| Code-Review | +1 |
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
| Code-Review | +1 |
ui/ozone/platform/wayland/wayland_buffer_manager_unittest.cc lgtm
| Auto-Submit | +1 |
| Commit-Queue | +1 |
Ok, ready for a reland after fixing AXObject
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
if (!std::isfinite(bounds_in_container.x())) {Can/should this code be moved into `GetRelativeBounds()`?
if (!std::isfinite(bounds_in_container.x())) {Can/should this code be moved into `GetRelativeBounds()`?
I'm not sure about that. GetRelativeBounds is a virtual and has (currently) 3 implementations. It felt like a smaller change to just put the sanitization here (once). PopulateAXRelativeBounds is also called in only 2 places the Action and the serialization path, GetRelativeBounds is called in more places. In short, this seems like a smaller blast radius.
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
goto done;why goto is necessary here? can we break?
if (bounds.transform) {is this test actually running? Can we make it more explicit / detemrinistic? this if should not be necessary here. if the test is testing what it is supposed to test, the values would be infinite and we would be checking the identity matrix.
| Auto-Submit | +1 |
| Commit-Queue | +1 |
why goto is necessary here? can we break?
There's a nested loop, so AFAIK, a goto is the cleanest way to do that.
is this test actually running? Can we make it more explicit / detemrinistic? this if should not be necessary here. if the test is testing what it is supposed to test, the values would be infinite and we would be checking the identity matrix.
Good catch. This was left over from earlier. Also fixed below.
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
| Commit-Queue | +2 |
Kaylee LubickCan/should this code be moved into `GetRelativeBounds()`?
I'm not sure about that. GetRelativeBounds is a virtual and has (currently) 3 implementations. It felt like a smaller change to just put the sanitization here (once). PopulateAXRelativeBounds is also called in only 2 places the Action and the serialization path, GetRelativeBounds is called in more places. In short, this seems like a smaller blast radius.
Reland "Add isfinite checks to Skia Mojo boundaries"
This is a reland of commit c110228380f30e0a3588d07ad0b691ecf1c41659
The reason we had to revert was because a few layout tests failed
in the force accessibility build. [1] That code path didn't prevent
infinities or NaNs from getting to the Mojo layer. I think it
should be squelching those, otherwise those values could cause
problems after serialization. Thus, I fixed the place in AXObject
where those problematic values seemed to sneak in. I also added
a unit test to make this easier to find.
While I was taking a second look, I added [[unlikely]] to the
mojo checks to help the compiler write better code for these checks.
Original change's description:
> Add isfinite checks to Skia Mojo boundaries
>
> The linked bug refers to NaNs sneaking in where they don't belong
> so I fixed those in the one spot listed in the bugs and other
> related spots. I added a unit test to verify the original buggy
> behavior was fixed.
>
> This deletes some tests that were added to make sure NaNs don't
> make it into Wayland, but because we are adding them here, they've
> been superceded and can be removed.
>
> The changes to lens/overlay find cases where we were sometimes
> dividing by zero, which put NaNs in the rects. This catches those
> more gracefully, fixing some of the failing tests.
>
> Bug: 500390256
> Fixed: 500390256
> Bug: 520300213
> Change-Id: I181112dc4bf8a77e242e95ffc3a31cfc9917c321
> Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/7897559
> Reviewed-by: Giovanni Ortuno Urquidi <ort...@chromium.org>
> Reviewed-by: Thomas Anderson <thomasa...@chromium.org>
> Reviewed-by: Juan Mojica <juanm...@google.com>
> Auto-Submit: Kaylee Lubick <kjlu...@chromium.org>
> Commit-Queue: Thomas Anderson <thomasa...@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#1644135}
[1] https://ci.chromium.org/ui/p/chromium/builders/ci/linux-blink-web-tests-force-accessibility-rel/47693/overview
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |