1857779 - Jagged outline on a CSS Codepen demo (https://codepen.io/alvaromontoro/pen/YzOjOWQ)

Status: NEW

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

Description

[Mayank Bansal]

Go to https://codepen.io/alvaromontoro/pen/YzOjOWQ

Observe that the joint of the right hand has a permanent jagged line/outline.

ER: Not so

I have observed this bug before, but assumed that it would be fixed bug 1850869/bug 1671784 and co.

Comment 1

[Mayank Bansal]

Attachment: about:suppport

Comment 2

[Mayank Bansal]

Attachment: Screenshot 2023-10-08 113423.png

I guess there are two types of jagged outlines here:

1. The one encircled in red ink - This jaggedness is consistent while the arm moves. So probably more important and something that would have been fixed by the recent clip-mask fixes
2. The two encircled in blue ink - They looks to be more of snapping?

Comment 3

[Blake]

mozregression isn't perfect as when those lines didn't appear, the tail didn't appear at certain zoom levels.

Last good build: 2020-12-12
First bad build: 2020-12-12
build_url: https://archive.mozilla.org/pub/firefox/nightly/2020/12/2020-12-12-21-29-13-mozilla-central/firefox-85.0a1.en-US.win64.zip
changeset: 8d6972b0b89b1303dbcfb566e16b1144303af83e
pushlog_url: https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=8d6972b0b89b1303dbcfb566e16b1144303af83e&tochange=8c4e9bba69eabce48600c6872cf3032229591fa3

Comment 4

[Blake]

(In reply to Blake from comment #3)

mozregression isn't perfect as when those lines didn't appear, the tail didn't appear at certain zoom levels.

Last good build: 2020-12-12
First bad build: 2020-12-12
build_url: https://archive.mozilla.org/pub/firefox/nightly/2020/12/2020-12-12-21-29-13-mozilla-central/firefox-85.0a1.en-US.win64.zip
changeset: 8d6972b0b89b1303dbcfb566e16b1144303af83e
pushlog_url: https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=8d6972b0b89b1303dbcfb566e16b1144303af83e&tochange=8c4e9bba69eabce48600c6872cf3032229591fa3

First bad build was the 13th with is this changelog: pushlog_url: https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=1130661c79c222fb1acd29b7ec5dc5202cdd0d2d&tochange=6567bdeb3c1e97a75f12fd72d92a6fb4a2cfcc91

Comment 5

[Mayank Bansal]

(In reply to Blake from comment #4)

(In reply to Blake from comment #3)

mozregression isn't perfect as when those lines didn't appear, the tail didn't appear at certain zoom levels.

Last good build: 2020-12-12
First bad build: 2020-12-12
build_url: https://archive.mozilla.org/pub/firefox/nightly/2020/12/2020-12-12-21-29-13-mozilla-central/firefox-85.0a1.en-US.win64.zip
changeset: 8d6972b0b89b1303dbcfb566e16b1144303af83e
pushlog_url: https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=8d6972b0b89b1303dbcfb566e16b1144303af83e&tochange=8c4e9bba69eabce48600c6872cf3032229591fa3

First bad build was the 13th with is this changelog: pushlog_url: https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=1130661c79c222fb1acd29b7ec5dc5202cdd0d2d&tochange=6567bdeb3c1e97a75f12fd72d92a6fb4a2cfcc91

This is probably just bug 1681921, which would enable Webrender on your machine (in the form of sw-wr).

Comment 6

[Mayank Bansal]

Attachment: reduced_testcase that is good on both Firefox and Nightly.html

Comment 7

[Mayank Bansal]

Attachment: reduced_testcase that is bad on both Firefox and Nightly.html

Comment 8

[Mayank Bansal]

I am confused here.
On the original testcase, Firefox is buggy and Chrome is good.
I tried to reduce the testcases, and i ended up with 2 reduced testcases - One is good on both browsers, the other is bad on both browsers.

Comment 9

[Nicolas Silva [:nical]]

The problem here is that a rotated rectangle should be drawn anti-aliased whereas if it is axis-aligned it should be drawn without AA (snapped to the pixel grid). If the browser decides to rasterize the rectangles rotated directly, then the anti-aliasing at its border causes a bit of transparency through which we can see a bit of the black rectangle under it. If the browser decides to rasterize the rectangle into a layer and then rotate it, then it is first drawn without AA and the black rectangle does not show through it.

With the simpler test case, chrome took a different layerization decision.

I don't think that the correct behavior is well specified here.