1451102 - Get layout/reftests/bugs/614272-1.svg passing with blob invalidation

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, defect)

Comment 1

[Aria Beingessner [:Gankra]]

Attachment: Bug 1451102 - roundout blobimage bounds and use them consistently.

Comment 2

[Aria Beingessner [:Gankra]]

Attachment: Bug 1451102 - roundout blobimage bounds and use that result consistently.

Tweaked to go back to scaling the DT

Comment 3

[Aria Beingessner [:Gankra]]

Attachment: Bug 1451102 - roundout blobimage bounds and use that result consistently.

Review commit: https://reviewboard.mozilla.org/r/234732/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/234732/

Comment 4

[Aria Beingessner [:Gankra]]

Comment on attachment 8965857 [details] [diff] [review]
Bug 1451102 - roundout blobimage bounds and use that result consistently.

Whoops this needed to be squashed with its predecessor!

Comment 5

[Aria Beingessner [:Gankra]]

Comment on attachment 8965923 [details]
Bug 1451102 - roundout blobimage bounds and use that result consistently.

Review request updated; see interdiff: https://reviewboard.mozilla.org/r/234732/diff/1-2/

Comment 6

[Jeff Muizelaar [:jrmuizel]]

Comment on attachment 8965923 [details]
Bug 1451102 - roundout blobimage bounds and use that result consistently.

https://reviewboard.mozilla.org/r/234732/#review240402

Comment 7

[Aria Beingessner [:Gankra]]

try build - before: https://treeherder.mozilla.org/#/jobs?repo=try&revision=e6aa5e5b8ee84b9ca368ec66c749cd36ada0769b

try build - after: https://treeherder.mozilla.org/#/jobs?repo=try&revision=86fe1aaa1a76b55f51dad0128532433a867a093a

Comment 8

[Jeff Muizelaar [:jrmuizel]]

It looks like this makes some things better and some things worse. There's a bunch of new failures in R6 that all look a similar cause.

It also looks like a bunch of tests might have been regressed by bug 1451761

Comment 9

[Aria Beingessner [:Gankra]]

As far as I can tell, the only regression is this huge pile of smil tests which are now losing pixels on the bottom/right: https://hg.mozilla.org/mozilla-central/raw-file/tip/layout/tools/reftest/reftest-analyzer.xhtml#logurl=https://queue.taskcluster.net/v1/task/YOaz9iU-Q-ygA7xa3dLaJA/runs/0/artifacts/public/logs/live_backing.log&only_show_unexpected=1

Just now taking a look into it, but if I had to guess our division trick is inadequate.

Sadly, I haven't been able to reproduce this regression locally.

Comment 10

[Aria Beingessner [:Gankra]]

Oh no wait, the non-gradient test cases here are also regressions: https://hg.mozilla.org/mozilla-central/raw-file/tip/layout/tools/reftest/reftest-analyzer.xhtml#logurl=https://queue.taskcluster.net/v1/task/BME9tV6cTomlCv2HQqSTIA/runs/0/artifacts/public/logs/live_backing.log&only_show_unexpected=1

looks like some text is disappearing.

Comment 11

[Aria Beingessner [:Gankra]]

After some investigation it appears all the regressions we're seeing here are intermittents caused by the bugs in Jeff's dirtyrect stuff. So we should be able to just land this now, and the issues will go away when that gets fixed.

Comment 12

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/d3c654a2185f
roundout blobimage bounds and use that result consistently. r=jrmuizel

Comment 13

[Natalia Csoregi [:nataliaCs]]

https://hg.mozilla.org/mozilla-central/rev/d3c654a2185f