359132 - ASSERTION: Some frame destructors were not called when closing Print Preview

Status: RESOLVED WORKSFORME

(Core :: Layout, defect)

Description

[Adam Guthrie]

Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9a1) Gecko/20061030 Minefield/3.0a1

Steps to reproduce:
1. Load the testcase.
2. File > Print Preview
3. Close Print Preview.

Results:
The following assertion appears in the console:
###!!! ASSERTION: Some frame destructors were not called: 'mFrameCount == 0', file /moz/trunk/mozilla/layout/base/nsPresShell.cpp, line 625

This is an unminimized testcase. It's sort of hard to minimize since it involves print previewing each time. You're welcome to try and reduce it further.

Comment 1

[Adam Guthrie]

Attachment: unminimized testcase

Comment 2

[Boris Zbarsky [:bzbarsky]]

Really need a minimal-ish testcase here if humanly possible.  Debugging this assertion is a PITA, since it fires _long_ after the problem really happened...  So the only way to debug is to step through frame construction line by line.  That's hundreds of lines of code per DOM node.

Comment 3

[Jesse Ruderman]

We can use Lithium to reduce the testcase if we can add some JavaScript to it that triggers Print Preview.  Who would know how to do that?

Comment 4

[Bernd]

I will try whether the trick from https://bugzilla.mozilla.org/attachment.cgi?id=236723 works for this.

Comment 5

[Boris Zbarsky [:bzbarsky]]

So the relevant assert is probably:

###!!! ASSERTION: bad geometric parent: 'IsChild(aNextInFlow)', file ../../../mozilla/layout/generic/nsBlockFrame.cpp, line 6090

with the stack:

#0  0xb638742e in nsBlockFrame::DeleteNextInFlowChild (this=0x8b15280, 
    aPresContext=0x8a95a58, aNextInFlow=0x8b29f1c)
    at ../../../mozilla/layout/generic/nsBlockFrame.cpp:6090
#1  0xb63877ce in nsBlockFrame::ReflowFloat (this=0x8b15280, aState=@0xbfff57dc, 
    aPlaceholder=0x8b17750, aFloatCache=0x8b28548, aReflowStatus=@0xbfff50dc)
    at ../../../mozilla/layout/generic/nsBlockFrame.cpp:6211
#2  0xb6390539 in nsBlockReflowState::FlowAndPlaceFloat (this=0xbfff57dc, 
    aFloatCache=0x8b28548, aIsLeftFloat=0xbfff4e48, aReflowStatus=@0xbfff50dc, 
    aForceFit=0) at ../../../mozilla/layout/generic/nsBlockReflowState.cpp:855
#3  0xb638ff9a in nsBlockReflowState::AddFloat (this=0xbfff57dc, 
    aLineLayout=@0xbfff51dc, aPlaceholder=0x8b17750, aInitialReflow=0, 
    aReflowStatus=@0xbfff50dc)
    at ../../../mozilla/layout/generic/nsBlockReflowState.cpp:630
#4  0xb63e7bac in nsLineLayout::AddFloat (this=0xbfff51dc, aFrame=0x8b17750, 
    aReflowStatus=@0xbfff50dc) at ../../../mozilla/layout/generic/nsLineLayout.h:257
#5  0xb63e416d in nsLineLayout::ReflowFrame (this=0xbfff51dc, aFrame=0x8b17750, 
    aReflowStatus=@0xbfff50dc, aMetrics=0x0, aPushedFrame=@0xbfff50d8)
    at ../../../mozilla/layout/generic/nsLineLayout.cpp:1004

Here the aNextInFlow is a placeholder, whose parent is the block that's |this| in frame 1.  Said |this| has no next in flow.

Interestingly, I can only reproduce this in a thebes build...

Probably still need a minimalish testcase to debug usefully.

Comment 6

[Bob Clary [:bc] (inactive)]

My first stop would be:
<http://developer.mozilla.org/en/docs/Mozilla_Embedding_FAQ:How_do_I...#How_do_I_print_preview_a_page.3F>
followed quickly by 
<http://lxr.mozilla.org/mozilla/source/embedding/browser/webBrowser/nsIWebBrowserPrint.idl> which _is_ scriptable. 

Comment 7

[Martijn Wargers (dead)]

Attachment: smaller, but not minimised testcase

Comment 8

[Boris Zbarsky [:bzbarsky]]

I don't see any asserts on Linux with the smaller testcase.  :(

Comment 9

[Martijn Wargers (dead)]

Comment on attachment 244434 [details]
smaller, but not minimised testcase

(In reply to comment #8)
> I don't see any asserts on Linux with the smaller testcase.  :(

Yeah, I don't see it either with the smaller testcase.
I guess I optimised it wrongly, so I should try again.

Comment 10

[Martijn Wargers (dead)]

Attachment: unminimised testcase2

Comment 11

[Martijn Wargers (dead)]

Attachment: Unminimised testcase3

I'm attaching these unminimised testcases to prevent my first testcase disaster.

Comment 12

[Martijn Wargers (dead)]

Attachment: minimised testcase

Ok, I finally managed to get a minimised testcase, that still asserts when closing print preview.

Comment 13

[Martijn Wargers (dead)]

With a slightly more minimised testcase, I get a different assertion when opening print preview, I filed bug 362084 for that.

Comment 14

[Boris Zbarsky [:bzbarsky]]

So... I can't seem to reproduce the asserts on linux with thebes builds now... either on the original testcase or the "minimised testcase".

Comment 15

[Martijn Wargers (dead)]

Comment on attachment 246782 [details]
minimised testcase

Oh, now I can't reproduce it with the minimised testcase.
(but I still can with Unminimised testcase3)

Comment 16

[Jesse Ruderman]

Martijn, do you still see this?  (I can't test on Mac because print preview goes through the normal print path on Mac.)

Comment 17

[Martijn Wargers (dead)]

No, I don't see this assertion anymore with a freshly build debug build.
Marking worksforme.