q is slanted the wrong way

[domoritz]

Describe the bug

To Reproduce

https://rsms.me/inter/lab/?varWeight=490&varSlant=10&varfont=1

Environment

• OS: macOS
• Browser: Chrome

[lordsutch]

I can verify this is also the case with macOS 10.14.6 using Inter.var.ttf, at least in the following apps:

• Chrome (77.0.3865.75)
• Font Book
• LibreOffice Writer (6.3.0.1)

The problem does not occur in Firefox (70.0b6), but this is likely because Firefox can't slant the two-way variable font at all.

The problem is not present in Inter-italic.var.ttf.

My guess is that the problem is the result of the q glyph just being a mirror of the p glyph, and the two-way variable font not accounting for the mirroring when applying the slant. Now whether this is a macOS bug or a bug in the font itself, I can't say.

[lordsutch]

I can report that this problem also exists on Chrome OS 76.0.3809.136, suggesting either the bug is shared between macOS and Chrome OS (and presumably all Freetype-based font rendering implementations) or it is a font bug.

[ToadKing]

I also experience the bug on Windows 10 in both Firefox Beta 70.0b9 and Chrome Dev 78.0.3904.17. Microsoft Edge 44.18362.329.0 does not show the bug.

[jcbp]

I experience the bug on Linux Mint Tessa (19.1) in Firefox 69.0.1 and Chromium 76.0.3809.100.
I also experience the bug with the opening exclamation mark (¡)

[thundernixon]

the problem is the result of the q glyph just being a mirror of the p glyph

Right, mirrored components don't work as expected with transformations. I'm not sure of the precise reason for this, but basically, it just is treated as a reference to the "parent" glyph, and mirrors that glyph's transformations.

This regressed between release 3.8 to 3.10. So, my guess is, the error was introduced either in makefile or misc/fontbuild, possibly with the introduction of the compact-font-names flag.

There is a function in the misc/fontbuild script that is meant to go through glyphs and decompose mirrored components, when generating UFOs to prep for the variable font. This function solved this problem previously.

From what I can tell by adding print statements into misc/fontbuild, the composedGlyphIsNonTrivial() function is never run. I assume this should probably be run as part of the make designspace flow.

[rsms]

This was a regression introduced in a recent upgrade of one of the 3rd party libraries used by fontbuild (fontmake).

Here is a build from master as is at the time of writing this comment.
Inter-3.11-ba71122abf.zip

[rsms]

New release is out that includes a fix: https://github.com/rsms/inter/releases/tag/v3.11