366128 - Crash with float and xul in <select> [@ nsFrameManager::CaptureFrameState] [@ nsFrameManager::CaptureFrameState]

Status: RESOLVED FIXED

(Core :: Layout: Floats, defect)

Description

[Jesse Ruderman]

Attachment: testcase (crashes on load or reload)

Loading this testcase in a debug build causes an assertion and a scary crash:

###!!! ASSERTION: Float frame has wrong parent: 'floatFrame->GetParent() == mBlock', file /Users/admin/trunk/mozilla/layout/generic/nsBlockReflowState.cpp, line 733

EXC_BAD_ACCESS (0x0001)
KERN_INVALID_ADDRESS (0x0001) at 0xddddddfd

Thread 0 Crashed:
0    nsIFrame::GetNextSibling() const + 20 (nsIFrame.h:738)
1    nsBlockFrame::CheckFloats(nsBlockReflowState&) + 548 (nsBlockFrame.cpp:6287)
2    nsBlockFrame::Reflow(nsPresContext*, nsHTMLReflowMetrics&, nsHTMLReflowState const&, unsigned&) + 2376 (nsBlockFrame.cpp:997)
3    nsSelectsAreaFrame::Reflow(nsPresContext*, nsHTMLReflowMetrics&, nsHTMLReflowState const&, unsigned&) + 248 (nsSelectsAreaFrame.cpp:260)

It also crash opt builds, but with a different signature, and only on reload.  The opt crashes are sometimes [@  nsFrameManager::CaptureFrameState] and sometimes somewhere else.

Comment 1

[Mats Palmgren (inactive)]

This is essentially the same underlying cause as bug 322436 - a float as
child of a XUL display type frame.

Comment 2

[Mats Palmgren (inactive)]

Attachment: wallpaper

Here's a wallpaper that veto creating a float frame as a child of a XUL
frame during frame construction.
This will of course trigger other assertions, eg:

###!!! ASSERTION: Disagreement about whether it's a block or not: 'fromLine->IsBlock() == fromLine->mFirstChild->GetStyleDisplay()->IsBlockLevel()', file nsBlockFrame.cpp, line 2267
###!!! ASSERTION: How'd we get a floated inline frame? The frame ctor should've dealt with this.: 'aReflowState.mStyleDisplay->mFloats == NS_STYLE_FLOAT_NONE', file nsLineLayout.cpp, line 1037

but I think those are probably less harmful to have than this bug.
The patch fixes the crash and makes the text appear in the second
testcase in bug 322436.

Comment 3

[Daniel Veditz [:dveditz]]

In ff1.5.0.9 and ff2.0.0.1 debug I get the "Float frame has wrong parent" assertion, but I don't see a crash in debug or opt builds. Should we take the wallpaper patch anyway just in case, or are we better leaving it alone since there's no crash?

Comment 4

[Jay Patel [:jay]]

Not blocking, but assigning to mats, let's see if we can figure out what the underlying problem is on the branches.  It'll be nice to take the patch if we think we need it.

Comment 5

[chris hofmann]

is the wall paper patch ready to go in on the trunk?

Comment 6

[chris hofmann]

mats, who would be a good reviewer?

Comment 7

[Mats Palmgren (inactive)]

Attachment: wallpaper

Updated to trunk

Comment 8

[Mats Palmgren (inactive)]

(In reply to comment #6)
> is the wall paper patch ready to go in on the trunk?

Yes, I believe so. 

> mats, who would be a good reviewer?

roc, bzbarsky or dbaron I guess.  I don't know if they have plans to fix
layout of floats in XUL in the 1.9 time frame... if not, then we probably
want something like this patch...

Comment 9

[Robert O'Callahan (:roc) (email my personal email if necessary)]

I can review it, I guess. But what is this doing to floats in XUL? Treating them as normal in-flow content? I suppose that's probably best...

Comment 10

[Boris Zbarsky [:bzbarsky]]

Isn't this equivalent to pushing a null float containing block before processing the kids of frames with these XUL display types, in the cases it covers?  Note that if I have an inline kid of a XUL frame we want to suppress floating of kids of the inline as well, which I don't think this patch does....

