Windows 7 32-bit mozilla-inbound opt test reftest-no-accel https://treeherder.mozilla.org/logviewer.html#?job_id=10037911&repo=mozilla-inbound 02:33:14 INFO - REFTEST TEST-UNEXPECTED-FAIL | file:///C:/slave/test/build/tests/reftest/tests/layout/reftests/writing-mode/abspos/s71-abs-pos-non-replaced-vrl-078-ref.xht | load failed: timed out waiting for test to complete (waiting for onload scripts to complete)

This is a recently added test (May 19) in bug 1079151. The error "waiting for onload scripts to complete" is a bit odd considering there are no onload scripts in the test/reference. There's no script at all AFAICT.

Yes, I don't see how the error message can possibly be true! However, looking in the log immediately preceding the failure, there's a suspicious-looking gfx failure: 02:28:14 INFO - REFTEST TEST-START | file:///C:/slave/test/build/tests/reftest/tests/layout/reftests/writing-mode/abspos/s71-abs-pos-non-replaced-vrl-078.xht 02:28:14 INFO - REFTEST TEST-LOAD | file:///C:/slave/test/build/tests/reftest/tests/layout/reftests/writing-mode/abspos/s71-abs-pos-non-replaced-vrl-078.xht | 12040 / 12432 (96%) 02:28:14 INFO - REFTEST TEST-LOAD | file:///C:/slave/test/build/tests/reftest/tests/layout/reftests/writing-mode/abspos/s71-abs-pos-non-replaced-vrl-078-ref.xht | 12040 / 12432 (96%) 02:28:14 INFO - ??????????p: N[GFX1]: Failed to create similar cairo surface! Size: Size(800,1000) Status: 1 02:28:14 INFO - JavaScript error: chrome://reftest/content/reftest.js, line 1393: NS_ERROR_FAILURE: 02:28:14 INFO - JavaScript error: chrome://reftest/content/reftest-content.js, line 995: TypeError: ret is undefined 02:33:14 INFO - REFTEST TEST-UNEXPECTED-FAIL | file:///C:/slave/test/build/tests/reftest/tests/layout/reftests/writing-mode/abspos/s71-abs-pos-non-replaced-vrl-078-ref.xht | load failed: timed out waiting for test to complete (waiting for onload scripts to complete) My guess is that the reported failure is really fallout from an internal gfx problem, perhaps lack of resources of some kind?

I agree, the root cause is likely the GFX failure: http://mxr.mozilla.org/mozilla-central/source/gfx/2d/DrawTargetCairo.cpp?rev=8538bc4d2cbd#1518

(In reply to Jonathan Kew (:jfkthame) from comment #3) > 02:28:14 INFO - ??????????p: N[GFX1]: Failed to create similar cairo > surface! Size: Size(800,1000) Status: 1 > 02:28:14 INFO - JavaScript error: chrome://reftest/content/reftest.js, line > 1393: NS_ERROR_FAILURE: > 02:28:14 INFO - JavaScript error: > chrome://reftest/content/reftest-content.js, line 995: TypeError: ret is > undefined These seem like the real problem. The second JS error is presumably in this code in SendInitCanvasWithSnapshot: var ret = sendSyncMessage("reftest:InitCanvasWithSnapshot")[0]; gHaveCanvasSnapshot = ret.painted; return ret.painted; I'm not sure where the first JS error is, but it seems likely that it is the result of the "Failed to create similar cairo surface" error, which is in DrawTargetCairo::CreateSimilarDrawTarget.

My theory is that this or some other bug does still block bug 1079151, because no matter whose fault it is that these 96 tests fail to run on Win no-accel, it's still significant to that bug that its 96 (or 192, or whatever) tests are going to get disabled if nobody does anything about making them not fail.

Wups, didn't mean to move the bug, just to make threatening noises at it.

See also bugs 1169165, 1171767, 1172376, and 1174379, which are the same thing happening in slightly different tests. (Even in this bug report, the logs show various different tests from the set being affected.) This looks like a graphics failure of some kind (see comments 3-5), and these tests just happen to come at roughly the point in the overall test run where something in gfx finally rolls over and dies. Milan, any chance someone could look into this from the Windows/gfx side?

Bug 1019563 is the original bug for the Win7 Ru "onload" failures, FWIW. I'm beginning to suspect that this issue is tied to OOM conditions (Win7 32-bit tends to be one of our canary platforms for such things). It's happening really frequently on PGO builds these days, though :(

The failure certainly could be from OOM. If this test is run on the debug, that error message would assert, and I guess we're not seeing that? Anyway, let's see the stack when it crashes: https://treeherder.mozilla.org/#/jobs?repo=try&revision=53ef539cb54e

This latest one doesn't seem to have any graphics errors in it: https://treeherder.mozilla.org/logviewer.html#?repo=b2g-inbound&job_id=2186655

Given the frequency, I'm likely going to have to hide Ru on Win7 pgo before too long. Bug 1019563 is still collecting the bulk of the stars and you can see how frequent it remains. Can we please get some eyes on this soon?

Not sure it's graphics, given comment 5, but let's see if we can gain traction. BenWa, can you take a look?

Which disables the entire writing-mode/abspos/ directory on Win7, since they've started randomly timing out the same way in accelerated reftests as well, on trunk and on aurora, and I'm bored with being the only one who knows enough about the problem to actually star them. On the bright side, we didn't add any other bustage to Ru on Win7 during the months it was hidden.

Pushed by mchang@mozilla.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/6f6ae4f2c283 Skip reftest/writing-mode/abspos/ on Windows 7. r=philor

Bulk assigning P3 to all open intermittent bugs without a priority set in Firefox components per bug 1298978.

I suspect this may have just transformed the problem into bug 1300355.