629143 - Investigate OS X 64 perma-orange on reftests img-height-meet-2.html, img-height-slice-2.html, img-width-meet-2.html, img-width-slice-2.html

Status: RESOLVED FIXED

(Core :: SVG, defect)

Description

[Daniel Holbert [:dholbert]]

Since earlier this morning, these tests have been perma-orange on OSX 64:

REFTEST TEST-UNEXPECTED-FAIL |
file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/svg/as-image/img-height-meet-2.html
| image comparison (==)
REFTEST TEST-UNEXPECTED-FAIL |
file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/svg/as-image/img-height-slice-2.html
| image comparison (==)
REFTEST TEST-UNEXPECTED-FAIL |
file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/svg/as-image/img-width-meet-2.html
| image comparison (==)
REFTEST TEST-UNEXPECTED-FAIL |
file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/svg/as-image/img-width-slice-2.html
| image comparison (==)

Log attached.

Description of what reftest-analyzer shows:
It's a single "block" in the grid that gets mispositioned in the *reference* case.  It's supposed to have a circle aligned with the left edge of the block, but instead, the circle is chopped in half, with the right half being left-aligned, and the left-half being right-aligned.

Comment 1

[Daniel Holbert [:dholbert]]

Attachment: reftest failure log

Comment 2

[Daniel Holbert [:dholbert]]

The first changeset I see with these failures is:
 http://hg.mozilla.org/mozilla-central/rev/a5f732abf109
which is clearly unrelated. (linux-only UI-styling change.)

I also noticed a few OS X 64 cycles since then that have *not* had this these failures -- so this isn't quite perma-orange, but it's near perma-orange.

I think I'm going to mark these tests as random on mac for now, pending some investigation, to stop the orange...

Comment 3

[Daniel Holbert [:dholbert]]

marked as random on our osx 64 tinderboxen:
http://hg.mozilla.org/mozilla-central/rev/0e7b88440e17

Comment 4

[Daniel Holbert [:dholbert]]

Attachment: testcase

(In reply to comment #0)
> It's a single "block" in the grid that gets mispositioned in the *reference*
> case.

So it's an <embed src="data:image/svg+xml[...]" height="20">
where the embedded SVG document has...
  viewBox="0 0 40 20 " preserveAspectRatio="xMidYMid meet"

I'm attaching a file that I *think* is the specific chunk of the reference case that's misrendered here.  On machines that hit this problem, I'd expect them to have the same problem when viewing this file.  (Trying to get access to an affected RelEng mac to test that theory...)

Comment 5

[Daniel Holbert [:dholbert]]

Hmmmm - so joe's got access to a RelEng mac that should be hitting this problem, and he took screenshots of the attached testcase as well as one of the busted reftest files.[1]

Both loaded correctly for him (no bug -- the circle was intact).  So I don't know what's going on.

So for now, I'm going to hold off on this bug, and watch the progress on bug 629016 (which probably is the same underlying issue as this bug).

[1] http://hg.mozilla.org/mozilla-central/raw-file/tip/layout/reftests/svg/as-image/img-width-meet-2-ref.html

Comment 6

[Joe Drew (not getting mail)]

Daniel, we've upgraded the test machines to 10.6.6, so you can push a build to try with the random-if removed to see if the tests pass now.

Comment 7

[Daniel Holbert [:dholbert]]

Thanks - I'm on it.

Comment 8

[Daniel Holbert [:dholbert]]

looks good! all green.  I even requested 2 bonus reftest runs for OS X64 Debug, since that's the tbpl row that was most orange (according to the mis-starrings of this on bug 607003), and those were green too.

So we can revert comment 3's test-disabling-push and close this bug.  I'll do that soon.

Comment 9

[Daniel Holbert [:dholbert]]

Re-enabled tests: http://hg.mozilla.org/mozilla-central/rev/7b4e4cbd95de