I thought we killed this snake long ago, but the upcoming testcase shows that the bounds of a relative span's absolute children aren't included in its overflow area.

This isn't quite minimal, but it's good enough to show the problem. The DIV "I am absolutely positioned." won't repaint properly if you dirty it. A moment's work with the layout debugger confirms that the relatively positioned span doesn't have the right combined area.

2 problems: -- nsLineLayout::RelativePositionFrames just ignored a span frame's pfd->mCombinedArea, wiping it out with a new combined area that's the union of its child inline frame combined areas. Solution: initialize the accumulated area with the span frame combined area instead of the span frame's bounds. -- nsPositionedInlineFrame was never computing the overflow area properly because someone noticed it was just being ignored and turned it off. Turn it back on. This fixes the bug. I took the liberty of cleaning up RelativePositionFrames a bit.

Comment on attachment 142399 [details] [diff] [review] fix How about getting rid of this declaration, > PerFrameData* pfd; using psd->mFrame->mCombinedArea directly here, > pfd = psd->mFrame; >+ combinedAreaResult = pfd->mCombinedArea; declaring pfd in the for-loop initializer, > for (pfd = psd->mFirstFrame; pfd; pfd = pfd->mNext) { and making this what VC++ will think is a shadowing variable? > if (psd->mFrame) { > pfd = psd->mFrame; More comments shortly...

Comment on attachment 142399 [details] [diff] [review] fix r+sr=dbaron if you've checked some testcases (e.g., bug 79315 and its fixed dependencies)

I'll fix up the pfds as you suggest, but I'll use a new name at the end to avoid shadowing.

Fix checked in. Testcases in 79315 and everything hanging off it work fine.

Actually marking fixed

I think the changes from this bug caused Bug 237734, backing out the checkin seems to make the testcase work every time.