1637405 - PUA script itemization

Status: RESOLVED FIXED

(Core :: Layout: Text and Fonts, defect)

Description

[corbett.dav]

User Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.14; rv:77.0) Gecko/20100101 Firefox/77.0

Steps to reproduce:

1. Install the font JuniusX from https://github.com/psb1558/Junicode-New/blob/ff2c8b05c6b77d2c298c7d4bc64778d9179f7952/fonts/JuniusX-Regular.ttf.
2. Go to data:text/html;charset=utf-8,<div style="font-family: JuniusX">m%EF%80%BE

Actual results:

The second character (a combining mark in the Private Use Area) is not centered above the first character (an ASCII m).

Expected results:

The PUA mark should be attached to its base using the anchor defined in the font.

The reason it doesn’t work is that the script itemizer keeps script=Unknown characters in their own separate runs; see https://github.com/harfbuzz/harfbuzz/issues/2401 for more information.

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

Bugbug thinks this bug should belong to this component, but please revert this change in case of error.

Comment 2

[Jonathan Kew [:jfkthame]]

Currently, gfxScriptItemizer checks for characters with Script=Inherited and Script=Common, and "adopts" them into the surrounding script run so that they are shaped together. I think we should do the same for Script=Unknown as well, which is the property value for PUA codepoints; while there's no way to know what script they are "supposed" to be, shaping them together with their context rather than as separate runs means that the font has at least a chance to apply any features it may have.

If PUA characters are used in the context of a complex script like Arabic or Indic, the shaper still won't have any basis for knowing how to apply features like 'init', 'medi', etc that depend on per-character shaping classes; but "global" features like 'liga', 'kern', mark attachments, etc., can be used by the font developer and should work as expected.

Comment 3

[Jonathan Kew [:jfkthame]]

Attachment: Bug 1637405 - Merge PUA characters with Script=Unknown with the surrounding script run during itemization (just like Script=Common). r=jrmuizel

If the script itemizer interrupts runs on encountering Script=Unknown, the PUA character will
be shaped separately from its context and therefore OpenType features cannot take effect.
Allowing characters with Script=Unknown to be "adopted" into the surrounding run makes glyph
shaping possible, where a font wants to apply OpenType features such as ligation, kerning,
or diacritic positioning between a PUA codepoint and the adjacent non-PUA characters.

Comment 4

[Jonathan Kew [:jfkthame]]

Attachment: Bug 1637405 - Add reftest for shaping with a PUA-encoded diacritic. r=jrmuizel

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/8e1cc786b9d3
Merge PUA characters with Script=Unknown with the surrounding script run during itemization (just like Script=Common). r=jrmuizel
https://hg.mozilla.org/integration/autoland/rev/6e8e4b967372
Add reftest for shaping with a PUA-encoded diacritic. r=jrmuizel

Comment 6

[Noemi Erli[:noemi_erli]]

https://hg.mozilla.org/mozilla-central/rev/8e1cc786b9d3
https://hg.mozilla.org/mozilla-central/rev/6e8e4b967372