Closed Bug 1089768 Opened 6 years ago Closed 6 years ago

Intermittent 113% Linux32 SVG Opacity Regression on Inbound (v.36) Oct 26 from push f51420708a03

Categories

(Core :: SVG, defect)

x86_64
Linux
defect
Not set
normal

Tracking

()

RESOLVED WORKSFORME

People

(Reporter: jmaher, Unassigned)

References

Details

(Keywords: perf, regression, Whiteboard: [talos_regression])

This isn't a straight regression. The numbers are jumping up and down between where they used to be and the regressed maximum.

Ubuntu HW 12.04 and this test also have a history of exhibiting this behavior every so often. I've no idea why every so often it jumps up often to double the amount of time it was previously taking on this OS.

Ubuntu HW 12.04 x64 is unchanged, as is typically the case.

As was the case for bug 1014789, I won't be surprised if we wontfix this.
No longer blocks: 1085321, 1088781
Summary: 107% Linux32 SVG Opacity Regression on Inbound (v.36) Oct 26 from push f51420708a03 → Intermittent 113% Linux32 SVG Opacity Regression on Inbound (v.36) Oct 26 from push f51420708a03
so this showed up on Aurora (v.35) from this push:
http://hg.mozilla.org/releases/mozilla-aurora/pushloghtml?changeset=f59f7d0605b8

Looking at the graph:
http://graphs.mozilla.org/graph.html#tests=%5B%5B225,52,33%5D%5D&sel=none&displayrange=90&datatype=running

it shows some multi modality in the test.  Quite possible the svg opacity test needs to be re-evaluated?
We tried to re-evaluate tsvg-opacity at bug 1027481, and we sort of concluded that it's still working well.

I never liked this test, but it seems(ed?) to be working, and I don't really know what to change in order to make it better or more reliable.

I think it would be best to understand first why did these changesets change the way the test behaves - only on Ubuntu 32. If we can do that then we could probably improve the test.

But as long as the same test which runs on seemingly consistent setup produces such noisy results without anyone understanding why, I don't think we have a starting point to re-evaluate the test.
Bas thought that the  might be to do with XLib backed cairo treating paths in a more optimized way to rectangles.
I do not see this on mozilla-beta, can we close this?
Status: NEW → RESOLVED
Closed: 6 years ago
Resolution: --- → WORKSFORME
You need to log in before you can comment on or make changes to this bug.