it appears we need to do linux32 testing, probably a priority behind windows, but something we should look into this quarter.
for reference, I built a docker image for ubuntu 16.04 in bug 1281179. I imagine we would need to take that work and duplicate it using ubuntu 12.04 x32 and cross reference any differences in the puppet configs.
I think we decided not to do linux32 testing? I remember asking about this a while back and being told it's not an issue.
We decided not to do linux32 *Talos*
no talos, but unittests- yes!
https://treeherder.mozilla.org/#/jobs?repo=try&revision=0fe33101a76352993ac66987c586d0a17617f8d3 is mostly jmaher's work. The cppunit failure is consistent with pulseaudio not running: In a one-click loaner, that test succeeds when pulseaudio is running and fails the same way if pulseaudio is killed.
Assignee: nobody → gbrown
right now there are a series of failures to sort out: 1) update web-platform-tests to have manifests which support linux32 as well as linux64 2) mochitest-plain-5 (non-e10s) has an issue with moz-icon 3) mochitest-media (mda task) has common failures 4) reftest issues, 1 moz-icon issue, bug 1319109, bug 1318953, and some gtk issues that can be fixed via manifest changes 5) xpcshell-6 appears to be a moz-icon issue What remains is figuring out moz-icon:// and mochitest-media jobs. Outside of that, we might have a lingering reftest issue or two left to investigate.
I'm guessing https://treeherder.mozilla.org/#/jobs?repo=try&revision=e3102a63b0db9fafef52ab6cf34d2ce82ab89bb4 is the one corresponding to comment 6. As the buildbot reftests are not currently visible, if these tests are useful (see also bug 1209932), then IMO it would be reasonable to just annotate failures as they are on taskcluster and file bugs for the failures. There may be some disagreement on that approach around bug 1310792. See bug 1309086 comment 9, but I'm not sure that Mats was aware of the visibility status when he made that comment.
(In reply to Joel Maher ( :jmaher) from comment #6) > 5) xpcshell-6 appears to be a moz-icon issue I found that gtk_icon_theme_lookup_by_gicon() failed where it succeeds on my laptop (the subsequent call to gtk_icon_theme_lookup_icon() also fails -- not sure if that is relevant). Perhaps we are missing an :i386 version of a gtk package? I tried installing libgtk2.0-dev:i386 but that failed (conflict with libgtk2.0-dev). https://hg.mozilla.org/mozilla-central/annotate/05328d3102efd4d5fc0696489734d7771d24459f/image/decoders/icon/gtk/nsIconChannel.cpp#l273 https://treeherder.mozilla.org/#/jobs?repo=try&revision=931251f09a3b4ffce376ee6fff9cc87f5d1f4077
I am not sure how gtk and mozicon work together. Also possible fixes might be to upgrade to 16.04? I found that hacking on a loaner didn't get me closer to a solution, primarily installing packages or trying to get additional information. I do think solving the mozicon stuff will get us very close to complete here.
AIUI nsIconChannel is looking for icons for mimetypes. On my system these would be in subdirectories of /usr/share/icons/, so I suggest comparing those folders on the different test machines, perhaps indicating different packages installed. On my system, conventional icon themes such as mate and oxygen provide "unknown" but the "hicolor" and "gnome" themes do not. I expect the test is looking for an icon for text/plain. Only the "oxygen" theme on my system has a text-plain.* icon. Many themes have text-x-generic* icons. Perhaps something should convert plain to x-generic somewhere. I don't know. If this also fails on 16.04, then maybe this is not working for users either, and the test has correctly detected a bug.
Thanks Karl. This does fail just the same on 16.04 (and a few other xpcshell tests fail on 16.04 - likely unrelated - so I'll concentrate on 12.04 for now). I'll check on /usr/share/icons.
I filed a few bugs which would be blocking us from making this green. They are: * mochitest-media job (trouble getting audio/video to work) * wpt green up (audio/video problems) * mozicon in xpcshell, 1 test case the rest would be landing what we have and scheduling the proper set of tests.
