Not evident how it is related to plugins and plugin process but after http://trac.webkit.org/changeset/112889 + http://trac.webkit.org/changeset/113703 3 tests started to crash. Most of the cases these are the following (somtimes it varies): plugins/change-widget-and-click-crash.html fast/frames/iframe-remove-after-id-change.html compositing/repaint/become-overlay-composited-layer.html These tests work on themselves, but fail if you run all tests.
Let me repeat some information from the previous bug report. I tried but I could not reproduce the regression reliably. When I runned tests in Ubuntu I saw 3 crashes. Not always the same ones. (This is the case on the bot as well.) The strange thing is that: with ulimit -c unlimited + echo core.%p > /proc/sys/kernel/core_pattern I only got one core dump for the whole test session. Seems like the other two crashes detected by nrwt are not real crashes. Furthermore, the core dump I got is unrelated to plugins or plugin process, it's about rendering. After all, I am a bit puzzled about how should I reproduce / fix the problem.
And it is P1/Critical, because they are crashes.
Just a reminder: These tests are still crashing on Qt-WK2 because of your _two weeks_ old buggy patch. It is the only changeset why the Qt-WK2 bots aren't green. Are you going to fix this bug or do you think if crashing tests are absolutely acceptable?
(In reply to comment #3) > Just a reminder: These tests are still crashing on Qt-WK2 because of your _two weeks_ old buggy patch. It is the only changeset why the Qt-WK2 bots aren't green. > > Are you going to fix this bug or do you think if crashing tests are absolutely acceptable? In fact you have no clue about what is crashing and why. Neither I, unfortunately. Please, rollout if you think it is necessary and it will solve the problem.
Rollout landed in http://trac.webkit.org/changeset/115293
(In reply to comment #5) > Rollout landed in http://trac.webkit.org/changeset/115293 I think the crashes disappeared because you re-unskipped the plugin tests with this rollout. The other changes made in the original patch - which was a pro file fix to enable building of the testnetscapeplugin and some build fixes to make it buildable - should be harmless. I am going roll them back so at some point we can examine the problem further with a working test environment, if you are not against it. Are you? :)
Not the skipping tests fixed the crashes. It was the first thing two weeks ago I tried. But after skipping tests nothing changed and tests still crashed. So please don't reland the change until you find what the problem is.
*** This bug has been marked as a duplicate of bug 93272 ***