372376 - Memory corruption with <tr style="display: inherit">

Status: VERIFIED FIXED

(Core :: Layout: Tables, defect)

Description

[Jesse Ruderman]

Attachment: testcase (causes memory corruption / crashes)

This testcase leads to crashes in *entirely* random places and times.  It might crash soon during GC, or it might crash while writing bookmarks during shutdown, or it might not crash at all.  You should restart Firefox after loading the testcase to avoid data loss in a crash later.

Memory corruption leading to random crashes, including "Illegal instruction" crashes --> sg:critical.

You're also likely to see these two assertions:

###!!! ASSERTION: wrong display type on rowframe: 'NS_STYLE_DISPLAY_TABLE_ROW == childFrame->GetStyleDisplay()->mDisplay', file /Users/jruderman/trunk/mozilla/layout/tables/nsTableRowFrame.cpp, line 1291

###!!! ASSERTION: invalid array index: 'i < Length()', file ../../dist/include/xpcom/nsTArray.h, line 318

and you might also see:

###!!! ASSERTION: span beyond the row size!: 'endColIndex + 1 <= row.Length()', file /Users/jruderman/trunk/mozilla/layout/tables/nsCellMap.cpp, line 2108

###!!! ASSERTION: Bogus row index?: 'Not Reached', file /Users/jruderman/trunk/mozilla/layout/tables/nsCellMap.cpp, line 399

Comment 1

[Jesse Ruderman]

Does the tr's display property have to be "inherit" to trigger this crash?  The computed style is "block" (why not "table"?), but if I change "inherit" to "block" in the testcase the assertions and crash and incorrect display go away.

Comment 2

[Martijn Wargers (dead)]

I think this might be related to bug 317876.

Comment 3

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

Attachment: valgrind warnings

Comment 4

[Bernd]

###!!! ASSERTION: wrong display type on rowframe: 'NS_STYLE_DISPLAY_TABLE_ROW
== childFrame->GetStyleDisplay()->mDisplay', file
/Users/jruderman/trunk/mozilla/layout/tables/nsTableRowFrame.cpp, line 1291

Is exactly bug 317876, the cure is outlined in https://bugzilla.mozilla.org/show_bug.cgi?id=323656#c5 which is beyond my capabilities.

Comment 5

[Window Snyder]

Critical security bugs need to have an owner.  Bern, DBarron, who should own this bug?

Comment 6

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

Whoever owns bug 323656.

Comment 7

[Window Snyder]

That one is also assigned to nobody.  Any ideas who should get them?

Comment 8

[Jesse Ruderman]

bz attached a patch last week and it already has review.  It sounds like he needs more time to test the patch before he's comfortable checking it in.

Comment 9

[Boris Zbarsky [:bzbarsky]]

Note to self:  The PresContext() landing bitrotted this.  Need to update...  The patch lives in the "leafframe" tree.

Comment 10

[Boris Zbarsky [:bzbarsky]]

Now that bug bug 323656 is fixed, is this still a problem?

Comment 11

[Jesse Ruderman]

A debug build from earlier today had a bunch of problems (random rendering, assertions, random crashes) with the testcase.  A debug build from just now does fine.  FIXED by the checkin for bug 323656.

Comment 12

[Daniel Veditz [:dveditz]]

The testcase crashes debug 1.5.0.12pre/2.0.0.4pre builds but with a different assertion (twice):

###!!! ASSERTION: Computed overflow area must contain frame bounds: 'aNewSize.width == 0 || aNewSize.height == 0 || aOverflowArea->Contains(nsRect(nsPoint(0, 0), aNewSize))', file c:/dev/ff2/mozilla/layout/generic/nsFrame.cpp, line 4410

The crash is due to the debug heap routines detecting damaged blocks (i.e. most likely sg:critical). Applying the patch from bug 323656 to see if it's the same thing as the trunk crash.

Comment 13

[Daniel Veditz [:dveditz]]

The patch from bug 323656 seems to fix the branch crash, too.

Comment 14

[Daniel Veditz [:dveditz]]

Appears fixed in 1.8.1.4pre by bug 323656. Adding keywords but also "qawanted" to make sure we double-check this one.

Comment 15

[Martijn Wargers (dead)]

Verified fixed on trunk and branches.
The testcase doesn't crash on load, or even reloading it a few times.

Comment 16

[Jesse Ruderman]

Crashtest checked in.