GitLab

Before this patch:
- We issue a CaptureDump callback from the main frame that has finished
  a test. If a secondary renderer issues a test finished, this routes
  via a browser and goes back to the main frame anyway.
- We issue several IPCs to the browser for each type of dump requested.
- After everything is done, we issue a test done IPC to the browser.

With this patch:
- We capture local dumps (temporarily*)
- We request that a browser initiates a CaptureDump via IPC.
- We get a mojo call to capture the dump with an output struct containing
  all the relevant fields, which the renderer populates from local dumps.
- When it finishes populating everything, it issues a callback with that
  struct.
- The browser then records all of the results from the struct and
  finishes.

The new approach seems to be a bit better in that we can move more logic
to the browser part and make the renderer part more synchronous. For
example, if the browser needs to initiate a layout dump, we can just do
that from the browser instead of requiring that the renderer sends an IPC
to the browser with that request.

Furthermore, because the end goal of this is to have OOPIF pixel dumps,
we need those to be initiated from the browser at some point. This also
aligns with the pattern where the browser is the part that initiates
a capture dump.

* - We would like to capture as little as possible locally and instead
    move things that capture things by asking the browser to the browser
    directly.

R=lukasza@chromium.org, mkwst@chromium.org

Bug: 667551
Change-Id: I07d2b16da687b679f6b3020732706e6bd7a77cd4
Reviewed-on: https://chromium-review.googlesource.com/972578
Commit-Queue: vmpstr <vmpstr@chromium.org>
Reviewed-by: Antoine Labour <piman@chromium.org>
Reviewed-by: Mike West <mkwst@chromium.org>
Reviewed-by: Łukasz Anforowicz <lukasza@chromium.org>
Cr-Commit-Position: refs/heads/master@{#545880}