Created attachment 8713405 [details] canvas_fbf.html After executing the exact same commands on a canvas there are some pixels that are not identical to the previous image data. In my testing the amount of pixels varies per run and sometimes they are all identical (this is why in my test I run it 10 times). I noticed this from pdf.js because our foward-back-foward tests are no longer passing on OSX. STR: 1) Open attached example Expected: background turns green Actual: background turns red After running mozgression I've narrowed it down to bug 1082598 The attached example is a simplified PDF that I've taken the canvas drawing commands from, it could probably be simplified further.
It would help if we could get a much more minimal test-case showing the divergence. Right now it is extremely difficult to dissect what is going on in this.
Created attachment 8715422 [details] canvas_fbf.html I've reduced the number of drawing commands, but I had to increase the number of rounds since it is now even more intermittent.
Created attachment 8715851 [details] a cleaned up version of the testcase Filed upstream Skia bug report here: https://bugs.chromium.org/p/skia/issues/detail?id=4894
It should be possible to get to the root cause of this pretty easily with rr.
Firefox: 45.0.1, Build ID 20160315153207 User Agent Mozilla/5.0 (Macintosh; Intel Mac OS X 10.8; rv:45.0) Gecko/20100101 Firefox/45.0 Hi reporter, I have tried to test this issue on the latest Firefox (45.0.1) release, latest Nightly (48.0a1 - Build ID: 20160324030447) build, but when I open the provided test cases I only see a page that has a green background. Can you please provide the commands that are required to reproduce this issue? Also, is this still reproducible on your end ? If yes, can you please retest this using latest FF release and latest Nightly build (https://nightly.mozilla.org/) and report back the results ? When doing this, please use a new clean Firefox profile, maybe even safe mode, to eliminate custom settings as a possible cause (https://goo.gl/PNe90E). Thanks, Cosmin.
Using "a cleaned up version of the testcase" on 48.0a1 (2016-03-24) I can still reproduce this, though it is still intermittent. I had to refresh the page ~5 times to get it to turn red.
Firefox: 48.0a1, Build ID: 20160324030447 User Agent Mozilla/5.0 (Macintosh; Intel Mac OS X 10.11; rv:48.0) Gecko/20100101 Firefox/48.0 Hi Berndan, I have tested this issue again on Mac OS 10.11 and I could reproduce it on the latest Nightly (48.0a1) build with the provided test case "a cleaned up version of the testcase" from comment 3. This is not reproducible on the latest Firefox (45.0.1) release so I have performed a regression. This are the results: Last good revision: d8e260e1535f0ff3877b21d6846380dedf449453 First bad revision: 6630a176477ef903dcf3057db755abc43ba5018d Pushlog: https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=d8e260e1535f0ff3877b21d6846380dedf449453&tochange=6630a176477ef903dcf3057db755abc43ba5018d 26:11.60 INFO: Looks like the following bug has the changes which introduced the regression: bug 1082598 Thanks, Cosmin.
I can't reproduce this. Is this bug still reproducible for anyone else?
The original test case still fails: https://bugzilla.mozilla.org/attachment.cgi?id=8713405