1250229 - Investigate IVS test failures on Android

Status: NEW

(Core :: Graphics: Text, defect, P3)

Description

[Brendan Dahl [:bdahl]]

Moving conversation from https://bugzilla.mozilla.org/show_bug.cgi?id=1248248#c12

Tests that are failing on android:
layout/reftests/font-face/ivs-1.html | image comparison (==), max difference: 255, number of differing pixels: 988
layout/reftests/font-face/cjkcisvs-1.html | image comparison (==), max difference: 255, number of differing pixels: 4294
layout/reftests/writing-mode/1248248-1-orientation-break-glyphrun.html | image comparison (==), max difference: 255, number of differing pixels: 1774

Comment 1

[Brendan Dahl [:bdahl]]

Attachment: image1_ivs-1.png

Comment 2

[Brendan Dahl [:bdahl]]

Attachment: image2_ivs-1.png

Comment 3

[Brendan Dahl [:bdahl]]

Attachment: image1_glyphrun.png

Comment 4

[Brendan Dahl [:bdahl]]

Attachment: image2_glyphrun.png

Comment 5

[Jonathan Kew [:jfkthame]]

From the images here, it appears there's a real bug of some kind. It looks like we're loading the correct test fonts, but the variation selectors are being ignored.

(The glyphrun images also show that we're apparently cutting off part of the test area; if this is a general problem across all tests, it suggests we could be missing important differences in some cases. In this particular case, though, we can still see enough that it's clear we are getting the non-variant glyphs.)

Question: do our android builds use our in-tree freetype, or do they link against a system freetype library? If the latter, can we find out what version is present?

Comment 6

[:Margaret Leibovic]

Could this be an issue with downloadable fonts?

Comment 7

[Jonathan Kew [:jfkthame]]

(In reply to :Margaret Leibovic from comment #6)
> Could this be an issue with downloadable fonts?

I don't think so, because the screenshots look to me like they've loaded the expected font, not used some other fallback; they've just failed to respect the variation selectors. Which could be due to an old freetype version, I think, IF we're using a system lib rather than our in-tree one.

Comment 8

[Sebastian Kaspari (:sebastian; :pocmo)]

(In reply to :Margaret Leibovic from comment #6)
> Could this be an issue with downloadable fonts?

It should not. For tests we push the fonts with the profile to the device. But just in case I'll flag gbrown: Do you think this could be caused by the changes we did for pushing the fonts with the profile?

Comment 9

[Geoff Brown [:gbrown]]

I cannot think of how. That was done in bug 1197716. I checked that font files are still in the reftest test zip and we still pass --extra-profile-file=fonts to the test harness, so they should be added to the profile.