It looks like this wasn't a transient problem, it also showed up in some of the mac failures in https://chromium-review.googlesource.com/c/chromium/src/+/1223149/3. But not exactly the same constellation. It's very strange. Trying again since I had to make some other unrelated changes.

Philip, I think your investigation is on the right path.

rebaseline-cl (or more specifically, optimize-baselines in this case) is WAI. The weird placement of baselines is due to mac10_11-blink-rel, mac10_12-blink-rel and mac10_13-blink-rel being different from others.

This is what the fallback tree looks like. I use "*" and "+" to represent the two different baselines, and "|" means this platform does not have a baseline (i.e. falls back to its parent).

* fuschia
* win
    | linux
+ mac
  * mac-retina
  | mac10.12
    | mac10.11
      * mac10.10

Yeah, I also suspsec that rebaseline-cl is correct here. Why some of the mac bots had different results is the mystery. szager@, any hypothesis of why that might be?

This might be related to an issue I brought up on blink-dev recently:

https://groups.google.com/a/chromium.org/d/msg/blink-dev/ivkXlBB3-XY/XkGdH8HiBwAJ

Basically, there are some tests in LayoutTests/intersection-observer/v2 that do this:

if (window.internals && internals.runtimeFlags) {
  internals.runtimeFlags.intersectionObserverV2Enabled = true;
}

If that js runs before the IntersectionObserver and IntersectionObserverEntry binding initialize code runs, then those extra attributes are added to the type templates.

Because the layout test runner re-uses content_shell instances to run multiple tests, it may be that the global namespace was polluted because one of the intersection-observer/v2 tests ran before the global-interface-listing test.

Here is the recent CL that added the tests:

https://chromium-review.googlesource.com/c/chromium/src/+/1212452

Oh, that would make sense. I thought that settings were reset between tests, but maybe not? Can you add cleanup code that resets the state back to the original?

Yeah, I'm working on just such a patch.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/be431e79c6ecdc6a097e90f9defa077a0af7e2af

commit be431e79c6ecdc6a097e90f9defa077a0af7e2af
Author: Stefan Zager <szager@chromium.org>
Date: Fri Sep 14 21:39:33 2018

[IntersectionObserverV2] Move all tests to virtual test suite

Previously, IOv2 layout tests would enable IOv2 in script via:

if (window.internals && internals.runtimeFlags) {
  internals.runtimeFlags.intersectionObserverV2Enabled true;
}

There are two problems with this approach:

1. By the time the above code runs, it may be too late: the prototype
for the IntersectionObserverEntry javascript type may already have been
created *without* the isVisible attribute. This was causing flaky test
failures, because the exact timing of when the prototype is created may
vary between runs.

2. Because the layout test runner re-uses content_shell processes for
multiple tests, the side effects from turning on the feature flag
affect tests that later use the same renderer process (see bug).

This CL also adds DisableIntersectionObserverV2ThrottleForTesting,
which eliminates the need to add a bunch of 100ms timers to the tests.
Slow tests are bad.

Finally, this CL reorders the handshake for cross-origin tests that
rely on postMessage, to ensure that both the embedding frame and the
child frame are properly initialized before starting the test
procedure.

BUG= 882280 , 883676 
R=chrishtr@chromium.org,foolip@chromium.org

Change-Id: I18e62c2cf099bf74ef4afc43f46a794c7a322582
Reviewed-on: https://chromium-review.googlesource.com/1220044
Commit-Queue: Stefan Zager <szager@chromium.org>
Reviewed-by: Chris Harrelson <chrishtr@chromium.org>
Cr-Commit-Position: refs/heads/master@{#591483}
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/TestExpectations
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/VirtualTestSuites
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/http/tests/intersection-observer/resources/v2-subframe.html
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/http/tests/intersection-observer/v2/cross-origin-effects.html
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/http/tests/intersection-observer/v2/cross-origin-occlusion.html
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/intersection-observer/resources/intersection-observer-test-utils.js
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/intersection-observer/v2/animated-occlusion.html
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/intersection-observer/v2/box-shadow.html
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/intersection-observer/v2/iframe-target.html
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/intersection-observer/v2/simple-effects.html
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/intersection-observer/v2/simple-occlusion.html
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/intersection-observer/v2/text-shadow.html
[rename] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/virtual/intersection-observer-v2/http/tests/intersection-observer/v2/README.txt
[copy] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/WebKit/LayoutTests/virtual/intersection-observer-v2/intersection-observer/v2/README.txt
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/blink/renderer/core/testing/internals.cc
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/blink/renderer/core/testing/internals.h
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/blink/renderer/core/testing/internals.idl
[modify] https://crrev.com/be431e79c6ecdc6a097e90f9defa077a0af7e2af/third_party/blink/renderer/platform/runtime_enabled_features.json5

rebaseline-cl was actually working as expected here, closing.