Created attachment 317414 [details]
testcase (crashes Firefox when closed)
Loading the testcase triggers:
###!!! ASSERTION: reflow roots should never split: 'status == NS_FRAME_COMPLETE', file /Users/jruderman/trunk/mozilla/layout/base/nsPresShell.cpp, line 6296
Closing the window triggers:
Crash [@ nsPropertyTable::PropertyList::Equals] dereferencing 0xddddde09.
Related to bug 399994?
Crashes for me using yesterday's linux nightly, when I close the testcase's tab.
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.1b2pre) Gecko/20081028 Minefield/3.1b2pre
OS --> All; Assignee --> me
Created attachment 345372 [details]
testcase 2 (crashes Firefox when closed)
This testcase is tweaked a bit.
- uses <br> instead of newline+"whitespace: pre" to trigger a break
- uses -moz-column-count:1 instead of -moz-column-width to trigger moz-col usage
- Adds backgrounds & a border to show what's going on
- Improved readability of style attribute, using newlines.
Like the previous one, this testcase crashes when I close or reload the tab.
FWIW, I also still get the assertion mentioned in comment 0, using both testcases.
###!!! ASSERTION: reflow roots should never split: 'status == NS_FRAME_COMPLETE', file /mozilla/layout/base/nsPresShell.cpp, line 6338
Created attachment 345746 [details]
backtrace of crash on testcase 2
Here's a backtrace of the crash triggered by testcase 2.
A few annotations:
- At stack level 8, aPropertyName is the atom for "ExcessOverflowContainersProperty".
- At stack levels 2-3, aPropertyName is the atom for "OverflowProperty". (This is passed in from level 4, which is GetOverflowFrames())
Basically, it looks like we're crashing because we're partway through deleting all the propertyLists in the propertyTable (stack level 12), and then we try to iterate across those same propertyLists and call methods on them ("Equals" in this case) (at stack level 1).
Shouldn't we be deleting frame properties when the frame is destroyed? Or do we suppress that during pres shell destruction?
We do, yes.
(In reply to comment #0)
> ###!!! ASSERTION: reflow roots should never split: 'status ==
> NS_FRAME_COMPLETE', file
> /Users/jruderman/trunk/mozilla/layout/base/nsPresShell.cpp, line 6296
I think this assertion is the same problem described in bug 468771 comment 8. Not sure if the crash is, though, although it seems likely.
Both testcase are now WFM in my mozilla-central debug build and nightly build.
("WFM" = No crash & no assertions at load, reload, or tab-closing)
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.2a1pre) Gecko/20090209 Minefield/3.2a1pre
My debug build is at changeset 5f349409c9d5 (up-to-date as of this morning).
This isn't yet fixed in 191, though. I just tested today's 191 nightly, and I got a crash:
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.1b3pre) Gecko/20090209 Shiretoko/3.1b3pre
Fix range for mozilla-central:
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.2a1pre) Gecko/20090204 Minefield/3.2a1pre
Built from http://hg.mozilla.org/mozilla-central/rev/4ada519b5c97
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.2a1pre) Gecko/20090205 Minefield/3.2a1pre
Built from http://hg.mozilla.org/mozilla-central/rev/5de9f1e51c68
Maybe bug 363247?
Bug 399994 is still crashing.
Yes, confirmed by local backout that it was fixed by bug 363247.
Great, thanks dbaron! And it looks like you've already requested approval1.9.1 on that bug's patch, so hopefully it'll be approved and able to land there soon.
I landed bug 363247 on 1.9.1, so marking this fixed1.9.1.
Verified fixed on OS X and Windows with:
Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9.2a1pre) Gecko/20090310 Minefield/3.2a1pre ID:20090310044308
Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9.1b4pre) Gecko/20090309 Shiretoko/3.1b4pre (.NET CLR 3.5.30729) ID:20090309034003
So this is blocking22.214.171.124+, but I can't reproduce it on my 1.9.0 build. Can anybody else?
dbaron: I can reproduce this using testcase 2 with Firefox 3.0.13 on Mac.
I get bp-082f9466-f9f5-4f9e-a608-a88b02090809.
I can't even get the testcase to emit a valgrind warning with DEBUG_TRACEMALLOC_FRAMEARENA. (And no assertion, either.)
Might be opt only? I definitely see it.
Verified fixed in Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:126.96.36.199pre) Gecko/2009081305 GranParadiso/3.0.14pre (.NET CLR 3.5.30729) for 188.8.131.52. I saw the crashes with the test cases here in 184.108.40.206.
Crashtest: Crashtest: http://hg.mozilla.org/mozilla-central/rev/4da43cad0331