1475827 - Missing and flickering tile borders on about:newtab

Status: VERIFIED FIXED

(Core :: Graphics: WebRender, defect, P2)

Description

[Darkspirit]

Attachment: testcase.html

GTX 1060, Win10 + Debian Testing

mozregression --good 2018-04-01 --bad 2018-07-14 --pref gfx.webrender.all:true browser.newtabpage.activity-stream.feeds.section.highlights:false browser.newtabpage.activity-stream.feeds.section.topstories:false browser.newtabpage.activity-stream.topSitesRows:3 browser.newtabpage.activity-stream.migrationExpired:true startup.homepage_welcome_url:'about:newtab'
> 6:07.16 INFO: Last good revision: 1344ca26ef0aec57cb0177f7089287b263c7a52b
> 6:07.16 INFO: First bad revision: 9896401414cb3c7cb5f1118b307b0267212c8aad
> 6:07.16 INFO: Pushlog:
> https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=1344ca26ef0aec57cb0177f7089287b263c7a52b&tochange=9896401414cb3c7cb5f1118b307b0267212c8aad

> 9896401414cb	Ed Lee — Bug 1454074 - Add chronological highlights, dark updates and bug fixes to Activity Stream. r=ursula

= new about:newtab

... made the testcase ...


good = flickering border of the first tile when changing window size
bad = above + Missing or too thick borders

mozregression --good 2018-04-01 --bad 2018-07-14 --pref gfx.webrender.all:true startup.homepage_welcome_url:'file:///C:/Users/Darkspirit/Desktop/testcase.html'
> 5:48.16 INFO: Last good revision: 9f436684f4f7425e7af421dec2ae39464ec34b2d
> 5:48.16 INFO: First bad revision: f0ef44027999330a69d31db01e654a22949e2726
> 5:48.16 INFO: Pushlog:
> https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=9f436684f4f7425e7af421dec2ae39464ec34b2d&tochange=f0ef44027999330a69d31db01e654a22949e2726

> 140454d5de2e	Kartikaya Gupta — Bug 1451453 - Update for API change in WR PR 2613. r=jrmuizel
> 8f46d004ff0d	Kartikaya Gupta — Bug 1451453 - Update for API change in WR PR 2600. r=jrmuizel
> f20cbdbdc3ac	Kartikaya Gupta — Bug 1451453 - Update webrender to commit 092ada1154b72fe71d2f227a5df0239586d2323a. r=jrmuizel

(has other regressions)

Comment 1

[Darkspirit]

Attachment: last good.mp4

last good:
mozregression --repo autoland --launch 9f436684f4f7425e7af421dec2ae39464ec34b2d --pref gfx.webrender.all:true startup.homepage_welcome_url:'https://bug1475827.bmoattachments.org/attachment.cgi?id=8992184'

Comment 2

[Darkspirit]

Attachment: first bad.mp4

mozregression --repo autoland --launch f0ef44027999330a69d31db01e654a22949e2726 --pref gfx.webrender.all:true startup.homepage_welcome_url:'https://bug1475827.bmoattachments.org/attachment.cgi?id=8992184'

Comment 3

[Aria Beingessner [:Gankra]]

Is this still an issue? Looks fine on my windows machine.

Comment 4

[Darkspirit]

It still behaves the same.
But if I change the window height about one pixel, it's just fine when changing the window width. So I would assume that this might be something like bug 1451439.

Comment 5

[Darkspirit]

Not fixed by bug 1451439.

Comment 6

[Darkspirit]

It's fixed in https://treeherder.mozilla.org/#/jobs?repo=try&revision=541116a30e5063aa896cdf54724d63bfac989427.

Comment 7

[Chris Pearce [:cpearce (Not reading bugmail)]]

I can't repro this on my hardware. The patch in the try push you linked to includes a number of different fixes. The testcase includes "box-shadow", which is one of the fixes. So I've split that fix out into a separate patch:

https://treeherder.mozilla.org/#/jobs?repo=try&revision=5d2e732af5973d3417cb3bcab5b1854a300b21fa

darkspirit: are you able to try the build in the above Try push to see if it fixes the issue for you? Thanks!

Comment 8

[Darkspirit]

Yep, it fixes the issue.

Comment 9

[Darkspirit]

It hasn't landed yet.

Comment 10

[Chris Pearce [:cpearce (Not reading bugmail)]]

(In reply to Jan Andre Ikenmeyer [:darkspirit] from comment #9)
> It hasn't landed yet.

Patch causes test failures. I am working through them and the consequences of the fix for them.

Comment 11

[Liz Henry (:lizzard) (relman/hg->git project)]

Marking fix-optional for 64. We could still take a patch for 65, and if it's verified and doesn't seem risky, could still take fixes for 64 as well.

Comment 12

[Jeff Muizelaar [:jrmuizel]]

Chris, what happens next here?

Comment 13

[Chris Pearce [:cpearce (Not reading bugmail)]]

In order for the tests to not fail, we need to fix https://github.com/servo/webrender/issues/3151 I think the issue there is in the border shadow shader, and I've not had much success trying to fix that. Ideally, someone who understands shaders and how WebRender's shaders work could fix that.

It's also not clear to me that rounding to device pixels before WR applies transforms will result in correct rendering, particularly in the case of subpixel positioning.

Comment 14

[Jeff Muizelaar [:jrmuizel]]

Nical, interested in picking up this one?

Comment 15

[Nicolas Silva [:nical]]

Attachment: Bug 1475827 - Change the way intermediate target rects are snapped. r=gw

Rounding out the rect causes fractional position of the rect to affect its final size.
If the size of a mask is rounded independently it better matches the size of the primitive that will read from it and avoids some noticeable asymmetry.

Comment 16

[Jeff Muizelaar [:jrmuizel]]

Do we know whether bug 1522351 helps here?

Comment 17

[Darkspirit]

Attachment: 2019-01-25 00-44-40.mp4

It doesn't look like bug 1522351 comment 7 had an effect on this.
In general, borders are not completely disappearing anymore, but it's still noticeable.

Comment 18

[Jeff Muizelaar [:jrmuizel]]

Glenn knows the root cause.

Comment 19

[Jeff Muizelaar [:jrmuizel]]

This seems fixed. If not please reopen.

Comment 20

[Cristian Comorasu, QA [:ccomorasu], Releae Desktop QA - Please contact [e-mail]]

I can confirm this issue is no longer reproducible. I verified using Fx 67.0b18 on Windows 10 x64, macOS 10.14.5 and Ubuntu 18.04 LTS.