New issue
Advanced search Search tips
Note: Color blocks (like █ or █) mean that a user may not be available. Tooltip shows the reason.

Issue 675499 link

Starred by 3 users
Status: Fixed
Owner:
█ jvanverth@chromium.org
Last visit > 30 days ago
Closed: Mar 2017
Cc:
fmalita@chromium.org
bsalo...@google.com
Components:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 2
Type: Bug-Regression



Sign in to add a comment

Circle rendering artifacts when under high cpu load

Project Member Reported by pdr@chromium.org, Dec 19 2016 
Chrome Version: 57.0.2950.4/dev, also tip-of-tree
OS: MacOS 10.11.6, low-dpi monitor

What steps will reproduce the problem?
(1) Open the attached testcase, or visit https://pr.gg/corruptiontest.html
(2) Notice the circle rendering bugs.

This testcase is a minimized perf benchmark and moves a bunch of circles on each frame. The bug does not seem to repro with rects instead of circles. The bug does not seem to repro with low numbers of circles either.

I bisected this down to:
https://chromium.googlesource.com/chromium/src/+log/8a02646e95830e7c67dc884280c94b219ae13e7b..c243409f8f9f7cfcae4264c1b1884cbef4feef3c

Which contains this skia roll:
https://chromium.googlesource.com/skia.git/+log/d207884bf5d1..221a4bb55b51

There are several patches in that range that could be the culprit. @Stephen, I don't know if your patch (https://chromium.googlesource.com/skia.git/+/49df8d17c56ee08ecf860289d501913d356f67dc) caused this but it's the first on the list. Could you please triage this?
corruptiontest.html
1.3 KB View Download
corruption.png
118 KB View Download

Comment 1 by ps.prave...@gmail.com, Mar 1 2017 

We are seeing the same issue when drawing hundreds of SVG circles.

Open the link below and notice lines criss-crossing the map - whereas only circles are expected to be displayed.
http://jsapi.maps.arcgis.com/home/webmap/viewer.html?url=http://servicesdev1.arcgis.com/5uh3wwYLNzBuU0Eu/ArcGIS/rest/services/PointFeatureMoreThan3000Features_CustomSR/FeatureServer/0&source=sd

Reproducible on macOS Sierra version 10.12.3 (16D32) + Chrome 56.0.2924.87 (64-bit).
NOT reproducible on Chrome Canary 58.0.3013.0 (64-bit).
(See attached screenshots)

This is a major issue for us as it affects thousands of users of our software. Can you confirm that this issue is indeed fixed in Canary and timeline for when it will land on Stable channel?
stable-56.0.2924.87.png
426 KB View Download
canary-58.0.3013.0.png
481 KB View Download

Comment 2 by senorblanco@chromium.org, Mar 2 2017

Cc: bsalo...@google.com
Owner: jvanverth@chromium.org
Very doubtful it's my change, which was only a refactor.

Bisecting from bad to good, it looks like this was fixed at https://chromium.googlesource.com/chromium/src/+/7273c6c6210628e39f02afa086ed025413cf8611, which includes this skia roll: https://skia.googlesource.com/skia.git/+log/1df161ab8a6a..2c49a4185865

I'm guessing it was fixed by Jim's https://skia.googlesource.com/skia.git/+/8cefe40ab094bfbea532761dad1a857eb3d4b831, in which case this is a dupe of https://bugs.chromium.org/p/chromium/issues/detail?id=690144. That's definitely in 58. Not sure if it was merged back to 57.

Comment 3 by jvanverth@google.com, Mar 2 2017 

I agree it's a dupe of 690144. It hasn't been merged back to 57 because some people are reporting a perf regression from the fix which I haven't been able to repro.

Comment 4 by schenney@chromium.org, Mar 3 2017

Labels: BugSource-Team PaintTeamTriaged-20161219

Comment 5 by jvanverth@chromium.org, Mar 6 2017

Status: Fixed (was: Assigned)
I'm marking this as fixed. A workaround for now should be to draw the circles as very slight ovals (as in, add something very small like 1.0e-4 to the radius in one direction). The perf regression that stopped the merge into M57 is  bug 695466 .
►

Sign in to add a comment