Rendering for Tibetan long vowel U is broken

[domschl]

On Linux and Android, the Tibetan syllable U (ཨཱུ) is rendered incorrectly.

The following site shows the browser-output when rendering hUM and U, together with bitmaps for the correctly rendered glyphs: Test-site

Affected are:

• Android (tested with 7.1.2beta)
• Linux (tested with harfbuzz 1.4.2)
• macOS & Chrome: rendering seems ok within macOS fonts Kokonor and Kailasa, but breaks with all other Tibetan fonts tested.

[KrasnayaPloshchad]

Confirmed on Windows 10.

Firefox 51.0.1


BabelPad 9.0.0.3


LibreOffice 5.3

What I got surprised are: with Microsoft Himalaya the vowel signs are upside down, with Jomolhari one vowel sign failed to join. Since HarfBuzz is also intergrated in LibreOffice (this includes new cross-platform text layout engine based on HarfBuzz), such problems become wider spread that need to pay more attentions to.

[eroux]

With latest git, running

./hb-view --output-format=png --output-file=U.png --language=tibt --script=tibt /usr/share/fonts/truetype/noto/NotoSansTibetan-Regular.ttf ཨཱུ

gives

(which is wrong)

[eroux]

After a git bisect, I can tell that the bug has been introduced in 2f560ee

[domschl]

@eroux : Nice find! Just changing the one line back seems to fix the rendering of long vowel U for me. I have tested on Arch linux, Chrome and LibreOffice work fine. Of course this needs to be checked by the original author of 2f560ee, don't know if this would break something else.

[jungshik]

@behdad, it seems that matching uniscribe breaks other fonts. What would be a resolution?

[eroux]

@jungshik what do you mean specifically? Is that related to Tibetan or is it a more general statement?

[jungshik]

@eroux, I'm referring to HarfBuzz commit 2f560ee which changed the behavior of Tibetan shaper to match the behavior of Uniscribe. That commit was found by @domschl with bisecting.

[eroux]

What I meant is: what are you referring to when you say

• "matching uniscribe": is it this change in particular or the general idea of matching uniscribe?
• "breaks other fonts": what exactly does get broken? which fonts? with what unicode sequence?

[domschl]

It was @eroux who found the cause of the problem. Currently, we are suggesting a possible fix (#443) which would repair the Tibetan, while keeping the improvements that were intended with 2f560ee . The fix requires a thorough review, which will be the next step.

[jungshik]

"matching uniscribe": is it this change in particular or the general idea of matching uniscribe?

My comment is in this bug so I meant this particular issue.

"breaks other fonts": what exactly does get broken? which fonts? with what unicode sequence?

This bug is about that, no? Microsoft Himalaya is fine, but other Tibetan fonts do not work as expected.

[eroux]

I'm still unsure what your point is... asking it in a different way, do you think #443 has some bugs left or is it fine for you?