Fix a regression where extension APIs would sometimes not be available on the

Fix a regression where extension APIs would sometimes not be available on the
web store.

BUG=86996
TEST=see bug for repro

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=90125

[Matt Perry, 2011-06-22 00:07:53]

jam: OWNERS approval
creis: please review

[Charlie Reis, 2011-06-22 00:52:31]

Is it possible to test this (e.g., in AppApiTest)?  Seems like we could kill an
app/extension process, reload it, and make sure the new one thinks it's an
extension process.

http://codereview.chromium.org/7211029/diff/2001/content/browser/renderer_hos...
File content/browser/renderer_host/render_process_host.h (right):

http://codereview.chromium.org/7211029/diff/2001/content/browser/renderer_hos...
content/browser/renderer_host/render_process_host.h:168: virtual void
PrepareForReuse() = 0;
Since we always call this (even on the first use), should we call it something
like PrepareForUse()?

http://codereview.chromium.org/7211029/diff/2001/content/browser/renderer_hos...
File content/browser/renderer_host/render_view_host.cc (right):

http://codereview.chromium.org/7211029/diff/2001/content/browser/renderer_hos...
content/browser/renderer_host/render_view_host.cc:114:
process()->PrepareForReuse();
Just as a sanity check, is it definitely the case that we want to support
sending messages to RPH before Init is called?  If so, then I think this is
right.  Just wanted to be sure it wasn't cleaner to wait until Init before
sending messages.

[jam, 2011-06-22 00:52:47]

is is_initialized_ still needed?

[Charlie Reis, 2011-06-22 00:54:33]

On 2011/06/22 00:52:47, John Abd-El-Malek wrote:
> is is_initialized_ still needed?

Yes-- we're concerned about messages that might be sent to a dead renderer
(e.g., after a crash) getting queued up and sent to the new renderer.  That's
why we moved away from Patch Set 1.

[jam, 2011-06-22 01:17:06]

rubber stamp lgtm

[Matt Perry, 2011-06-22 21:14:12]

I tried creating an api test, but I couldn't replicate whatever is happening on
the web store. I'm not sure what sequence of events results in the RPH being
reused. It doesn't seem to reproduce consistently manually either.

http://codereview.chromium.org/7211029/diff/2001/content/browser/renderer_hos...
File content/browser/renderer_host/render_process_host.h (right):

http://codereview.chromium.org/7211029/diff/2001/content/browser/renderer_hos...
content/browser/renderer_host/render_process_host.h:168: virtual void
PrepareForReuse() = 0;
On 2011/06/22 00:52:31, creis wrote:
> Since we always call this (even on the first use), should we call it something
> like PrepareForUse()?

I renamed to EnableSendQueue to reflect what it actually does.

http://codereview.chromium.org/7211029/diff/2001/content/browser/renderer_hos...
File content/browser/renderer_host/render_view_host.cc (right):

http://codereview.chromium.org/7211029/diff/2001/content/browser/renderer_hos...
content/browser/renderer_host/render_view_host.cc:114:
process()->PrepareForReuse();
On 2011/06/22 00:52:31, creis wrote:
> Just as a sanity check, is it definitely the case that we want to support
> sending messages to RPH before Init is called?  If so, then I think this is
> right.  Just wanted to be sure it wasn't cleaner to wait until Init before
> sending messages.

I think so. is_initialized_ was originally added expressly for this purpose (we
just didn't reset it properly).

[Charlie Reis, 2011-06-22 22:20:48]

Thanks-- LGTM.

http://codereview.chromium.org/7211029/diff/7001/content/browser/renderer_hos...
File content/browser/renderer_host/render_process_host.h (right):

http://codereview.chromium.org/7211029/diff/7001/content/browser/renderer_hos...
content/browser/renderer_host/render_process_host.h:165: // Call this allow
queueing of IPC messages that are sent before the process
Nit: Call this to allow