1954762 - Reflow orthogonal child frame during nsLayoutUtils::IntrinsicForAxis if its intrinsic block size is needed

Status: RESOLVED FIXED

(Core :: Layout, defect)

Description

[Jonathan Kew [:jfkthame]]

With the improved support for orthogonal flow sizing implemented in bug 1949064, I think it will now be feasible for us to reflow orthogonal child frames when needed during IntrinsicForAxis, so that we can get a valid block size rather than using a potentially uninitialized or stale value.

Comment 1

[Jonathan Kew [:jfkthame]]

It turns out that since bug 1949064 fixed the available-size computation for orthogonal flows, this becomes quite simple: we can do a "measuring reflow" of an orthogonal child by setting up a dummy parent ReflowInput (similar to what Grid code does), and using the resulting desired size of the child.

This resolves a large number of the existing test failures related to orthogonal flows. (This includes a couple that are part of interop-2025, but also dozens of other WPT writing-modes tests.)

Once this is landed, we should revisit any real-world examples in existing reports (bug 1955008) to check if they are adequately resolved, and investigate whether there are cases where the "dummy" parent RI is insufficient and a "real" ancestor RI is needed for context (bug 1955921).

Comment 2

[Jonathan Kew [:jfkthame]]

Attachment: Bug 1954762 - If nsLayoutUtils::IntrinsicForAxis needs the intrinsic block-size of an orthogonal child frame, do an immediate measuring reflow to get its desired size. r=#layout

This reflows the child frame using a "dummy" parent ReflowInput, and does not provide
a valid containing block. This means proper positioning of the child would not be
possible in vertical-rl writing mode; but we don't apply positioning here, as we only
care about its desired size and don't actually update the child's mRect (we don't call
FinishReflowChild or equivalent), so this is OK.

Comment 3

[Jonathan Kew [:jfkthame]]

Attachment: Bug 1954762 - Remove failure metadata for WPT tests that now pass. r=#layout

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/5406da763b22
If nsLayoutUtils::IntrinsicForAxis needs the intrinsic block-size of an orthogonal child frame, do an immediate measuring reflow to get its desired size. r=layout-reviewers,emilio,TYLin
https://hg.mozilla.org/integration/autoland/rev/edb5a7edfcde
Remove failure metadata for WPT tests that now pass. r=TYLin

Comment 5

[Cristian Tuns]

https://hg.mozilla.org/mozilla-central/rev/5406da763b22
https://hg.mozilla.org/mozilla-central/rev/edb5a7edfcde