Closed Bug 1317047 Opened 7 years ago Closed 7 years ago

Intermittent writing-modes-3/logical-physical-mapping-001.html == writing-modes-3/logical-physical-mapping-001-ref.html | image comparison, max difference: 255, number of differing pixels: 21663

Categories

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

defect
Not set
normal

Tracking

()

RESOLVED FIXED
mozilla54
Tracking Status
firefox52 --- fixed
firefox-esr52 --- fixed
firefox53 --- fixed
firefox54 --- fixed

People

(Reporter: intermittent-bug-filer, Assigned: manishearth)

Details

(Keywords: intermittent-failure, Whiteboard: [stockwell fixed])

Attachments

(2 files)

this increased in frequency the last week, can we do some basic investigation here- feel free to ni back to me.
Flags: needinfo?(gbrown)
This test fails mostly on debug, mostly on e10s, and mostly on windows, but does affect a variety of platforms.

The test was added in bug 1316172 and, except for short periods, has been failing intermittently ever since.

:manishearth -- Can you make this test more reliable? Otherwise, consider disabling it, at least on Windows Debug.
Flags: needinfo?(gbrown) → needinfo?(manishearth)
Looks like a table layout bug.

https://hg.mozilla.org/mozilla-central/raw-file/tip/layout/tools/reftest/reftest-analyzer.xhtml#logurl=https://archive.mozilla.org/pub/firefox/tinderbox-builds/mozilla-central-win64-debug/1486942246/mozilla-central_win8_64-debug_test-reftest-e10s-1-bm110-tests1-windows-build6.txt.gz&only_show_unexpected=1

This can be worked around by making each cell the same width, since the table isn't the main feature of the test.

I do think the actual table layout intermittent behind this should be fixed too, but I don't know how to go about that. Xidorn, any idea who would be able to look into that?
Assignee: nobody → manishearth
Status: NEW → ASSIGNED
Flags: needinfo?(manishearth) → needinfo?(xidorn+moz)
The table-layout bug may be a manifestation of our problems with intrinsic inline size of orthogonal frames; see bug 1332555. (Compare also bug 1311208, bug 1227616, and I'm sure there are others...)
Comment on attachment 8841105 [details]
Bug 1317047: Make logical-physical-mapping-001 less dependent on flaky table layout;

https://reviewboard.mozilla.org/r/115444/#review116974
Attachment #8841105 - Flags: review?(xidorn+moz) → review+
Pushed by xquan@mozilla.com:
https://hg.mozilla.org/integration/autoland/rev/3fb3d10919d0
Make logical-physical-mapping-001 less dependent on flaky table layout; r=xidorn
https://hg.mozilla.org/mozilla-central/rev/3fb3d10919d0
Status: ASSIGNED → RESOLVED
Closed: 7 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla54
jfkthame can probably have a look at that issue and it seems he has already commented about that :)
Flags: needinfo?(xidorn+moz)
Jonathan, can you take a look at this, we are still seeing failures at a pretty good rate on win7/8 debug reftests.
Flags: needinfo?(jfkthame)
Whiteboard: [stockwell needswork]
Could you point me at a treeherder link for a push with failures, to take a look at logs/reftest images/etc? Thanks.
Flags: needinfo?(jfkthame) → needinfo?(jmaher)
sure, from today on autoland:
https://treeherder.mozilla.org/#/jobs?repo=autoland&revision=87bf65872b562834b3b13eea0074c249cc355503&selectedJob=81031845

the win-debug r1 job failed (and is selected).  My eyes for what is incorrect yield confusion- I will trust you know what is right here :)
Flags: needinfo?(jmaher)
That's....weird. :manishearth's patch here gave all those table cells a specific size of 80px x 80px, and when I view the testcase locally, I see square cells. Yet in those screenshots there is clearly some width variation going on, and they are far from square anyhow.

Oh.... I think I see what may be going on. The table cell sizes are actually specified as 100px by 100px (because <td> has width:80px, height: 80px, but also has padding:10px, which contributes 20px altogether in each dimension). And there are 9 cells per row, for a total of 900px, plus whatever the border separating the cells contributes.

But our reftest page size is only 800px wide, IIRC. And so the table ends up getting "compressed" horizontally when the reftest harness loads these files. And that compression behaves somewhat inconsistently due to the intrinsic-size-of-orthogonal-flow bug mentioned earlier (comment 17). The flakiness is worst on slower systems, no doubt, which is why the failures tend to show up in debug builds.

I think we can fix this, therefore, by just making the table cells smaller such that they retain their specified sizes, and don't have to be horizontally compressed to fit into the reftest viewport. I'll push a try job to see how that goes... https://treeherder.mozilla.org/#/jobs?repo=try&revision=47ebae9813a598e3e98680196042f08c7a38e9e7.
So far, try run looks green; more re-triggers on the way, but I'm hopeful this will finally fix the flakiness.
Attachment #8843073 - Flags: review?(manishearth)
Attachment #8843073 - Flags: review?(manishearth) → review+
https://hg.mozilla.org/integration/mozilla-inbound/rev/1402d02dd61e0f69c00da5a765d6083a627d35eb
Bug 1317047 followup, make the table cells smaller to avoid horizontal compression of columns. r=manishearth
Whiteboard: [stockwell needswork] → [stockwell fixed]
Jonathan pushed the follow-up to 53 as well while it was still on Aurora.
https://hg.mozilla.org/releases/mozilla-aurora/rev/a219509df28e
You need to log in before you can comment on or make changes to this bug.