Theta hat doesn't have y offset

[mitchellh]

Character: θ̂
Font: https://www.jetbrains.com/lp/mono/ (non-variable version)
Backends: Freetype and CoreText (both appear the same)
Harfbuzz version: 8.2.2, 8.3.0, and 74e8971 all exhibit the issue

$ hb-shape --output-format=json ~/Library/Fonts/JetBrainsMono-Regular.ttf θ̂ | jq .
[
  {
    "g": "theta",
    "cl": 0,
    "dx": 0,
    "dy": 0,
    "ax": 600,
    "ay": 0
  },
  {
    "g": "uni0302",
    "cl": 0,
    "dx": 0,
    "dy": 0,
    "ax": 0,
    "ay": 0
  }
]

If I follow the advance directly and render, the rendering appears like this (enlarged for effect):

Notice the hat is inside the theta. I reimplemented my shaping using CoreText directly for shaping, and it provides the following information:

codepoint=952 glyph=445 pos=graphics.geometry.Point{ .x = 0.0e+00, .y = 0.0e+00 } advance=graphics.geometry.Size{ .width = 1.8e+01, .height = 5.2875000000000005e+00 } index=0 cluster=0
codepoint=770 glyph=372 pos=graphics.geometry.Point{ .x = 1.8e+01, .y = 5.2875000000000005e+00 } advance=graphics.geometry.Size{ .width = 0.0e+00, .height = -5.2875000000000005e+00 } index=1 cluster=0

Note the y advance being set (advance.height) and then being unset. This results in the theta-hat rendering correctly. Other applications (unsure if they're using hb) appear to render theta-hat correctly with this font as well. I'm unsure if I'm just configuring Harfbuzz incorrectly, but I'm unable to find a solution to this. Thank you for an excellent library.

[khaledhosny]

This is a font bug. The font does not have mark positioning for the theta glyph, but it has it for other glyphs. Core Text seems to be doing some fallback mark positioning here, but HarfBuzz does not do this unless the font has no mark positioning data at all.

[mitchellh]

Got it, I'll report the issue to JB. Thank you.

EDIT: Reported here: JetBrains/JetBrainsMono#660

[behdad]

This is a font bug. The font does not have mark positioning for the theta glyph, but it has it for other glyphs. Core Text seems to be doing some fallback mark positioning here, but HarfBuzz does not do this unless the font has no mark positioning data at all.

It's really curious that CoreText does this level of fallback, since the glyph might be designed to just fit... Maybe @nedley can tell us what CoreText is doing.

[mitchellh]

If it's helpful, I did some additional testing. The following fonts all render θ̂ correctly with CoreText shaping and do not with HarfBuzz. I tried to pick fonts that were built-in to the system.

• Andale Mono
• Menlo
• Monaco
• Courier New

It feels strange to assume all of these fonts have bugs...?

[khaledhosny]

• Andale Mono: does not have combining circumflex
• Menlo: looks fine:
  
• Monoca: Does not have theta
• Courier New: Has mark positioning only for Arabic and Hebrew, and additionally claasifies combining circumflex as base glyph.

[nedley]

CoreText may perform fallback mark positioning when a mark glyph has not been substituted or attached.

[behdad]

Humm. Interesting. Thanks.

I know in the past the argument against doing that for us was "what if the designer put it in the right place by design?". I'll experiment a bit and see if we can do the same without breaking the world. AFAIU DWrite doesn't do that.

We probably decide this on a per-shaper basis...

cc @dscorbett

[jfkthame]

As a possible heuristic: maybe try this only for marks with CCC=220 (below) or CCC=230 (above), not other classes; and perhaps initially just in the generic shaper, not in the script-specific ones.

[behdad]

What system and what HB version are you testing on?

• Andale Mono
• Menlo

With Menlo on MacOS 13.3.1 (22E261) it works fine with HarfBuzz for me. I don't get the exact same positioning as CoreText, but it doesn't look wrong.

• Monaco

My Mac's Monaco doesn't support theta.

• Courier New

It feels strange to assume all of these fonts have bugs...?