after implementation of the reflow branch to trunk the rendering of http://www.vobis.com/shop.html seems to have regressed (see screenshot #1). the site is valid, but has some css errors. it works with Fx2 and Fx3-pre-a1 builds prior to reflow-branch-merging (like firefox-3.0a1.en-US.win32-2006-12-06-07).

thanks to polidobj for providing the testcase

screenshot of rendering the testcase with Fx3, latest WebKit, Opera 9.5alpha (thanks to phiw13)

Note: without 'overflow:hidden' on the outside, wrapping div, the test case renders the same in Gecko 1.9 and WebKit.

I find it odd that the overflow style has an effect, given that the div should be shrink-wrapping anyway. I'd think the pref width (which includes the margins) would just be that 305px width even with overflow:visible....

This seems like something we should fix, and it shouldn't be that hard. Who wants it?

So I guess what happens here is that nsIFrame::InlinePrefWidthData::ForceBreak accumulates the widths of all floats and then adds the result into the line width. So the large negative margin on the purple float "pulls back" the pref-width of the line to something fairly small, ensuring that the red float gets cut off. IMHO it would make more sense and be more compatible (looking at that testcase rendering, thanks!) to just ignore floats that don't change the available space.

This fixes this testcase, making our rendering compatible with MSIE8 and Opera. And I think it makes sense. Running reftests now.

Reftests pass.

--> P2, this doesn't block the beta

Mike, this could have some web-compat fallout, so if we _can_ take it for the beta, we really should...

I think we should get this into beta5 if we possibly can

I'll make this P1 again in the hope someone notices the above comments :-)

We've noticed: waiting on dbaron's review ...

Comment on attachment 311118 [details] [diff] [review] fix? r+sr=dbaron

Comment on attachment 311300 [details] [diff] [review] fix with reftest a1.9b5=beltzner based on dbaron's review of above

Checking in layout/generic/nsFrame.cpp; /cvsroot/mozilla/layout/generic/nsFrame.cpp,v <-- nsFrame.cpp new revision: 3.791; previous revision: 3.790 done RCS file: /cvsroot/mozilla/layout/reftests/bugs/423385-1-ref.html,v done Checking in layout/reftests/bugs/423385-1-ref.html; /cvsroot/mozilla/layout/reftests/bugs/423385-1-ref.html,v <-- 423385-1-ref.html initial revision: 1.1 done RCS file: /cvsroot/mozilla/layout/reftests/bugs/423385-1.html,v done Checking in layout/reftests/bugs/423385-1.html; /cvsroot/mozilla/layout/reftests/bugs/423385-1.html,v <-- 423385-1.html initial revision: 1.1 done Checking in layout/reftests/bugs/reftest.list; /cvsroot/mozilla/layout/reftests/bugs/reftest.list,v <-- reftest.list new revision: 1.410; previous revision: 1.409 done

REFTEST UNEXPECTED FAIL: file:///builds/slave/trunk_centos5/mozilla/layout/reftests/pixel-rounding/border-height-5.html REFTEST UNEXPECTED FAIL: file:///builds/slave/trunk_centos5/mozilla/layout/reftests/pixel-rounding/border-height-6.html REFTEST UNEXPECTED FAIL: file:///builds/slave/trunk_centos5/mozilla/layout/reftests/pixel-rounding/border-width-5.html REFTEST UNEXPECTED FAIL: file:///builds/slave/trunk_centos5/mozilla/layout/reftests/pixel-rounding/border-width-6.html Failed on Linux and Win32... seems to pass on Mac, though. Backed out patch and the reftest.list addition.

Broken reftests? Bad patch? At this point I'd like someone to reiterate the *need* to have this in the beta, as we're blocking on it, and we shipped 4 betas without it and didn't even relnote or get a lot of dupes.

(In reply to comment #19) > REFTEST UNEXPECTED FAIL: > file:///builds/slave/trunk_centos5/mozilla/layout/reftests/pixel-rounding/border-height-5.html > REFTEST UNEXPECTED FAIL: > file:///builds/slave/trunk_centos5/mozilla/layout/reftests/pixel-rounding/border-height-6.html > REFTEST UNEXPECTED FAIL: > file:///builds/slave/trunk_centos5/mozilla/layout/reftests/pixel-rounding/border-width-5.html > REFTEST UNEXPECTED FAIL: > file:///builds/slave/trunk_centos5/mozilla/layout/reftests/pixel-rounding/border-width-6.html > > Failed on Linux and Win32... seems to pass on Mac, though. Backed out patch and > the reftest.list addition. > Aren't these due to bug 417178?

Relanded. Checking in layout/generic/nsFrame.cpp; /cvsroot/mozilla/layout/generic/nsFrame.cpp,v <-- nsFrame.cpp new revision: 3.793; previous revision: 3.792 done Checking in layout/reftests/bugs/reftest.list; /cvsroot/mozilla/layout/reftests/bugs/reftest.list,v <-- reftest.list new revision: 1.414; previous revision: 1.413 done

with Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9b5pre) Gecko/2008032605 Firefox/3.0.0.0 ID:2008032605 testcase is rendered like MSIE8 beta1/Opera 9.5 build 9841 and http://www.vobis.com/shop.html looks fine too => VERIFIED FIXED roc = my hero :D