RESOLVED FIXED206194
webkitpy: Pass ImageDiff commands in a single write 
https://bugs.webkit.org/show_bug.cgi?id=206194
Summary webkitpy: Pass ImageDiff commands in a single write
Jonathan Bedard
Reported 2020-01-13 14:14:54 PST
It seems that ImageDiff can get misaligned and timeout after the Python 3 transition.
Attachments
Patch (1.25 KB, patch)
2020-01-13 14:17 PST, Jonathan Bedard 		no flags
DetailsFormatted DiffDiff
Patch (7.51 KB, patch)
2020-01-14 12:15 PST, Jonathan Bedard 		no flags
DetailsFormatted DiffDiff
Show Obsolete (1) View All Add attachment proposed patch, testcase, etc.
Jonathan Bedard
Comment 1 2020-01-13 14:17:40 PST
Created attachment 387566 [details] Patch
Alexey Proskuryakov
Comment 2 2020-01-13 19:31:20 PST
The patch appears incomplete. But also, how can this be caused by Python3, does this imply a bug elsewhere.
Jonathan Bedard
Comment 3 2020-01-14 08:50:33 PST
(In reply to Alexey Proskuryakov from comment #2) > The patch appears incomplete. But also, how can this be caused by Python3, > does this imply a bug elsewhere. The uploaded patch was actually intended to test EWS. The bug is real, though. I think this could be caused by a bug elsewhere (especially since it flakily reproduces), but I can bisect this to the Python 3 compatibility change, and I've made some significant progress towards narrowing down what part of that change actually causes the problem.
Alexey Proskuryakov
Comment 4 2020-01-14 09:30:01 PST
It is also possible that the flakiness is caused by actually getting different results (e.g. there is a preceding test that changes color settings permanently, affecting subsequent tests).
Jonathan Bedard
Comment 5 2020-01-14 10:21:04 PST
(In reply to Alexey Proskuryakov from comment #4) > It is also possible that the flakiness is caused by actually getting > different results (e.g. there is a preceding test that changes color > settings permanently, affecting subsequent tests). The root cause seems to be comparing images with different sizes, though. I've narrowed it down to a few lines of code that have changed in image_diff.py, so I'll have a fix shortly.
Alexey Proskuryakov
Comment 6 2020-01-14 10:42:37 PST
if all of these are different sizes, we should also file a bug to get the tests fixed, it's not really OK for them to rely on automatic resizing.
Jonathan Bedard
Comment 7 2020-01-14 11:28:40 PST
(In reply to Alexey Proskuryakov from comment #6) > if all of these are different sizes, we should also file a bug to get the > tests fixed, it's not really OK for them to rely on automatic resizing. If it is the fault of the tests, then yes, we should/ But I'm not certain that's the root of the problem. I think the root of the problem might actually be ImageDiff getting tripped up when we feed it one image at a time instead of both images at the same time, ImageDiff is reporting one of the images to be exactly twice the size of the other.
Jonathan Bedard
Comment 8 2020-01-14 12:15:30 PST
Created attachment 387686 [details] Patch
Radar WebKit Bug Importer
Comment 9 2020-01-14 13:15:20 PST
<rdar://problem/58578775>
Jonathan Bedard
Comment 10 2020-01-14 13:32:13 PST
Committed r254532: <https://trac.webkit.org/changeset/254532>
Alexey Proskuryakov
Comment 11 2020-01-14 22:56:05 PST
*** Bug 206179 has been marked as a duplicate of this bug. ***
Alexey Proskuryakov
Comment 12 2020-01-14 22:57:11 PST
> If it is the fault of the tests, then yes, we should/ But I'm not certain > that's the root of the problem. I agree that this is not the root of the problem at hand, but the tests are broken too, I don't think that they need to do this, or are allowed to.
Note You need to log in before you can comment on or make changes to this bug.