479154 - test_dont_use_document_colors.html seems to fail occasionally on linux

Status: NEW

(Core :: CSS Parsing and Computation, defect)

Description

[Jeff Muizelaar [:jrmuizel]]

I don't see the failure when running the test individually and only sometimes see it when running the full mochitest. Gavin has seen the problem as well.

I get the following failure:
46444 ERROR TEST-UNEXPECTED-FAIL | /tests/layout/style/test/test_dont_use_document_colors.html | color is blocked - got "rgb(0, 0, 0)", expected "rgb(16, 16, 16)"

Comment 1

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Only the one failure?  All the other tests pass?

Comment 2

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: debugging patch

Could you try this patch and attach what it outputs if the test fails?  (And, optionally, the comparison for when it passes, although I can generate that pretty easily here.)

(This patch was written assuming that it's only that one test that fails, and that it's always the same test that fails.  If it's actually a whole bunch of tests that fail, I'd guess the problem is more likely an issue with the timers.)

Comment 3

[Jeff Muizelaar [:jrmuizel]]

Attachment: Debuging log from test failure

Here's a log (http://people.mozilla.org/~jmuizelaar/log3.out) from an entire mochitest run that failed. The only test in test_dont_use_document_colors.html that it fails is the one mentioned above.

I had to munge the patch a little bit because I was using an older tree. Hopefully, the results are still valid. I've attached the patch that I used, it's against a tree with c7206236523e as the tip.

Comment 4

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: another debugging patch

So it doesn't look like the simple timing issue I was worried about, but I think it could be a timing/ordering issue related to the order the pres contexts get notified about the preference change... in particular, your log has one occurrence of "Considering whether to ignore color:yellow declaration at level 4." after the pref change stuff starts happening but before I would have expected it.  So this adds some debugging code to get a little more information about that.

I'm also puzzled by where the rgb(16,16,16) color comes from in the first place.  I'd expect the default color to be the rgb(0, 0, 0), although it could be rgb(16,16,16).  But I wouldn't expect it to be both at once.  So I added some debugging of the default color stuff too.


So could you try again this patch?  Also, when you're logging, could you run the mochitest with --console-level=INFO so the mochitest output gets mixed in?

Finally, since this has stack dumps inside the output, could you then pipe the output through tools/rb/fix-linux-stack.pl before attaching it?