1770273 - Reftest 1673510-1.html fails on Win7, apparently due to the web-font failing to load

Status: RESOLVED WONTFIX

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

Description

[Daniel Holbert [:dholbert]]

Reftest layout/reftests/writing-mode/1673510-1.html seems to reliably fail on Windows 7.

It has been annotated as skipped-on-win7 ever since it was landed, with comment "test font doesn't load reliably on win7", i.e. most-likely it was flaky on Win7 (like lots of downloadable-font tests were) due to bug 1499459.

Bug 1499459 seems to have now gone away, though this test still fails on Win7 -- note due too the font failing to load, but due to the test legitimately failing to render the character with the expected orientation (vertical vs. horizontal).

Try run where I removed the skip-if annotation:
https://treeherder.mozilla.org/jobs?repo=try&revision=0b1039b3532a48f3863aceb593237b305e09740e&searchStr=R2

jfkthame, do you know what might be going on here? I imagine this might be hard to investigate without a Win7 environment locally available, but at least it seems to be a reliable failure (based on the fact that it failed in its first run on R-fis and R-swr-fis, at least; I've retriggered bug haven't gotten results yet).

For now I'll just bump the skip-if annotation comment to mention this bug (since it's implicitly referencing bug 1499459 right now, which is obsolete/no-longer-relevant).

Comment 1

[Daniel Holbert [:dholbert]]

Attachment: reftest failure log, suitable for reftest-analyzer

Here's the log snippet for the reftest failure, from this failure:
https://treeherder.mozilla.org/jobs?repo=try&revision=0b1039b3532a48f3863aceb593237b305e09740e&searchStr=R2&selectedTaskRun=QGTDo05EQza9fqakHw8XjA.0
https://treeherder.mozilla.org/logviewer?job_id=378657455&repo=try&lineNumber=17784

In the testcase screenshot, the first squiggle is horizontal (looks like a backwards ~), and the second and third squiggle are upright (looks like a stylized S).

In the reference case screenshot, all three squiggles are horizontal (looks like a backwards ~).

Comment 2

[Daniel Holbert [:dholbert]]

(In reply to Daniel Holbert [:dholbert] from comment #1)

In the reference case screenshot, all three squiggles are horizontal (looks like a backwards ~).

Side note: this^ seems a bit surprising, since when I load the testcase & reference case directly in Linux Nightly, all three squiggles are vertical.

If my local (consistent) rendering is the correct result, then in the win7 test failure, the mismatching parts (squiggles 2 and 3) would actually be places where the testcase is rendering correctly and the reference case is rendering incorrectly.

Comment 3

[Jonathan Kew [:jfkthame]]

The wavy-dash glyphs in the win7 failure screenshots do not look like they come from the correct test font at all; I presume they're coming from a Windows fallback font, which doesn't fully support the relevant vertical-mode glyphs.

So it looks to me like the font is simply failing to load at all on win7. No idea why offhand; maybe the older DirectWrite version there doesn't like something about it. I would suggest we just annotate it as failing (or skip) and move on.

Comment 4

[Daniel Holbert [:dholbert]]

I think you're right, yeah. I just noticed that the Win7 failure's rendering looks exactly like how Windows 11 renders these files when the font load is blocked. (On my ThinkPad with Windows 11, when I load hg-hosted versions of the testcase and reference, I get results that match exactly what I described in comment 1, AND I see this in my web console:

downloadable font: failed to start download [...]
Content Security Policy: The page’s settings blocked the loading of a resource at https://hg.mozilla.org/mozilla-central/raw-file/tip/layout/reftests/fonts/ShipporiMincho-Regular.subset.otf (“default-src”).

(with the external resource blocked, as-expected for stuff hosted on hg.m.o.)

Comment 5

[Daniel Holbert [:dholbert]]

(In reply to Jonathan Kew (:jfkthame) from comment #3)

I would suggest we just annotate it as failing (or skip) and move on.

Note, it was already annotated as skipped (originally under the umbrella of intermittent bug 1499459). But since it seems to fail reliably (per my try run in comment 0), I think it'd be better to annotate it as failing. That way we'll find out if the situation changes (e.g. if the issue goes away). We can always go back to marking it as skipped if it turns out to be intermittent, though.

Comment 6

[Daniel Holbert [:dholbert]]

Attachment: Bug 1770273: Annotate reftest 1673510-1.html as reliably failing on win7 (instead of skipped). r?jfkthame

Based on the retriggers on Try, the test seems to always fail on this platform.
So let's mark it as failing instead of skipped, so that we find out if this
issue becomes fixed at some point.

Comment 7

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/d270cb82e282
Annotate reftest 1673510-1.html as reliably failing on win7 (instead of skipped). r=jfkthame

Comment 8

[bszekely]

https://hg.mozilla.org/mozilla-central/rev/d270cb82e282

Comment 9

[BugBot [:suhaib / :marco]]

The leave-open keyword is there and there is no activity for 6 months.
:dholbert, maybe it's time to close this bug?
For more information, please visit auto_nag documentation.

Comment 10

[Daniel Holbert [:dholbert]]

This bug is tracking the fact that the test fails. The patch that landed was just adjusting the annotation to expect failure (so that we find out if it starts working); but for now the issue remains & let's keep the bug open.