695108 - Android-only sporadic "REFTEST TEST-UNEXPECTED-FAIL | reftests/svg/filter-extref-differentOrigin-01.svg | image comparison (==)"

Status: RESOLVED FIXED

(Core :: SVG, defect)

Description

[Daniel Holbert [:dholbert]]

I added the test "filter-extref-differentOrigin-01.svg" last week, as part of Bug 686013.  It checks that external SVG filters from a different origin will not load.

Correct rendering (no filter) is green.
Bad rendering (filter applied) is black.

Today, it failed sporadically (twice in a row) on Android R2:
https://tbpl.mozilla.org/php/getParsedLog.php?id=6892206&tree=Firefox
https://tbpl.mozilla.org/php/getParsedLog.php?id=6893270&tree=Firefox

This makes sense, because the reftest is served over file:// on Desktop (where url(../whatever) is different-origin, but over http on Android (where url(../whatever is same-origin).

To fix this, we need to:
 - mark it as fails-if(Android)
 - add a <use> element that pulls in the external resource, so that it *is* accessible, we'll load it synchronously.  (this will make it perma-"fail" on Android instead of sporadic-fail)

Comment 1

[Daniel Holbert [:dholbert]]

I'll fix this up, as described at the end of Comment 0.

Comment 2

[Boris Zbarsky [:bzbarsky]]

Or make the test always be an HTTP test and use an actual cross-origin load?

Comment 3

[Daniel Holbert [:dholbert]]

AFAIK, our reftest harness doesn't support cross-origin for HTTP tests.

However, as bz points out on IRC, we could convert this into a mochitest and test actual cross-origin loads there, since the mochitest harness has multiple fake domains available for use.

I think I'll fix this up per comment 0 to stop the randomorange, and then file a followup on adding an additional mochitest for "real" cross-origin loads.

Comment 4

[Daniel Holbert [:dholbert]]

Attachment: patch: add <use> and flag as fails-on-android

Comment 5

[Daniel Holbert [:dholbert]]

(running this through TryServer right now)

Comment 6

[Daniel Holbert [:dholbert]]

This passed TryServer:
  https://tbpl.mozilla.org/?tree=Try&rev=949912e60891

I did 6 runs of Android R2, for good measure, and the 5 of those that are complete were all successful, modulo known-randomorange.  (They failed this bug's tweaked reftest, as expected & asserted in the reftest manifest).

Reftest-runs were green on Linux & Linux64, too.

Comment 7

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 567590 [details] [diff] [review]
patch: add <use> and flag as fails-on-android

r=me but maybe make the bug# comment point to the followup bug you file?

Comment 8

[Daniel Holbert [:dholbert]]

Filed bug 695385 on adding a mochitest, and updated the reftest.list comment to point to that bug instead of this one.

Landed: https://hg.mozilla.org/integration/mozilla-inbound/rev/3fe4c90c9e1f

Comment 9

[Marco Bonardo [:mak]]

https://hg.mozilla.org/mozilla-central/rev/3fe4c90c9e1f

I assume this can be mark fixed, if not please reopen.