Bug 207296 – Nullptr crash in RenderStyle::isFlippedBlocksWritingMode when fragment flow gains a new in-flow descendant

• Component: Layout and Rendering
• Status: RESOLVED FIXED

[Jack]

<rdar://56967115>

    #0 0x4b1c42ab1 in WebCore::RenderStyle::writingMode() const (Safari_ASAN_252152_93985f3693290f9e1b9273fb7bc493e9eb44c361.app/Contents/Frameworks/WebCore.framework/Versions/A/WebCore:x86_64+0x1c40ab1)
    #1 0x4b405e128 in WebCore::RenderStyle::isFlippedBlocksWritingMode() const (Safari_ASAN_252152_93985f3693290f9e1b9273fb7bc493e9eb44c361.app/Contents/Frameworks/WebCore.framework/Versions/A/WebCore:x86_64+0x405c128)
    #2 0x4b495d0fa in WebCore::RenderBox::flipForWritingMode(WebCore::LayoutPoint const&) const (Safari_ASAN_252152_93985f3693290f9e1b9273fb7bc493e9eb44c361.app/Contents/Frameworks/WebCore.framework/Versions/A/WebCore:x86_64+0x495b0fa)
    #3 0x4b4bd811f in WebCore::RenderMultiColumnFlow::physicalTranslationFromFragmentToFlow(WebCore::RenderMultiColumnSet const*, WebCore::LayoutPoint const&) const (Safari_ASAN_252152_93985f3693290f9e1b9273fb7bc493e9eb44c361.app/Contents/Frameworks/WebCore.framework/Versions/A/WebCore:x86_64+0x4bd611f)
    #4 0x4b4bd7df6 in WebCore::RenderMultiColumnFlow::mapAbsoluteToLocalPoint(unsigned int, WebCore::TransformState&) const (Safari_ASAN_252152_93985f3693290f9e1b9273fb7bc493e9eb44c361.app/Contents/Frameworks/WebCore.framework/Versions/A/WebCore:x86_64+0x4bd5df6)
    #5 0x4b49c5521 in WebCore::RenderBoxModelObject::mapAbsoluteToLocalPoint(unsigned int, WebCore::TransformState&) const (Safari_ASAN_252152_93985f3693290f9e1b9273fb7bc493e9eb44c361.app/Contents/Frameworks/WebCore.framework/Versions/A/WebCore:x86_64+0x49c3521)
    #6 0x4b49c5521 in WebCore::RenderBoxModelObject::mapAbsoluteToLocalPoint(unsigned int, WebCore::TransformState&) const (Safari_ASAN_252152_93985f3693290f9e1b9273fb7bc493e9eb44c361.app/Contents/Frameworks/WebCore.framework/Versions/A/WebCore:x86_64+0x49c3521)
    #7 0x4b49c5521 in WebCore::RenderBoxModelObject::mapAbsoluteToLocalPoint(unsigned int, WebCore::TransformState&) const (Safari_ASAN_252152_93985f3693290f9e1b9273fb7bc493e9eb44c361.app/Contents/Frameworks/WebCore.framework/Versions/A/WebCore:x86_64+0x49c3521)
    #8 0x4b4bfda9a in WebCore::RenderObject::absoluteToLocal(WebCore::FloatPoint const&, unsigned int) const (Safari_ASAN_252152_93985f3693290f9e1b9273fb7bc493e9eb44c361.app/Contents/Frameworks/WebCore.framework/Versions/A/WebCore:x86_64+0x4bfba9a)
    #9 0x4b40670ea in WebCore::FrameView::convertFromContainingViewToRenderer(WebCore::RenderElement const*, WebCore::IntRect const&) const (Safari_ASAN_252152_93985f3693290f9e1b9273fb7bc493e9eb44c361.app/Contents/Frameworks/WebCore.framework/Versions/A/WebCore:x86_64+0x40650ea)
    #10 0x4b4067ee8 in WebCore::FrameView::convertFromContainingView(WebCore::IntRect const&) const (Safari_ASAN_252152_93985f3693290f9e1b9273fb7bc493e9eb44c361.app/Contents/

[Jack]

Root cause analysis from Alan:

Without the [outer.style.transition = "1s”] this is what we end up with: 

before [range.extractContents] call
B-----L- --        DIV RenderBlock at (0,0) size 1222x0 renderer->(0x1360f9420) node->(0x12f59afc0)
B---YGL-           RenderMultiColumnFlowThread at (0,0) size 603x0 renderer->(0x1360f97b0) [Rs:0x0 Re:0x0]
BA----L- --            DIV RenderBlock at (8,8) size 304x154 renderer->(0x1360f9550) node->(0x12f59b050)
N------- --                 IFRAME RenderIFrame at (0,0) size 304x154 renderer->(0x1360f9680) node->(0x12f59b0e0)

after [range.extractContents] call
B------- -+        DIV RenderBlock at (0,0) size 1222x0 renderer->(0x1360f9420) node->(0x12f59afc0) layout->[self][normal child]
B------- -+          DIV RenderBlock at (8,8) size 304x154 renderer->(0x1360f9550) node->(0x12f59b050) layout->[self][normal child]
N------- -+            IFRAME RenderIFrame at (0,0) size 304x154 renderer->(0x1360f9680) node->(0x12f59b0e0) layout->[self]
^^Note that missing RenderMultiColumnFlowThread and the inner <div> is no longer absolutely positioned.

and when the [outer.style.transition = "1s”] is added back the after state is this ->
B-----L- -+        DIV RenderBlock at (0,0) size 1222x0 renderer->(0x1360f9420) node->(0x1360fbd60) layout->[self][normal child]
B---YGL-           RenderMultiColumnFlowThread at (0,0) size 603x0 renderer->(0x1360f97b0) [Rs:0x0 Re:0x0] layout->[self][normal child]
B------- -+            DIV RenderBlock at (8,8) size 304x154 renderer->(0x1360f9550) node->(0x1360fbdf0) [Rs:0x0 Re:0x0] layout->[self][normal child]
N------- -+              IFRAME RenderIFrame at (0,0) size 304x154 renderer->(0x1360f9680) node->(0x1360fa2a0) layout->[self]
^^the inner <div> is no longer absolutely positioned but we’ve still have a fragmented flow.

It looks like when the inner <div> goes from out-of-flow to in-flow, we don’t check if the enclosing fragmented flow needs a new set.

[Jack]

Created attachment 389894 [details]
Patch

[Jack]

EWS test "Mac-debug-wk1" fails because of <rdar://46103517>. Pending on investigation.

[Ryosuke Niwa]

Comment on attachment 389894 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=389894&action=review

> Source/WebCore/ChangeLog:10
> +        Reviewed by NOBODY (OOPS!).

This line should appear below URL but before the description (surrounded by blank lines as you did here).

[Jack]

Created attachment 391076 [details]
Patch

[Jack]

Submit the change with expected crash in layout test due to bug 202805.

[Ryosuke Niwa]

<rdar://problem/49687828>

[Aakash Jain]

Comment on attachment 391076 [details]
Patch

commit-queue still doesn't support security bugs. Please land manually. See: https://bugs.webkit.org/show_bug.cgi?id=201939

[Ryosuke Niwa]

This is not a security bug.

[Ryosuke Niwa]

Committed r257129: <https://trac.webkit.org/changeset/257129>

[Ryosuke Niwa]

*** Bug 208317 has been marked as a duplicate of this bug. ***