In Firefox, the fiddle does not trigger this bug and the table appears to be a sane size.

I think Morten fixed something like this

This does sound somewhat familiar, especially the bogus table width.

Deleted: tc.html 296 bytes

tc.html 296 bytes View Download

The table sets its max intrinsic width to 1000000 (from kTableMaxWidth, I guess). That's as close to infinity as we get, and all browsers (except Presto) agrees on this IMHO weird behavior.

Deleted: fixed-table-problem.html 244 bytes

fixed-table-problem.html 244 bytes View Download

So.. in fixed-table-problem.html we're at least constrained by the containing block of the shrink to fit container, but there's no such thing inside a flexbox to save us.

Re comment #3: That must have been https://chromium-review.googlesource.com/c/chromium/src/+/1095220 - i.e.  bug 850566 .

In this case, though, I'm not sure if we want to prevent the table algorithm from propagating "infinite" values. The fix for  bug 850566  included changes for both flex and grid, and it would be natural to do the same in this case.

But it looks like at least for grid, all browsers end up with "infinite" wide items in such situations. For flex, both Edge and Firefox limit the width to the width of the flex container. Which looks pretty strange since there are multiple flex lines.

Attaching grid demo. Behaves identically in Chrome, Firefox and Edge.

Deleted: demo-grid.html 284 bytes

demo-grid.html 284 bytes View Download

I won't be working on this any time soon, so unassigning.

I have a fix for flex, this is in fact a bug in our implementation.

Re "pretty strange", https://drafts.csswg.org/css-flexbox/#algo-cross-item requires laying this out as fit-content.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/5918bde3b81e8d418ba290ad0df7a751601d56b7

commit 5918bde3b81e8d418ba290ad0df7a751601d56b7
Author: Christian Biesinger <cbiesinger@chromium.org>
Date: Mon Nov 12 21:57:41 2018

[css-flexbox] ChildIntrinsicLogicalWidth should use fit-content, not max-content

This function needs to match the actual sizing we use. But, since
this is the cross-axis size, we use fit-content:
https://drafts.csswg.org/css-flexbox/#algo-cross-item

See LayoutBox::SizesLogicalWidthToFitContent and LayoutBox::ComputeLogicalWidthUsing
for where we do that fit-content sizing; this code makes us
match the ComputeLogicalWidthUsing calculation.

This bug is currently somewhat hard to trigger because it requires
a combination of all of:
- flex-wrap: wrap
- flex-direction: column
- More than one flex line
- Not using align-items: flex-start or align-content: flex-start
- No explicit width on the flex item
- The max-content width needs to be larger than the width of flexbox

The reason this does not matter in other cases is because we only use
this function to set FlexLine::cross_axis_extent, which we overwrite in
AlignFlexLines if we only have one line. And if we do have more than
one line, we only use this value for aligning items and lines, so
it only matters if we don't use flex-start alignment.

This bug will be much easier to trigger once bug 599828 is fixed (there
are at least two real-world sites that are affected by this once
that bug is fixed)

Bug:  876749 
Change-Id: I673026bedfaf0e4c4a25830ef686b200ec8f54a1
Reviewed-on: https://chromium-review.googlesource.com/c/1327746
Commit-Queue: Christian Biesinger <cbiesinger@chromium.org>
Reviewed-by: Morten Stenshorne <mstensho@chromium.org>
Cr-Commit-Position: refs/heads/master@{#607342}
[modify] https://crrev.com/5918bde3b81e8d418ba290ad0df7a751601d56b7/third_party/WebKit/LayoutTests/TestExpectations
[add] https://crrev.com/5918bde3b81e8d418ba290ad0df7a751601d56b7/third_party/WebKit/LayoutTests/external/wpt/css/css-flexbox/flex-wrap-002.html
[add] https://crrev.com/5918bde3b81e8d418ba290ad0df7a751601d56b7/third_party/WebKit/LayoutTests/external/wpt/css/css-flexbox/flex-wrap-003.html
[modify] https://crrev.com/5918bde3b81e8d418ba290ad0df7a751601d56b7/third_party/blink/renderer/core/layout/layout_flexible_box.cc

Testcase works in latest canary.

As per comment 9, for Grid we match the other browsers, and for Flex we do now too, so closing this.