Assertion failure: !mFrame->IsAbsolutelyPositioned() (abspos child frames should not be treated as flex items), at /Users/jruderman/trees/mozilla-central/layout/generic/nsFlexContainerFrame.cpp:1057 This assertion was added in: changeset: http://hg.mozilla.org/mozilla-central/rev/da32dcdc4121 user: Daniel Holbert date: Tue Jan 21 14:52:32 2014 -0800 summary: Bug 783470 part 5: Add assertion to ensure we don't create FlexItems for abspos children. rs=mats

A similar testcase trips: Assertion failure: aFrame1->IsFlexItem() && aFrame2->IsFlexItem() (this method only intended for comparing flex items), at layout/generic/nsFlexContainerFrame.cpp:809

Yeah, IsFlexItem() checks IsAbsolutelyPositioned(), so comment 2 boils down to the same underlying problem -- IsAbsolutelyPositioned() is lying. nsIFrame::IsAbsolutelyPositioned() just checks the "position" property (along with !IsSVGText()), and that's not sufficient to say whether the frame is *actually* being absolutely positioned. MathML frames are constructed in nsCSSFrameConstructor with the FCDATA_DISALLOW_OUT_OF_FLOW flag, which seems to make "position:absolute" have no actual effect (aside from confusing IsAbsolutelyPositioned). So, I need to find a way to tell if the frames are *actually* absolutely positioned. (Maybe check their NS_FRAME_OUT_OF_FLOW bit, or something like that.)

I think this should do it. This makes us use the frame-state bit instead of IsAbsolutelyPositioned() in the general "IsFlexItem" function, and check the bit (as well as for nsPlaceholderFrames, for good measure) in the FlexItem constructor assertions. I'm going to give this a Try run, and if it does well there, I'll request review.

Try run: https://tbpl.mozilla.org/?tree=Try&rev=f7588abcb58a (BTW, the patch includes the attached testcase, along with a second modified version with two <munderover> elements instead of one. In unpatched builds, that second testcase makes us trigger the assertion that Jesse mentioned from comment 2.)

Comment on attachment 8382749 [details] [diff] [review] fix v1 Try is mostly-done and looks good. Requesting review.

Comment on attachment 8382749 [details] [diff] [review] fix v1 r=mats