Created attachment 8821183 [details] [diff] [review] linux32_jmaher_s_original_work_a_little_help adding a patch here which is our latest WIP- still missing support for flash as web-platform-tests have issues, and the mozicon support (2 different tests)
Created attachment 8821612 [details] [diff] [review] enable tc linux32 tests We need to skip 2 wpt tests for flash failures and 1 xpcshell + 1 mochitest for mozicon failures (dependent bugs). Otherwise, this image works fine. mochitest-bc is slow on tc linux32, as it is on tc linux64: I'm both increasing # chunks and max-run-time as a belt-and-suspenders type fix. https://treeherder.mozilla.org/#/jobs?repo=try&revision=22b7ee16da27eb45b432bfd04df1b1252707565a
Comment on attachment 8821612 [details] [diff] [review] enable tc linux32 tests Review of attachment 8821612 [details] [diff] [review]: ----------------------------------------------------------------- r- for the test sets, otherwise this is great. If you want to land this now and focus on a refactor later, I believe :dustin has a refactor in the works- lets just add firefox-ui-tests at a minimum. ::: taskcluster/ci/desktop-test/test-sets.yml @@ +77,5 @@ > > +linux32-tests: > + - cppunit > + - crashtest > + - external-media-tests we need firefox-ui-tests in here- also how is this different from linux64 tests? could we find a way to split into a shared test set? ::: taskcluster/ci/desktop-test/tests.yml @@ +292,4 @@ > linux64-jsdcov/opt: 7200 > linux64-ccov/opt: 7200 > linux64/debug: 5400 > + linux32/debug: 5400 I suspect */debug will be 5400, too bad we cannot make it that.
Attachment #8821612 - Flags: review?(jmaher) → review-
(In reply to Joel Maher ( :jmaher) from comment #15) > we need firefox-ui-tests in here- also how is this different from linux64 > tests? could we find a way to split into a shared test set? Are you sure about firefox-ui-tests? I don't see buildbot linux32 firefox-ui-tests on central currently. Differences between linux32 and linux64 reflect current differences for buildbot linux32: - no firefox-ui-tests - no plain reftests - no unaccelerated reftests
(In reply to Geoff Brown [:gbrown] from comment #16) > Differences between linux32 and linux64 reflect current differences for > buildbot linux32: > - no firefox-ui-tests > - no plain reftests > - no unaccelerated reftests Actually, there's more... - mochitest-valgrind - web-platform-tests-wdspec - talos :jmaher verified that linux32 firefox-ui-tests are desired and should be easy to add -- I'll do that.
Created attachment 8821650 [details] [diff] [review] enable tc linux32 tests Now includes firefox-ui tests, which look fine on try: https://treeherder.mozilla.org/#/jobs?repo=try&revision=2ae885e3a35509686d49cf3e6ca6e4b7ea47d60b I'll look into reftests next, but will treat that as a separate patch, and perhaps a separate bug, depending on how review timing works out over the holidays.
Comment on attachment 8821650 [details] [diff] [review] enable tc linux32 tests Review of attachment 8821650 [details] [diff] [review]: ----------------------------------------------------------------- lets do reftests and Wd in other bugs- keep in mind on inbound I believe a refactor just landed so we need to rebase and fix any conflicts here- proably in test-sets.yml.
Attachment #8821650 - Flags: review?(jmaher) → review+
Created attachment 8822525 [details] [diff] [review] enable tc linux32 tests Rebased for recent changes to tc yml files. r=jmaher https://treeherder.mozilla.org/#/jobs?repo=try&revision=abd43fd072ee5bfc910345e49b2862f27b8be734
Pushed by email@example.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/5ddb2e58bb74 Run linux32 tests in taskcluster; r=jmaher
sorry had to back this out for linux web platform failures, https://treeherder.mozilla.org/logviewer.html#?job_id=66460982&repo=mozilla-inbound&lineNumber=2776
Backout by firstname.lastname@example.org: https://hg.mozilla.org/integration/mozilla-inbound/rev/c216c29d0c18 Backed out changeset 5ddb2e58bb74
I don't understand the failures and don't see them when I push to try. I'm going to push again today, breaking the patch down into a few different components to try to isolate the problem if it recurs.
Pushed by email@example.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/afa19305d88a Update Ubuntu 12.04 docker image for linux32; r=jmaher
Pushed by firstname.lastname@example.org: https://hg.mozilla.org/integration/mozilla-inbound/rev/4a298007ca24 Update Ubuntu 16.04 docker image for linux32; r=jmaher
Pushed by email@example.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/b33eab5ae47e Run linux32 tests in taskcluster (tier 2); r=jmaher
Status: NEW → RESOLVED
Last Resolved: a year ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla53
You need to log in before you can comment on or make changes to this bug.