The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/2c36561594b11b0b35b4294f59fe8e639670b6df

commit 2c36561594b11b0b35b4294f59fe8e639670b6df
Author: johnme <johnme@chromium.org>
Date: Fri Oct 21 17:41:36 2016

Mark css3/filters/effect-reference-after.html flaky

BUG=658305
TBR=senorblanco@chromium.org
NOTRY=true
NOTREECHECKS=true

Review-Url: https://chromiumcodereview.appspot.com/2439143002
Cr-Commit-Position: refs/heads/master@{#426834}

[modify] https://crrev.com/2c36561594b11b0b35b4294f59fe8e639670b6df/third_party/WebKit/LayoutTests/TestExpectations

There appear to be a non-trivial amount of "almost" flakes among the css3/filters/effect-reference* tests:

http://test-results.appspot.com/dashboards/flakiness_dashboard.html#tests=css3%2Ffilters%2Feffect-reference

Of course hard to say if these all are related in some way, but thought I'd mention it.

(Been looking at css3/filters/effect-reference-delete.html for a bit for unrelated reasons, and there I can occasionally repro locally. Will see if I manage to find something out.)

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/3353a7a6003262aae1515ac9541a519ec23c8c1c

commit 3353a7a6003262aae1515ac9541a519ec23c8c1c
Author: fs <fs@opera.com>
Date: Mon Oct 31 10:34:39 2016

Make sure to always reset the cached filter in ReferenceFilterOperation

BUG=658305
CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2

Review-Url: https://codereview.chromium.org/2453033004
Cr-Commit-Position: refs/heads/master@{#428678}

[modify] https://crrev.com/3353a7a6003262aae1515ac9541a519ec23c8c1c/third_party/WebKit/Source/core/paint/FilterEffectBuilder.cpp

After looking at this test some more it seems like a sequence like the following leads to the flaky failing result:

1) The entire document is parsed (or, at least enough for the filter to be located )
2) A layout (et.c) happens before image dimensions has arrived, so we compute a filter chain with a reference box of 0x0. Hence the primitive region will also be 0x0.
3) Image dimensions (and the image data itself) is ready (loaded), but the filter chain is not invalidated, and hence there's no output.

The above can be verified by setting 'filterUnits' to userSpaceOnUse and explicitly specifying dimensions for the filter region (if not, the the default values of 120% will resolve against the 0x0 viewport of the <svg>.)

This issue is blocking  issue 603396  (mojo-loading) as with mojo the failure rate increases. See [1] and [2] for example.

I have a more deterministic test case. Apply [3] (a test-only change), run the http server (Tools/Scripts/run-blink-httpd) and access to the file (http://localhost:8080/css/css-filter.html). The expected behavior is that two boxes appear after two seconds pass. I can see the expected behavior with Chrome 54.0.2840.100 (the current stable) but only one box appear with ToT (aa5cbb1a6f3500536feec0c97acdd72e1c56dbb4).

senorblanco@, is there any status update?

1: https://codereview.chromium.org/2553153002/
2: https://build.chromium.org/p/tryserver.chromium.linux/builders/linux_chromium_rel_ng/builds/351291
3: https://codereview.chromium.org/2555353003/

I haven't had time to look into this further. fs@, do you think you could take another look?

I bisected based on the info in c#5, and landed at fbeebf2905898067b4def80b945aacabe9d413e9. However that Cl would've just made this issue more visible because then we actually started to clip to the filter region/primitive subregion (which both are 0x0 in this case.)

Per c#4, are we missing an invalidation when the image is updated?

Yes, of the (stored) filter chain (which is size-dependent.) The fix is made a tad bit more complex (I think) because of how the size used to resolve the filter chain is the size including descendants.

I'll add flaky tests to TestExpectations.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/c269a43c7619112f4096f27a9b5d5312103bed33

commit c269a43c7619112f4096f27a9b5d5312103bed33
Author: yhirano <yhirano@chromium.org>
Date: Tue Dec 13 13:27:54 2016

Mark layout tests using 'filter: url(...)' as flaky

BUG=658305
R=fs@opera.com
NOTRY=true

Review-Url: https://codereview.chromium.org/2567113004
Cr-Commit-Position: refs/heads/master@{#438154}

[modify] https://crrev.com/c269a43c7619112f4096f27a9b5d5312103bed33/third_party/WebKit/LayoutTests/TestExpectations

 Issue 663581  has been merged into this issue.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/7469f74a88dea8965e646230fc716d4dcd0c0e4e

commit 7469f74a88dea8965e646230fc716d4dcd0c0e4e
Author: yhirano <yhirano@chromium.org>
Date: Tue Jan 24 08:36:53 2017

Mark css3/filters/effect-reference-external-stylesheet.html as flaky

BUG=658305
R=fs@opera.com
NOTRY=true

Review-Url: https://codereview.chromium.org/2654713003
Cr-Commit-Position: refs/heads/master@{#445681}

[modify] https://crrev.com/7469f74a88dea8965e646230fc716d4dcd0c0e4e/third_party/WebKit/LayoutTests/TestExpectations

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/b034403ab50b3a27bda35d6cf72efec57f2f1217

commit b034403ab50b3a27bda35d6cf72efec57f2f1217
Author: qyearsley <qyearsley@chromium.org>
Date: Tue Feb 07 17:50:47 2017

Remove flaky expectations with update-test-expectations script

The update-test-expectations script is meant to remove
flaky expectations when it appears that according to the
most recent results on the flakiness dashboard for that test,
the test is no longer flaky.

