1675167 - Intermittent transform/partial-prerender-translate-2.html == transform/partial-prerender-translate-2-ref.html | image comparison, max difference: 255, number of differing pixels: 7000

Looks like we just need some slight increases on the fuzzy thresholds here -- hiro, would you mind taking a look? (I think you wrote this test, so you might have some insights into that vs. whether there's an even-better fix) Looks like this is a fairly frequent failure (particularly on Mac) as shown in comment 71.

I ran across this bug because I just happened to hit this on a Try run ( https://treeherder.mozilla.org/logviewer?job_id=384943102&repo=try&lineNumber=6661 ) In my case, the test failed with max difference: 57, number of differing pixels: 1000, where the max-difference is just a bit higher than the annotated threshold, fuzzy(0-43,0-1800)

The mismatching pixels are all in the extreme far right column of pixels, and it seems to be a lighter shade of green in the testcase vs. the reference case for some reason.

Flags: needinfo?(hikezoe.birchill)

Hiroyuki Ikezoe (:hiro)

Comment 73

•

2 years ago

(In reply to Daniel Holbert [:dholbert] from comment #72)

Looks like we just need some slight increases on the fuzzy thresholds here -- hiro, would you mind taking a look? (I think you wrote this test, so you might have some insights into that vs. whether there's an even-better fix) Looks like this is a fairly frequent failure (particularly on Mac) as shown in comment 71.

I ran across this bug because I just happened to hit this on a Try run ( https://treeherder.mozilla.org/logviewer?job_id=384943102&repo=try&lineNumber=6661 ) In my case, the test failed with max difference: 57, number of differing pixels: 1000, where the max-difference is just a bit higher than the annotated threshold, fuzzy(0-43,0-1800)

The pixels are all on the extreme far right column of pixels, and it seems to be a lighter shade of green in the testcase vs. the reference case for some reason.

Yeah. In the testing case, the lither area is the rightmost 1px of the green box, next to the 1px area there should be white area so that, I guess, the lither color is a result of anti-aliasing with the white area.

From the perspective of partial-prerender animations, it will not matter since, I believe, users won't notice the anti-aliasing results during moving the animation. So addinf fuzzy-if would be fine.

That's said, I wonder the anti-aliasing result implies WebRender has recently changed something related anti-aliasing? I guess, I've heard that WebRender did refactor clip related things so that it did start anti-aliasing outside of the viewport? CCing Glenn.

Flags: needinfo?(hikezoe.birchill)

Daniel Holbert [:dholbert]

Comment 74

•

2 years ago

(In reply to Hiroyuki Ikezoe (:hiro) from comment #73)

So addinf fuzzy-if would be fine.

That's said, I wonder the anti-aliasing result implies WebRender has recently changed something related anti-aliasing?

Two clarifications:

The test already has a fuzzy-if annotation (no need to add a new one); we just need to update its max-difference metric, I think.
This intermittent issue isn't a particularly-recent thing; it just happened to come to my attention today. It goes back a year at least (for webrender test runs, which I think are more of our test-runs now). E.g. comment 27 had 11 failures in a single week, about a year ago. And in Comment 49, ~5 months ago, we had 33 failures in a single week.

So: probably not a recent WebRender behavior-change; maybe just a WebRender quirk that's been present all along, and got more frequent (in terms of reports) over the past year due to us switching to WebRender-only.

Comment hidden (Intermittent Failures Robot)

Joel Maher ( :jmaher ) (UTC -8)

Updated

•

2 years ago

Blocks: 1788323

Joel Maher ( :jmaher ) (UTC -8)

Updated

•

2 years ago

Status: REOPENED → RESOLVED

Closed: 4 years ago → 2 years ago

Resolution: --- → DUPLICATE