commit 7178f5c1387636c7e5dfec90a533bcadb5c0c0c1
author Philip Rogers <pdr@chromium.org> Mon Jul 09 18:52:59 2018
committer Commit Bot <commit-bot@chromium.org> Mon Jul 09 18:52:59 2018
tree 3c3c2232d4adec99f0703046a1e0685be7c3c939
parent d0855ce5d8e550e243803522771e4c7ed5aec81a

Revert "Transmit the response's body datapipe in CommitNavigation()."

This reverts commit 7521deccbb7d3d4a7700e6e007e679a8706d8be0.

Reason for revert: Caused failures on fast/history/history-back-initial-vs-final-url.html (and possibly other tests, see https://crbug.com/861808)

Original change's description:
> Transmit the response's body datapipe in CommitNavigation().
> 
> # Summary of the current behavior:
> 
> When a navigation occurs, the mojo::URLLoaderClient interface is first
> bound to the content::NavigationURLLoaderImpl in the browser process.
> When the browser knows in which renderer process the navigation will
> commit, the interface is bound to the content::URLLoaderClientImpl in
> the renderer process to continue the navigation. The switching from one
> to another happens when mojo::URLLoaderClient::OnReceiveResponse() is
> called. This is described here: https://goo.gl/Rrrc7n
> 
> # What are we trying to do:
> 
> After the mojo::URLLoaderClient interface is unbound in the browser
> process and bound again in the renderer process, it takes some amount of
> time to receive URLLoader::OnStartLoadingResponseBody(). It may be worth
> sending the data pipe with RenderFrameHostImpl::CommitNavigation() and
> give it to the renderer process immediatly instead of waiting for it. It
> would allow the Renderer process to get the main resource's data
> earlier, especially when the renderer process is busy.
> 
> Note: It doesn't look like there are any strong reasons to make
>       URLLoaderClient::OnReceiveResponse(response_head) and
>       URLLoaderClient::OnStartLoadingResponseBody(response_body)
>       separate messages, maybe at some point, they could be merged into:
>       URLLoaderClient::OnReceiveResponse(response_head, response_body)
> 
> # What this CL does:
> 
> This CL makes the switch to happen in
> URLLoaderClient::OnStartLoadingResponseBody(). The data pipe is
> transmitted in CommitNavigation().
> 
> This is enabled/disabled using the NavigationImmediateResponse experiment
> The goal is to look for performance improvement or regressions on
> Canary.
> 
> Bug: 831155
> Change-Id: Id6cf667fdc1482baf27f41aab754e58d9a5a569b
> Reviewed-on: https://chromium-review.googlesource.com/1100830
> Reviewed-by: Yutaka Hirano <yhirano@chromium.org>
> Reviewed-by: Peter Kasting <pkasting@chromium.org>
> Reviewed-by: Kinuko Yasuda <kinuko@chromium.org>
> Commit-Queue: Arthur Sonzogni <arthursonzogni@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#573276}

TBR=pkasting@chromium.org,kinuko@chromium.org,nasko@chromium.org,yhirano@chromium.org,clamy@chromium.org,arthursonzogni@chromium.org

Change-Id: I22eac2d660f9cf1d8128331b547c304af729a85a
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: 831155
Reviewed-on: https://chromium-review.googlesource.com/1129659
Reviewed-by: Philip Rogers <pdr@chromium.org>
Commit-Queue: Philip Rogers <pdr@chromium.org>
Cr-Commit-Position: refs/heads/master@{#573392}