Closed Bug 1266933 Opened 8 years ago Closed 8 years ago

Skia causes strange line on a website

Categories

(Core :: Graphics, defect)

defect
Not set
normal

Tracking

()

VERIFIED FIXED
mozilla49
Tracking Status
firefox47 --- unaffected
firefox48 + verified
firefox49 --- verified
firefox50 --- verified

People

(Reporter: soeren.hentzschel, Assigned: mchang)

References

Details

(Keywords: regression, Whiteboard: [gfx-noted])

Attachments

(3 files)

Attached image firefox48-bug.png
There is a strange line on a website which seems to be a Firefox bug because this line is only visible in Firefox 48 Nightly and not in Firefox 45 Stable or Firefox 47 Developer Edition and not in other browsers like Chrome.

URL of the website: https://forum.sozone.de/

I can't provide a test case because I don't know what the reason for this line is. It's reproducible with a new Firefox profile. Tested on OS X 10.11.5 Beta.
Keywords: regression
12:57.85 INFO: Last good revision: 3587b25bae302c1eed72968dbd7cef883e715948 (2016-03-22)
12:57.85 INFO: First bad revision: efe7d026ac641759838dd3897c37892e37e5b244 (2016-03-23)
12:57.85 INFO: Pushlog:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=3587b25bae302c1eed72968dbd7cef883e715948&tochange=efe7d026ac641759838dd3897c37892e37e5b244

btw. e10s on/off makes no difference.
Further bisected:

30:27.70 INFO: Last good revision: 9873db170eaa08b203fd39fbf8f02fcb8358a10e
30:27.70 INFO: First bad revision: 66ee9fcbbbd3f207c3fcc141aa2da12ca9534894
30:27.70 INFO: Pushlog:
https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=9873db170eaa08b203fd39fbf8f02fcb8358a10e&tochange=66ee9fcbbbd3f207c3fcc141aa2da12ca9534894

30:29.42 INFO: Looks like the following bug has the changes which introduced the regression:
https://bugzilla.mozilla.org/show_bug.cgi?id=1207332

I can confirm that disabling skia "solves" the problem.
Blocks: skia-osx
Summary: Firefox 48: strange line on a website (but not in Firefox 47 and not in Chrome) → Skia causes strange line on a website
Component: CSS Parsing and Computation → Graphics
Assignee: nobody → mchang
Whiteboard: [gfx-noted]
We don't pass the repeat image mode here [1], but even in that case, CG doesn't seem to sample outside the bounds or Skia does? I have a patch that passes the extend mode from layout to [1], which I think is right but I also want to figure out what's going on with the repeat mode again between skia and CG. This is just a note for me to ask tomorrow.

[1] https://dxr.mozilla.org/mozilla-central/source/gfx/thebes/gfxUtils.cpp?from=gfxUtils.cpp#556
See Also: → 1221840
Attached patch repeat.patchSplinter Review
This was happening because of a repeat resampling artifact. CG doesn't support repeat resampling, chrome does some weird combination of repeat resampling and strict tiling, Edge supports repeat resampling. We enable repeat resampling in Skia. This patch passes down the repeat mode from CSS down to [1] when creating the scaled image and when repeating the scaled image.

[1] https://dxr.mozilla.org/mozilla-central/source/gfx/thebes/gfxUtils.cpp?from=gfxUtils.cpp#556
Attachment #8748754 - Flags: review?(jmuizelaar)
Attachment #8748754 - Flags: review?(jmuizelaar) → review+
https://hg.mozilla.org/mozilla-central/rev/bc653ab527f5
Status: NEW → RESOLVED
Closed: 8 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla49
Comment on attachment 8748754 [details] [diff] [review]
repeat.patch

Approval Request Comment
[Feature/regressing bug #]: Skia content on OS X, bug 1207332
[User impact if declined]: Can have a line through some websites, see screenshot.
[Describe test coverage new/current, TreeHerder]: Manual
[Risks and why]: Low, only affects rare cases where we repeat a scaled image multiple times.
[String/UUID change made/needed]: None
Attachment #8748754 - Flags: approval-mozilla-aurora?
Comment on attachment 8748754 [details] [diff] [review]
repeat.patch

Recent regression, manual testing for fix on m-c. Let's uplift and verify the fix
Attachment #8748754 - Flags: approval-mozilla-aurora? → approval-mozilla-aurora+
Flags: qe-verify+
Depends on: 1274726
I tried to reproduce this issue on Firefox 48.0a1 (2016-03-23) and on Firefox 48.0a1 (2016-04-22), using the STR from Comment 0, without success.
The tests were performed on Windows 10 x64, Mac OS X 10.11.1 and on Ubuntu 16.04 x64.
I've also tried with the gfx.content.azure.backends pref changed to skia.
Sören, are you still able to reproduce this issue on the affected builds?
If the issue is reproducible on you're side, can you please provide more information about this issue (STR, build under test and OS)?
Flags: needinfo?(cadeyrn)
The only step to reproduce the issue is to visit the website from comment 0. I can still reproduce the issue in older builds and verify that it's fixed in current builds.

Apple OS X 10.11.6 Beta (15G7b), MacBook Pro with Retina display.
Flags: needinfo?(cadeyrn)
Thanks Sören for the information.
I managed to reproduce this issue on MacBook Pro with Retina displayed and on Firefox 48.0a1 (2016-03-24). It seems that only on this device the issue is reproducible.
I confirm that the issue is no longer reproducible on Firefox 48.0b1, Firefox 49.0a2 (2016-06-14) or on Firefox 50.0a1 (2016-06-13).
The tests were performed on MacBook Pro OS X 10.11.2, Windows 10 x64 and on Ubuntu 16.04 x64.
Status: RESOLVED → VERIFIED
Flags: qe-verify+
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: