Fix for swapping in prerendered RenderViews with provisional data sources

The RenderView currently only updates the navigation state
for the currently displayed data source in
OnDisplayPrerenderedPage.  If there's a provisional data
source, it still thinks its being prerendered when it
becomes the main data source.

This CL fixes that issue.  To do this, it moves the
|is_prerendering_| flag over to the RenderView.  While
not a strictly necessary design change, things are a bit
less bug-prone this way.

Also changes how |was_started_as_prerender| is set.

BUG=73173
TEST=none, yet
(manual test:  Go to http://peter.sh and refresh it a few times, with prerender enabled and a debug build.  Notice lack of DCHECK assertion in OnDisplayPrerenderedPage)

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=75903

[mmenke, 2011-02-17 22:16:12]

I'm not quite sure how to create a browser test for this without having some
sort of a race condition.

The TestServer can be used to create a page that doesnt' send and data for X
seconds, but depending on events on a timer being slower than passing a couple
messages back and forth between processes doesn't seem like a good way to make
robust tests.

The new code could theoretically result in some prerendered pages being
double-counted in the histograms, but think it's the best solution, for now. 
Also don't think it will be that common.

More problematic is what to do for histograms in the case of html redirects, an
issue I'm punting on for the moment.

[cbentzel, 2011-02-18 13:56:47]

LGTM

What are the cases where double-counting will happen?

http://codereview.chromium.org/6532031/diff/8001/chrome/renderer/render_view.cc
File chrome/renderer/render_view.cc (right):

http://codereview.chromium.org/6532031/diff/8001/chrome/renderer/render_view....
chrome/renderer/render_view.cc:4715: WebDataSource* povisional_ds =
provisional_ds - although at least the misspelling was consistent

[mmenke, 2011-02-18 14:53:04]

On 2011/02/18 13:56:47, cbentzel wrote:
> LGTM
> 
> What are the cases where double-counting will happen?

It depends on when "WebKit::WebFrameClient::willClose" is called (by webkit).  I
believe prerendered pages with HTML/Javascript redirects/refreshes that are
swapped into a TabContents after the redirect/refresh has started, but before
any of the body of the response page has been received will be double counted -
once for the page being swapped out, and once for the page being swapped in.

http://codereview.chromium.org/6532031/diff/8001/chrome/renderer/render_view.cc
File chrome/renderer/render_view.cc (right):

http://codereview.chromium.org/6532031/diff/8001/chrome/renderer/render_view....
chrome/renderer/render_view.cc:4715: WebDataSource* povisional_ds =
On 2011/02/18 13:56:47, cbentzel wrote:
> provisional_ds - although at least the misspelling was consistent

Fixed, though I prefer my more creative spelling.

[jam, 2011-02-18 23:07:19]

I don't understand this code, but changes to RV are fine to me.

[mmenke, 2011-02-18 23:20:20]

On 2011/02/18 23:07:19, John Abd-El-Malek wrote:
> I don't understand this code, but changes to RV are fine to me.

When a navigation starts, a provisional data source with its own navigation
state is created.  The data source for the displayed page (Even if it's a new
RenderView, with a blank page) stays around, with its own NavigationState until
we receive some data from the provisional data source.

I wasn't correctly handling the case when a prerendered page was first displayed
at a point in time where it had a provisional data source (In addition to the
one currently in use).  This fixes that problem.

[mmenke, 2011-02-18 23:41:02]

On 2011/02/18 23:20:20, Matt Menke wrote:
> On 2011/02/18 23:07:19, John Abd-El-Malek wrote:
> > I don't understand this code, but changes to RV are fine to me.
> 
> When a navigation starts, a provisional data source with its own navigation
> state is created.  The data source for the displayed page (Even if it's a new
> RenderView, with a blank page) stays around, with its own NavigationState
until
> we receive some data from the provisional data source.
> 
> I wasn't correctly handling the case when a prerendered page was first
displayed
> at a point in time where it had a provisional data source (In addition to the
> one currently in use).  This fixes that problem.

To clarify - I was only updating the current page's NavigationState, rather than
the provision page's one, which always had the |is_prerendering| flag set.  The
main frame's data source can also have the flag set after the first navigation. 
So life is just much simpler with the flag as part of the RenderView itself
instead of the NavigationState for a frame.

[jam, 2011-02-23 17:44:38]

thanks for the explanation :)