Closed Bug 1363945 Opened 7 years ago Closed 7 years ago

Intermittent forms/text-control-baseline-1.html == forms/text-control-baseline-1-ref.html | image comparison, max difference: 2, number of differing pixels: 1

Categories

(Core :: Layout: Form Controls, defect)

Product:
Core
Component:
Layout: Form Controls
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

()

Status:
RESOLVED FIXED
Milestone:
mozilla58
Tracking Flags:
Tracking Status
firefox57 --- fixed
firefox58 --- fixed

People

(Reporter: intermittent-bug-filer, Unassigned)

References

Details

(Keywords: intermittent-failure, Whiteboard: [fennec-scouting][stockwell fixed:other])

Whiteboard: [fennec-scouting]
android/debug is the primary root cause here:
https://brasstacks.mozilla.com/orangefactor/?display=Bug&bugid=1363945

rates have picked up, but not so concerning.
failure rate is on android debug 95% of the time, the rate has dropped in recent times luckily
Whiteboard: [fennec-scouting] → [fennec-scouting][stockwell unknown]
Flags: needinfo?(gbrown)
Failures here have been an on-going, low-frequency problem for months, on Android only.

Beginning with the changes for bug 1381434, Android failures became near perma-fail:

https://treeherder.mozilla.org/#/jobs?repo=autoland&filter-searchStr=android%20reftest&tochange=f6d9016ac75aee2d94a3a6e4ad53ca6326d672a6&fromchange=9aa27ded4a4e1360813b3af5e6aeaae1094ad597

:schien -- Do you understand how your changes affected this test? Can you make the test more reliable on Android, or can we address this with fuzzy-if()?
Flags: needinfo?(gbrown) → needinfo?(schien)
Blocks: 1381434
See Also: → 1363940
As in bug 1363945, we should just annotate this with fuzzy-if().
Pushed by gbrown@mozilla.com:
https://hg.mozilla.org/integration/mozilla-inbound/rev/5676964b1a22
Mark text-control-baseline-1.html as fuzzy on Android; r=me,test-only
Pushed by gbrown@mozilla.com:
https://hg.mozilla.org/integration/mozilla-inbound/rev/652b1c36693c
Follow-up: really mark text-control-baseline-1.html as fuzzy on Android, r=me,test-only
Whiteboard: [fennec-scouting][stockwell disable-recommended] → [fennec-scouting][stockwell fixed:other]
https://hg.mozilla.org/mozilla-central/rev/5676964b1a22
https://hg.mozilla.org/mozilla-central/rev/652b1c36693c
Status: NEW → RESOLVED
Closed: 7 years ago
status-firefox58: --- → fixed
Resolution: --- → FIXED
Target Milestone: --- → mozilla58
The pixel difference is always on the top-left most corner of the first input or text area. I did a try experiment that make text-control-baseline-1-ref.html to load exact the same content as text-control-baseline-1.html and it still shows the pixel difference. The result looks weird to me because we have different rendering result while rendering the same HTML content.
https://treeherder.mozilla.org/#/jobs?repo=try&revision=dec048b968a91f50ad6939afd92a650f66d41353&selectedJob=135095209

The RGB value difference of that pixel is about 1 or 2 and this pixel is on the edge of the text area. My guess is that this delta is caused by anti-aliasing or alpha blending which doing floating point operations. @dholbert do you thinks it worth to dig in more on this bug?
Flags: needinfo?(schien) → needinfo?(dholbert)
(In reply to Shih-Chiang Chien [:schien] (UTC+8) (use ni? plz) from comment #34)
> The pixel difference is always on the top-left most corner of the first
> input or text area. I did a try experiment that make
> text-control-baseline-1-ref.html to load exact the same content as
> text-control-baseline-1.html and it still shows the pixel difference. The
> result looks weird to me because we have different rendering result while
> rendering the same HTML content.

Yeah, there's some non-determinacy here, so the same content might load differently.  As jfkthame noted in bug 1363940 comment 27, the OMT font load might be influencing the number of times we layout/paint, or something like that.

> My guess is that this delta is caused by
> anti-aliasing or alpha blending which doing floating point operations.

Makes sense. Fortunately, it's extremely subtle and not user-perceivable (and only on a single platform).

> @dholbert do you thinks it worth to dig in more on this bug?

I don't think so; I think we can consider this dealt with.
Flags: needinfo?(dholbert)
If so, the fuzzy-if() sounds like a good solution to deal with this intermittent failure.
Great!  (And, looks like gbrown took care of landing that already, in comment 29.)
You need to log in before you can comment on or make changes to this bug.