1171773 - [css-grid] ASSERTION: writing-mode mismatch with grid

Status: RESOLVED FIXED

(Core :: Layout, defect)

Description

[Mats Palmgren (inactive)]

Attachment: testcase #1

STR
1. set the pref layout.css.grid.enabled to true
2. load the attached testcase

ACTUAL RESULT
The assertion occurs inside the nsHTMLReflowState ctor when accessing
aAvailableSpace.ISize(mWritingMode).  The nsHTMLReflowState is
created here:
http://hg.mozilla.org/mozilla-central/annotate/5b4c240e1a36/layout/generic/nsGridContainerFrame.cpp#l1264

Comment 1

[Jonathan Kew [:jfkthame]]

Actually, it's not even necessary to enable vertical mode in order to hit this; just setting direction:rtl on the child <span> will hit the same assertions.

When reflowing the child, we need to convert appropriately between the grid container's writing mode and the child's.

Comment 2

[Jonathan Kew [:jfkthame]]

Attachment: Handle reflowing a grid child with different writing mode

I think something along these lines is what we want here. But clearly we need some additional tests, too...

Comment 3

[Mats Palmgren (inactive)]

Comment on attachment 8615882 [details] [diff] [review]
Handle reflowing a grid child with different writing mode

Nit: s/grid child/grid item/ in the commit message (so that it excludes
abs.pos.)

(A vertical abs.pos. child still isn't rendered correctly, but I suspect
we need bug 1079151 for that.  It's also broken with a block CB so it's
not just grid.)

Thanks for the fix!

(and don't worry about tests - I'll make sure to include some in
bug 1151212 that involves different writing modes.  I actually
discovered this while writing those tests!)

Comment 4

[Mats Palmgren (inactive)]

I'm a bit surprised our fuzz testing haven't caught this assertion yet.
Perhaps it's not run with both layout.css.vertical-text.enabled and
layout.css.grid.enabled set to true?

Comment 5

[Jonathan Kew [:jfkthame]]

Attachment: Handle reflowing a grid item with different writing mode from its grid container

Minor update to account for margins on the grid item, which weren't handled correctly in the previous version. Carrying over r=mats.

Comment 6

[Jonathan Kew [:jfkthame]]

Attachment: Reftest for grid item with different direction from its container

OK, I won't spend a lot of time on testcases here, but here's one I wrote earlier that hits these assertions, and runs cleanly with the patch. (It was when I added margins to this that I noticed the original patch didn't handle them right.) A more extensive set of tests, including vertical mode (which still has shrink-to-fit-related issues) would still be valuable, of course.

Comment 7

[Jesse Ruderman]

(In reply to Mats Palmgren (:mats) from comment #4)
> I'm a bit surprised our fuzz testing haven't caught this assertion yet.
> Perhaps it's not run with both layout.css.vertical-text.enabled and
> layout.css.grid.enabled set to true?

My fault for not removing "ASSERTION: writing-mode mismatch" from my known-bugs list after another bug was fixed. I should also generate more complex values for grid-template-columns and grid-template-rows, and change the writing-mode property more frequently.

Comment 8

[Jonathan Kew [:jfkthame]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/e3057db49f4e
https://hg.mozilla.org/integration/mozilla-inbound/rev/f1c34cec973b

Comment 9

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/e3057db49f4e
https://hg.mozilla.org/mozilla-central/rev/f1c34cec973b