Comment 11

[Window Snyder]

mats can you write the patch that Boris described?

Comment 12

[Jesse Ruderman]

roc attached a patch to bug 322436.

Comment 13

[Jesse Ruderman]

roc, this still crashes for me, with

###!!! ASSERTION: not in child list: 'found', file /Users/jruderman/trunk/mozilla/layout/base/nsCSSFrameConstructor.cpp, line 1774

Comment 14

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Attachment: fix

Followup fix. We should call PushFloatContainingBlock(newFrame) if newFrame is a real block.

Comment 15

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 267536 [details] [diff] [review]
fix

Looks reasonable.

Comment 16

[Jesse Ruderman]

Better, but now I get

###!!! ASSERTION: Please push a real float containing block!: '!aNewFloatContainingBlock || aNewFloatContainingBlock->GetContentInsertionFrame()-> IsFloatContainingBlock()', file /Users/jruderman/trunk/mozilla/layout/base/nsCSSFrameConstructor.cpp, line 1285

every time I open a Firefox browser window.

Comment 17

[Jesse Ruderman]

Should the comment above the modified code be updated too?

Comment 18

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Attachment: updated patch

Sorry Boris ... this is better. XUL IFRAMEs and other subdocuments should not be pushed as float containing blocks.

Comment 19

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 268168 [details] [diff] [review]
updated patch

I'd reverse the condition of the ?: and the results; having the '!' there just seems to confuse things unnecessarily.

That said, why were we ever constructing any kids under the subdocument?  That shouldn't be happening, given that they're leaves, no?

Comment 20

[Boris Zbarsky [:bzbarsky]]

Oh, we're just avoiding an assert.  Nevermind.  ;)

Comment 21

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Checked in with comment addressed.

Comment 22

[Robert Strong (they/them - no direct email)]

I don't crash on the 1.8 branch but I do get the following assertion
###!!! ASSERTION: Float frame has wrong parent: 'floatFrame->GetParent() == mBlock', file c:/moz/mozilla_1_8_branch/mozilla/layout/generic/nsBlockReflowState.cpp, line 847

Jesse, is this assertion important enough to get fixed on the branch?

Comment 23

[Jesse Ruderman]

bz or roc would know better than me whether this fix is needed on branch.

Comment 24

[Robert O'Callahan (:roc) (email my personal email if necessary)]

I think this should be fixed on branch. A bad frame tree is very risky.

Comment 25

[Daniel Veditz [:dveditz]]

Comment on attachment 268168 [details] [diff] [review]
updated patch

approved for 1.8.1.5 and 1.8.0.13, a=dveditz for release-drivers

Comment 26

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Can we push this to 1.8.1.6/1.8.0.14? I'm not comfortable pushing this into the branch in a rush.

Comment 27

[Mats Palmgren (inactive)]

This was actually checked in earlier today to fix bug 384344, which is a merge
of this fix and bug 322436.  If you're not comfortable with it we should
back that out.  Sorry for the confusion, I should have let it be handled
by the respective bugs instead of course... my bad.

Comment 28

[Robert O'Callahan (:roc) (email my personal email if necessary)]

No, I guess that's OK. At least I'm not on the hook :-)

Comment 29

[Carsten Book [:Tomcat]]

verified fixed 1.8.1.5 using  Mozilla/5.0 (Macintosh; U; Intel Mac OS X; en-US; rv:1.8.1.5) Gecko/2007071317 Firefox/2.0.0.5 and on Windows XP - no crash on testcase - adding verified keyword

Comment 30

[juan becerra [:juanb]]

Verified on MacIntel using Thunderbird version 1.5.0.13 (20070809). Used Thunderbrowse extension to load testcase. No crash. 

Comment 31

[Daniel Veditz [:dveditz]]

Comment on attachment 268168 [details] [diff] [review]
updated patch

clearing approval 1.8.1.7 request, this ended up landing in 1.8.1.5 after all (see above comments)

Comment 32

[Jesse Ruderman]

Crashtest checked in.