322436 - WARNING: nsBlockFrame::CheckFloats: Explicit float list is out of sync with float cache

Status: RESOLVED FIXED

Description

[Jesse Ruderman]

Steps to reproduce:
1. Load the testcase in a trunk debug build.

Result:

###!!! ASSERTION: not in child list: 'nsFrameList(aChildFrame->GetParent()->GetFirstChild(listName)) .ContainsFrame(aChildFrame)', file /Users/admin/trunk/mozilla/layout/base/nsCSSFrameConstructor.cpp, line 1888

Comment 1

[Jesse Ruderman]

Created attachment 207582 [details]
testcase

Comment 2

[Jesse Ruderman]

bz, here's one of the testcases you asked for in bug 319280 comment 13.

Comment 3

[Boris Zbarsky [:bz]]

Created attachment 207583 [details]
What I think should be an even smaller testcase

This testcase shows what I think is the underlying problem -- the float being dropped from the frame tree altogether.  I'll bet money the reason is that BuildFloatList doesn't walk into boxes and that the answer is to nuke BuildFloatList.

Comment 4

[Jesse Ruderman]

I don't see the assertion any more (after BuildFloatList removal), but I do see

WARNING: nsBlockFrame::CheckFloats: Explicit float list is out of sync with float cache: file /Users/admin/trunk/mozilla/layout/generic/nsBlockFrame.cpp, line 7068

and the second testcase's "This text should show up" text still doesn't show up.

Comment 5

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

Yeah, we're still not reflowing the float. However, I believe we have changed the error from something potentially catastrophic to something merely incorrect.

Comment 6

[Jesse Ruderman]

That's good to know.

Each warning is accompanied by a screenful of frame dumps (?), so if I see the warning often while fuzzing, I'll ask you to make at least the frame dump be DEBUG_roc or something.

Comment 7

[Mats Palmgren (vacation - back in August)]

Created attachment 224570 [details] [diff] [review]
Patch to mute the frame dumps

(In reply to comment #6)
> I'll ask you to make at least the frame dump be DEBUG_roc or something.

I'll second that request.

Comment 8

[Mats Palmgren (vacation - back in August)]

Comment on attachment 224570 [details] [diff] [review]
Patch to mute the frame dumps

Landed on trunk.

Comment 9

[Mats Palmgren (vacation - back in August)]

A float inside a Box where the float containing block is an ancestor
of the Box is going to trigger this warning.
So, what is the real fix to this bug? do we want to add the float
of the nearest block reflow state? or wrap it in a anonymous block
during frame construction? or something else?

In the meantime, we could make CheckFloats() be silent for this
particular case, since it's a known problem (allowing us to see any
remaining issues with the float cache).

Comment 10

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

You mean a XUL box?

Comment 11

[Mats Palmgren (vacation - back in August)]

Yes - anything that uses nsFrame::BoxReflow()
http://bonsai.mozilla.org/cvsblame.cgi?file=/mozilla/layout/generic/nsFrame.cpp&rev=3.667&root=/cvsroot&mark=5534#5532
It doesn't seem to add the float to the block reflow state (which eventually
adds it to the float cache for the line), like nsLineLayout::ReflowFrame does:
http://bonsai.mozilla.org/cvsblame.cgi?file=/mozilla/layout/generic/nsLineLayout.cpp&rev=3.238&root=/cvsroot&mark=1013-1018#992

There is also this block:
http://bonsai.mozilla.org/cvsblame.cgi?file=/mozilla/layout/generic/nsFrame.cpp&rev=3.667&root=/cvsroot&mark=5657-5659#5648
which made me think it could be intentional.

Comment 12

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

I don't know how that should be handled. Perhaps we should just ignore the float property on elements that aren't in a block formatting context?

Comment 13

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

A fix for this could be backported to the 1.0.x branch, which might be useful. I'll try to work on it.

Comment 14

[Jesse Ruderman]

Nominating because this blocks an [sg:critical] crash bug.

Comment 15

[chris hofmann]

Is this ready for check-in on the trunk?  the problem blocks some of jesse's testing.

also do we know if it really does fix bug 366128, or will more work be needed for that one.

Comment 16

[Jesse Ruderman]

The patch in this bug was just to reduce the amount of console output when this bug occurs.  There isn't a patch for this bug yet afaik.

Comment 17

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

This is basically the same as bug 366128. We should fix it by pushing a null float containing block for XUL box frames, as Boris suggests:
https://bugzilla.mozilla.org/show_bug.cgi?id=366128#c10

Comment 18

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

Created attachment 266862 [details] [diff] [review]
fix

It should be this simple.

Comment 19

[Boris Zbarsky [:bz]]

Comment on attachment 266862 [details] [diff] [review]
fix

Would that it were so.  GetFloatContainingBlock also needs to be changed.

Comment 20

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

Created attachment 266892 [details] [diff] [review]
fix v2

Indeed.

Comment 21

[Boris Zbarsky [:bz]]

Comment on attachment 266892 [details] [diff] [review]
fix v2

I assume you checked that IsBoxFrame() is true if and only if the frame is created by ConstructXULFrame, right?  If so, r+sr=bzbarsky.

Comment 22

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

There was one issue --- ConstructXULFrame creates non-box blocks for <label>, <text> and <description>. So I added an IsBoxFrame check here:

+    nsFrameConstructorSaveState floatSaveState;
+    if (newFrame->IsBoxFrame()) {
+      aState.PushFloatContainingBlock(nsnull, floatSaveState, PR_FALSE,
+                                      PR_FALSE);
+    }

and checked it in.

Comment 23

[Daniel Veditz [:dveditz] PTO until August]

Mats incorporated this into a branch fix for bug 384344

Comment 24

[Jesse Ruderman]

I committed my testcase as a crashtest.

I also committed bz's testcase as a reftest.  I assumed the current rendering is correct.