In variant /2 we use cc layers as the source of generating hit test regions, I'll investigate into this bug.

Also I think we can unblock the issue, /2 variant is hidden --use-viz-hit-testing-surface-layer, which is by default is disabled.

/2 variant does not submit (4, 5) because it is completely covered by the other iframe. We put a logic in cc hit test data provider that eliminates completely covered surfaces.

I can confirm that if we change the position of "frame2" to left: 30px in page_with_two_iframes.html(https://cs.chromium.org/chromium/src/content/test/data/frame_tree/page_with_two_iframes.html?q=/frame_tree/page_with_two_iframes.html&sq=package:chromium&g=0&l=1), we will also submit (4, 5) in variant /2.

jonross@, can you take a look at this and confirm that?

That does lead to submitting (4, 5) in variant /2, while continuing to submit it in variant /1.

Looking at that page_with_two_iframes.html test they are creating completely overlapping iframes. Is it expected that this would lead to one not submitting hit test data?

If so this does expose a flaw in the testing strategy of  issue 785986 . As if an obscured region never has its submitted hit test data provided to the browser, then how can one detect that the submission occurred? Though alternatively it could be that tests shouldn't wait on hit test data from obscured regions.

I think in the long term, we want to roll out /2 logic for performance issues. I personally think variant /2 should have different checks in test from /1.

In the future when we generate hit test data based on blink's knowledge, we will experience more differences between current draw quad hit testing and "paint" hit testing. I think the solution is to not wait for obscured region if /2 flag is enabled?

Xianda, can the iframe on the top specify that it doesn't want some type of events, so that we have to target the iframe on the bottom for those events? Basically I'm concerned if eliminating obscured surfaces going to give us correct results in all cases

Afaik in cc, if a layer is hit testable it accepts both types of events. (I could be wrong here)

I think in v3 hit testing we'll focus on that problem. Before that is implemented I can certainly submit those regions.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/79f35d0e90605e72245b8c1ea5d560e5f9e0b6ab

commit 79f35d0e90605e72245b8c1ea5d560e5f9e0b6ab
Author: jonross <jonross@chromium.org>
Date: Tue Jun 19 15:00:40 2018

Update SitePerProcessHitTestBrowserTest.HitTestStaleDataDeletedView hit test waiting

Currently SitePerProcessHitTestBrowserTest.HitTestStaleDataDeletedView is using
a legacy method for waiting for hit test data. This change updates it to use
the new HitTestRegionObserver.

However there is a bug in the /2 variant of the test, there overlapping hit test
regions are not submitted. Due to this I have it exit early in that config.

TEST=SitePerProcessHitTestBrowserTest.HitTestStaleDataDeletedView

Bug:  846798 
Change-Id: I410d493f1abc7334e4cc21c3c4edceb4e2edd8c9
Reviewed-on: https://chromium-review.googlesource.com/1102721
Reviewed-by: Ken Buchanan <kenrb@chromium.org>
Commit-Queue: Jonathan Ross <jonross@chromium.org>
Cr-Commit-Position: refs/heads/master@{#568447}
[modify] https://crrev.com/79f35d0e90605e72245b8c1ea5d560e5f9e0b6ab/content/browser/site_per_process_hit_test_browsertest.cc
[modify] https://crrev.com/79f35d0e90605e72245b8c1ea5d560e5f9e0b6ab/testing/buildbot/filters/viz.content_browsertests.filter

Reproduced today ahead of triage

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/bd4ae48dae4a5c62c9513d0080301d2baffebdfa

commit bd4ae48dae4a5c62c9513d0080301d2baffebdfa
Author: Xianda Sun <sunxd@chromium.org>
Date: Wed Oct 31 18:06:11 2018

VizHitTest: Do not skip OOPIF even if it's occluded

We previously employed an optimization that skips OOPIFs occluded by
parent frame's layers. This optimization is problematic as the layer's
bounds do not necessarily represent its hit test region, and it works
incorrectly when the occluding layers have pointer-events property.

This CL removes that optimization.

Bug:  846798 , 896786
Change-Id: Id33c7d658cfd677ab3eb496371a213de2423d349
Reviewed-on: https://chromium-review.googlesource.com/c/1296821
Reviewed-by: Ken Buchanan <kenrb@chromium.org>
Reviewed-by: Robert Flack <flackr@chromium.org>
Commit-Queue: Xianda Sun <sunxd@chromium.org>
Cr-Commit-Position: refs/heads/master@{#604329}
[modify] https://crrev.com/bd4ae48dae4a5c62c9513d0080301d2baffebdfa/cc/trees/layer_tree_host_impl.cc
[modify] https://crrev.com/bd4ae48dae4a5c62c9513d0080301d2baffebdfa/content/browser/site_per_process_hit_test_browsertest.cc
[add] https://crrev.com/bd4ae48dae4a5c62c9513d0080301d2baffebdfa/content/test/data/frame_tree/page_with_occluded_iframes.html