i khiu aH ka" (without the quotes). I am not sure I understood your last question... if I want te reproduce it, I just copy/paste the unicode in LibreOffice and I select the font... Otherwise with hb-view: ./hb-view --output-file=foo.png --output-format=png TestLig.ttf གཚོའི་ཁིའུ་ཨཱཿཀ What else do you need to know? Thank you, -- Elie Should be enough - in the future please make sure to read everything carefully, the bug needs to go back to UNCONFIRMED status for us to know it needs triaged once you provide info requested by QA staff :) Thanks so much for the clear explanation, going to attempt to reproduce after installing those packages Oh, sorry for the workflow... if you are under linux, m17n can be a little tricky to make work... Some clues here: http://www.digitaltibetan.org/index.php/Tibetan_Input_Method_for_Linux_%28Gnome%29 (I used uim). Also their EWTS parser is quite buggy... you shouldn't encounter bugs with the string I've given you though... Also, hb-view is only available when you compile harfbuzz, in the utils/ directory. Thank you, I'm installing TISE to see if I can reproduce this. I can reproduce this after installing TISE on Windows 7, using LibreOffice 4.0.2.2. OK, this is odd. When I open the attached document in Word 2010, I am getting exactly the same thing. Created attachment 78660 [details]
docx document with the issue.
If you want some details: http://lists.freedesktop.org/archives/harfbuzz/2013-April/003101.html If I understand correctly there are two bugs: one with incorrect spacings, and one with the ligature at the beginning that is made or not... Thank you, -- Elie caolanm->khaled: seeing as the original report suggests that "harfbuzz does it right", and we're using harfbuzz by default for 4.1, is this now fixed ? (In reply to comment #14) > caolanm->khaled: seeing as the original report suggests that "harfbuzz does > it right", and we're using harfbuzz by default for 4.1, is this now fixed ? Some of this issue is still remaining and I have it on my radar. I believe you can close this, or reassign to HarfBuzz and someone will find the correct discussion thread on the HarfBuzz list and add it here. http://lists.freedesktop.org/archives/harfbuzz/2013-April/003101.html At the Ngapi hackfest in February we changed the generic shaper to zero advance of glyphs that are from characters that are non-spacing marks in Unicode. Before that, the decision to zero mark advance was based on the GDEF class of the glyph. Testing shows that Uniscribe uses one or the other behavior in different shapers and now we try to do the same. It was brought to my attention however, that if a mark glyph ligates with following non-mark glyphs, HarfBuzz zeros the advance while Uniscribe doesn't. This makes sense. We already have logic in GSUB for synthetic GDEF, to categorize the ligature as mark if all components are mark. I think we want the same logic to be used for the Unicode general category. Ie, if a non-mark is present, change the category of the result to non-mark. Here's a font for testing: https://bugs.freedesktop.org/attachment.cgi?id=72363 Here's the sequence: 0f42,U+0F7C,U+0F60,U+0F72,U+0F42 Uniscribe output: [uni0F42=0+356|uni0F7C0F600F72=2+1782|uni0F42=4+356] HB: [uni0F42=0+356|uni0F7C0F600F72=0+0|uni0F42=4+356] Note that the ligature in question has GDEF class 3 (mark), so it's definitely the Unicode-based logic that is involved here, not GDEF-based. I'll go ahead and add the logic to do this, though it's getting a bit uglier than I like. commit 7e08f1258da229dfaf7e1c4b5c41e5bb83906cb0 Author: Behdad Esfahbod <behdad@behdad.org> Date: Mon May 27 14:48:34 2013 -0400 Don't zero advance of mark-non-mark ligatures If there's a mark ligating forward with non-mark, they were inheriting the GC of the mark and later get advance-zeroed. Don't do that if there's any non-mark glyph in the ligature. Sample test: U+1780,U+17D2,U+179F with Kh-Metal-Chrieng.ttf Also: Bug 58922 - Issue with mark advance zeroing in generic shaper |
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.
Created attachment 72363 [details] Font showing the bug When using writer if I type the text གཚོའི་ཁིའུ་ using the attached font TestLig.ttf, I can see a buggy result (I'll attach the result in a later note), with an enormous space between the two syllables. The correct result is basically the same but without the space, I'll attach a correct result also in a later note. This bug comes from a quite rare use case in the layout engine: the font contains in a ccmp table (lookup 2) the ligature: uni0F7C(ོ) uni0F60(འ) uni0F72(ི) -> uni0F7Cuni0F60uni0F72 : the first is zero-width mark apllying to a glyph (in my example ཚ), and the second is not (the third doesn't really matter here), the result is thus a non-zero-width mark... Some OT layout engines also have difficulties with this case: - ConTeXt/LuaTeX had the same bug but I reported it and it's now fixed - on Debian stable, the layout engine behind gedit has it too - harfbuzz seems to handle it correctly. Please tell me if I can provide more information or if I should report it to some other place (ICU ?).