Open Bug 1126530 Opened 9 years ago Updated 2 years ago

On Win8, reftest "image-object-position-with-background-2.html" renders with gradient of fuzzy pixels at the bottom of each page, if imgFrame::Optimize is disabled

Categories

(Core :: Graphics: ImageLib, defect)

x86_64
Linux
defect

Tracking

()

People

(Reporter: dholbert, Unassigned)

References

Details

(Whiteboard: [gfx-noted])

Attachments

(1 file)

Seth pushed a Try run with gDisableOptimize set to "true" in imgFrame:
  https://tbpl.mozilla.org/?tree=Try&rev=a619a8707963

(This was experimental; we should only depend on this optimization for performance, but he's looking into whether we depend on it for correctness as well.)

This Try run had failures for "image-object-position-with-background-2.html", on Windows 8 x64. (both opt + debug).  Win7 & WinXP are all fine. (I'm not sure whether the relevant difference is the Windows version or the 64bit-ness.)

The failure looks like a real bug. The reftest has an opaque image that's split across several pages, and the testcase's reftest-snapshot shows the image & its background gradually blending together at the bottom of each page (with the image fading out & the background fading in). This blending is smeared across the bottom 4-5px of the image's content-box on each page.  I'm attaching the reftest log for this test-failure, for use with reftest-analyzer.xhtml.

I'm guessing what's going on is that we render the image into some buffer, with an antialiased bottom-edge for some reason (so, the bottom edge of pixels in this buffer is semitransparent) -- and then when we scale that up to render, this bottom row of semitransparent pixels get scaled up to several rows.

Anyway, this is a bug. We should fix it.
Attachment #8555501 - Attachment description: reftest-log.txt → reftest log for failure (w/ data URIs)
[ni=seth to tie this to the bug on making this imgFrame optimization pref-controllable, if/when he files that bug]
Flags: needinfo?(seth)
Flags: needinfo?(seth)
See Also: → 1126163
Depends on: 1126634
I talked to Bas about this issue and he suggested that this bug may be specific to the D2D 1.1 backend, which apparently we're only using on the Windows 8 reftest machines right now.
BTW, this is probably in reality a graphics bug, but I think leaving it in this component is fine until someone has time to investigate and identify the specific issue.
See Also: → 1128229
Depends on: 1128229
See Also: 1128229
Severity: normal → S3
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: