GitLab

We currently have a C++ views object owning an Cocoa object which then owns a
bridge back into some other C++ interfaces. We can simplify this by having the
top-level C++ object handle those interfaces and leave the Cocoa object to
purely manage the SFChooseIdentityPanel.

This doesn't fix the https://crbug.com/987744 but is some cleanup along the
way. Before we add more cases to that logic, reduce the number of places we
spread that logic out.

This does result in some RunUntilIdle() calls in the tests no longer quite
waiting long enough, I suspect because the Cocoa event loop is involved. I've
switched them to specifically wait for the relevant object to be destroyed,
which seems more robust in general. It seems that was also the cause of the
lifetime mismatch in https://crbug.com/1041175 so now we don't need a WeakPtr.

Unfortunately the diff here is kind of hard to read. I had to reorder some of
the classes so they could refer to each other. (Note the old bridge object was
similarly interleaved between the @interface and @implementation.)

Bug: 987744, 1041175
Change-Id: Icccc9e18cdfc64444a6496748dc106be6233896a
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1978946
Commit-Queue: David Benjamin <davidben@chromium.org>
Reviewed-by: Ryan Sleevi <rsleevi@chromium.org>
Reviewed-by: Elly Fong-Jones <ellyjones@chromium.org>
Cr-Commit-Position: refs/heads/master@{#734237}