BUG= 491764 , 498539 , 575766 , 600248 ,658305, 663838 , 663840 ,663848, 663851 , 663853 , 663872 , 663874 , 663877 , 663879 ,664817,664839, 664840 , 664841 , 664842 , 664846 , 664850 , 664855 , 664856 , 666991 , 671480 , 671618 , 672204 , 673296 , 673632 , 674720 ,674858, 678488 , 683800 

Review-Url: https://codereview.chromium.org/2679543005
Cr-Commit-Position: refs/heads/master@{#448663}

[modify] https://crrev.com/b034403ab50b3a27bda35d6cf72efec57f2f1217/third_party/WebKit/LayoutTests/TestExpectations

This is not a P1 issue. All of the css3/filters/effect-reference... tests are flaky according to the TestExpectations.

Just to summarize, 26 tests are now marked PASS/FAIL, which means they're effectively getting no test coverage.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/97d33b7704eb40400fbce5e10512dea9bde839e1

commit 97d33b7704eb40400fbce5e10512dea9bde839e1
Author: Vitalii Iarko <vitaliii@chromium.org>
Date: Wed Aug 23 13:55:33 2017

Ignore effect-reference-colorspace-hw.html timeouts because flaky.

TBR=shend@chromium.org

Bug: 658305,758133
Change-Id: I4fb6e2fd2a076eddd4d74b897c1dd5ba79b36c9b
Reviewed-on: https://chromium-review.googlesource.com/628736
Reviewed-by: vitaliii <vitaliii@chromium.org>
Commit-Queue: vitaliii <vitaliii@chromium.org>
Cr-Commit-Position: refs/heads/master@{#496673}
[modify] https://crrev.com/97d33b7704eb40400fbce5e10512dea9bde839e1/third_party/WebKit/LayoutTests/TestExpectations

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/f5df9c50a89feb7769b5847d8fa81e1c4e9ffbb8

commit f5df9c50a89feb7769b5847d8fa81e1c4e9ffbb8
Author: Friedrich Horschig <fhorschig@chromium.org>
Date: Wed Oct 11 14:43:46 2017

Mark css3/filters/effect-reference-subregion-hw.html as Timeout

This flaky test is starting to fail due to timeouts. Accept this until
the underlying issue (i.e. shader logic) is fixed.

TBR=yhirano@chromium.org, fs@opera.com

Bug: 773648, 658305
Change-Id: Ibb248836d068f5c1b38a5a312aa9f64836c537f5
Reviewed-on: https://chromium-review.googlesource.com/712714
Commit-Queue: Friedrich Horschig <fhorschig@chromium.org>
Reviewed-by: Friedrich Horschig <fhorschig@chromium.org>
Cr-Commit-Position: refs/heads/master@{#507980}
[modify] https://crrev.com/f5df9c50a89feb7769b5847d8fa81e1c4e9ffbb8/third_party/WebKit/LayoutTests/TestExpectations

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/0c6213902cc35b2c9c954bafdf61a844a94cbe8c

commit 0c6213902cc35b2c9c954bafdf61a844a94cbe8c
Author: Reilly Grant <reillyg@chromium.org>
Date: Tue Oct 17 23:05:50 2017

Mark css3/filters/effect-reference-subregion-hw.html as Timeout

This flaky test is starting to fail due to timeouts. Accept this until
the underlying issue (i.e. shader logic) is fixed.

The original change (r507980) attempting to mark this test as Timeout
accidentally changed the line for
css3/filters/effect-reference-subregion.html instead.

TBR=yhirano@chromium.org, fs@opera.com
NOTRY=true

Bug: 773648, 658305
Change-Id: Ie52945d5158037e600ac24a04533d6d0efaecec9
Reviewed-on: https://chromium-review.googlesource.com/724224
Reviewed-by: Reilly Grant <reillyg@chromium.org>
Commit-Queue: Reilly Grant <reillyg@chromium.org>
Cr-Commit-Position: refs/heads/master@{#509579}
[modify] https://crrev.com/0c6213902cc35b2c9c954bafdf61a844a94cbe8c/third_party/WebKit/LayoutTests/TestExpectations

Apologies, applied the wrong component in bulk.

 Issue 797823  has been merged into this issue.