1546133 - Mysterious color shadow was incorrectly painted near around the texts

Status: NEW

(Core :: Graphics: Text, defect, P3)

Description

[Alastor Wu [:alwu]]

I found that there are a lots of WebVTT ref tests failed [1] because some mysterious unexpected color showing around the text, it looks like a colorful shadow, but the ref files don't have such CSS property.

The completed try result is in here [2].

[1] https://hg.mozilla.org/mozilla-central/raw-file/tip/layout/tools/reftest/reftest-analyzer.xhtml#logurl=https://queue.taskcluster.net/v1/task/MHt1ISX1SVWHssmXl9Jo5A/runs/0/artifacts/public/logs/live_backing.log&only_show_unexpected=1

[2] https://treeherder.mozilla.org/#/jobs?repo=try&revision=9b88a91bd510481988a4f66b1968276a6faddb24&selectedJob=241364222

Comment 1

[Jonathan Kew [:jfkthame]]

Looks like the tests are being rendered with grayscale antialiasing, while the reference is using cleartype (subpixel) AA.

The cleartype "color fringing" on the reference does look surprisingly strong; I wonder if it's not using the correct gamma, or something like that?

Comment 2

[Alastor Wu [:alwu]]

Hi, Jonathan,
Do you think if this issue is easy to get a fix? Or it's a complex issue and I should disable those tests on Windows and wait for someone to take a look for this issue?
Thank you!

Comment 3

[Jonathan Kew [:jfkthame]]

Lee is more likely to know if there's an easy way to fix this, I think; I'm not sure of all the exact criteria that go into determining what kind of antialiasing will be used when the text is painted.

Comment 4

[Lee Salzman [:lsalzman]]

As far as the text backends are concerned, whatever AA layout asks for is what they end up getting. I think maybe Matt or Markus would have a better idea of the situations where AA gets disabled or not by FLB above...

Comment 5

[Matt Woodrow (:mattwoodrow)]

It might be better if the reference file also used a <video> (or maybe a <canvas>?) to put under the text, so that the only difference between the two is the text (normal vs WebVTT).

The current case has dynamic content under the text in the test file (the video), and static under the text in ref (a white div), so we take quite different rendering paths, depending on platform and backend.

Comment 6

[Alastor Wu [:alwu]]

(In reply to Matt Woodrow (:mattwoodrow) from comment #5)

It might be better if the reference file also used a <video> (or maybe a <canvas>?) to put under the text, so that the only difference between the two is the text (normal vs WebVTT).

The current case has dynamic content under the text in the test file (the video), and static under the text in ref (a white div), so we take quite different rendering paths, depending on platform and backend.

Good idea, using <video> in reference file makes more sense.

Comment 7

[Alastor Wu [:alwu]]

Take basic-ref.html as an example [1], I made a example to put the text over the <video> [2]. Is that what you meant?

[1] https://searchfox.org/mozilla-central/source/testing/web-platform/tests/webvtt/rendering/cues-with-video/processing-model/basic-ref.html
[2] https://codepen.io/anon/pen/dEmLap

Comment 8

[Matt Woodrow (:mattwoodrow)]

I'm not sure if that will work, since from the display list side, there isn't actually any video content there.

I think you need to do the same as the test variant, where you load a white mp4/webm, set the autoplay attribute, and make the test wait for the onplaying event.

Comment 9

[Alastor Wu [:alwu]]

Attachment: Bug 1546133 - part1 : use <video> for reference tests.

In order to use same rendering path for test and reference files, we should also use <video> in reference files.

Comment 10

[Alastor Wu [:alwu]]

Attachment: Bug 1546133 - part2 : enable wpt on Windows.

As we have let test and reference file using same rendering path, we can re-enable those wpt on Windows.

Comment 11

[Alastor Wu [:alwu]]

Hi, Matt,
This issue still happened on Win7 [1], even if we used <video> tag for the reference files. Do you have any idea about why this issue only happen on Win7?
Thank you.

[1] https://treeherder.mozilla.org/#/jobs?repo=try&revision=afb2de6153a07bfac010202f914d9bf51452c5d3&selectedJob=248821538

Comment 12

[Matt Woodrow (:mattwoodrow)]

Do you need to include class="reftest-wait" and the reftest-wait.js script?

https://searchfox.org/mozilla-central/source/testing/web-platform/tests/css/css-layout-api/child-constraints/available-size-for-percentages-invalid.https.html#2

Comment 13

[Alastor Wu [:alwu]]

After adding class="reftest-wait" and including reftest-wait.js, the issue still happened [1]. It seems that we still have some rendering issue on Windows even if we use <video> on both test and reference files.

[1] https://treeherder.mozilla.org/#/jobs?repo=try&revision=17d97a24c8ffd9d06dfefc27525291a8b38efbc1&selectedJob=248992626

Comment 14

[Alastor Wu [:alwu]]

I believe that we should still fix this bug, not working around by changing tests.