Lazy background pages now load in response to message passing.

Lazy background pages now load in response to message passing.

I refactored the pending event stuff so that it handles generic tasks. The
first enqueued task for an extension will start its lazy background page, and
tasks are run once the page finishes loading. Events and messages now share
this mechanism so that either one can activate the page.

BUG=81752
TEST=no

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=127216

[Matt Perry, 2012-03-14 23:09:36]

+miranda for chrome/browser/profiles OWNERS review. Note that the change to
Profile is temporary, until the fix for
http://code.google.com/p/chromium/issues/detail?id=104095 can be checked in.

[Miranda Callahan, 2012-03-15 00:12:32]

LGTM, with comment addressed. cc-ing erg so he knows what's up.

http://codereview.chromium.org/9704031/diff/3002/chrome/browser/profiles/prof...
File chrome/browser/profiles/profile_impl.cc (right):

http://codereview.chromium.org/9704031/diff/3002/chrome/browser/profiles/prof...
chrome/browser/profiles/profile_impl.cc:142: COMPILE_ASSERT(sizeof(ProfileImpl)
<= 720u, profile_impl_size_unexpected);
I think you're going to regress this assertion -- if this is temporary, that's
fine. But you'll need to change this number -- and ping erg about this.

[Matt Perry, 2012-03-15 20:08:32]

http://codereview.chromium.org/9704031/diff/3002/chrome/browser/profiles/prof...
File chrome/browser/profiles/profile_impl.cc (right):

http://codereview.chromium.org/9704031/diff/3002/chrome/browser/profiles/prof...
chrome/browser/profiles/profile_impl.cc:142: COMPILE_ASSERT(sizeof(ProfileImpl)
<= 720u, profile_impl_size_unexpected);
On 2012/03/15 00:12:32, Miranda Callahan wrote:
> I think you're going to regress this assertion -- if this is temporary, that's
> fine. But you'll need to change this number -- and ping erg about this.

The assertion doesn't fire for me... Maybe people have removed bits without
updating this number.

[Yoyo Zhou, 2012-03-15 23:56:02]

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
File chrome/browser/extensions/extension_message_service.cc (right):

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
chrome/browser/extensions/extension_message_service.cc:188:
ExtensionMessageService::ExtensionMessageService(Profile* profile)
Just pass in the LazyBackgroundTaskQueue instead.

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
chrome/browser/extensions/extension_message_service.cc:236: return;
Can this ever happen?

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
chrome/browser/extensions/extension_message_service.cc:443: profile =
profile->GetOriginalProfile();
I'm not fully convinced this is needed.

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
File chrome/browser/extensions/extension_message_service.h (right):

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
chrome/browser/extensions/extension_message_service.h:128: void
PendingCloseChannel(int port_id, ExtensionHost*) {
Why is there an extra unused argument here? It doesn't appear to be passed in
where it's Bound. Same in PendingPostMessage.

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/la...
File chrome/browser/extensions/lazy_background_task_queue.cc (right):

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/la...
chrome/browser/extensions/lazy_background_task_queue.cc:80: // Balance the
keepalive in AddPendingTask.
Should this be inside the for loop at 72?

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/la...
chrome/browser/extensions/lazy_background_task_queue.cc:94: if
(host->extension_host_type() ==
We still need a Profile check on the source. Since this observes all sources,
when profile 2's extension's lazy background page stops loading, profile 1 will
try to process it and presumably hit the NOTREACHED at 67.

[Matt Perry, 2012-03-16 00:39:30]

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
File chrome/browser/extensions/extension_message_service.cc (right):

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
chrome/browser/extensions/extension_message_service.cc:188:
ExtensionMessageService::ExtensionMessageService(Profile* profile)
On 2012/03/15 23:56:03, Yoyo Zhou wrote:
> Just pass in the LazyBackgroundTaskQueue instead.

Done.

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
chrome/browser/extensions/extension_message_service.cc:236: return;
On 2012/03/15 23:56:03, Yoyo Zhou wrote:
> Can this ever happen?

Yes. This code starts the background page in response to a message from a
content script. (See the api test.)

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
chrome/browser/extensions/extension_message_service.cc:443: profile =
profile->GetOriginalProfile();
On 2012/03/15 23:56:03, Yoyo Zhou wrote:
> I'm not fully convinced this is needed.

I assure you it is :). If we don't do this, then the PendingChannel will be
registered with the incognito profile in line 448. Then when we later post a
message, we use that same profile to add the PendingPostMessage task in line
377. Since it's the incognito profile, it doesn't run in response to the regular
bg page load, so it never gets executed.

(ditto for PendingCloseChannel)

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
File chrome/browser/extensions/extension_message_service.h (right):

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
chrome/browser/extensions/extension_message_service.h:128: void
PendingCloseChannel(int port_id, ExtensionHost*) {
On 2012/03/15 23:56:03, Yoyo Zhou wrote:
> Why is there an extra unused argument here? It doesn't appear to be passed in
> where it's Bound. Same in PendingPostMessage.

These are curried callback functions. The lazy background task queue passes in
the ExtensionHost. PendingCloseChannel and PendingPostMessage don't use the host
param, though. I'll add a comment.

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/la...
File chrome/browser/extensions/lazy_background_task_queue.cc (right):

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/la...
chrome/browser/extensions/lazy_background_task_queue.cc:80: // Balance the
keepalive in AddPendingTask.
On 2012/03/15 23:56:03, Yoyo Zhou wrote:
> Should this be inside the for loop at 72?

No, we only increment the count when adding the first task, so we decrement when
removing the last.

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/la...
chrome/browser/extensions/lazy_background_task_queue.cc:94: if
(host->extension_host_type() ==
On 2012/03/15 23:56:03, Yoyo Zhou wrote:
> We still need a Profile check on the source. Since this observes all sources,
> when profile 2's extension's lazy background page stops loading, profile 1
will
> try to process it and presumably hit the NOTREACHED at 67.

Good catch. Done.

[Yoyo Zhou, 2012-03-16 01:04:55]

LGTM, just 1 more comment.

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
File chrome/browser/extensions/extension_message_service.cc (right):

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
chrome/browser/extensions/extension_message_service.cc:236: return;
On 2012/03/16 00:39:30, Matt Perry wrote:
> On 2012/03/15 23:56:03, Yoyo Zhou wrote:
> > Can this ever happen?
> 
> Yes. This code starts the background page in response to a message from a
> content script. (See the api test.)

Hmm, I think I meant to ask if MaybeAddPendingOpenChannelTask could ever return
false, i.e. the process is not there but it's not a lazy background page. But it
seems like this is handled in OpenChannelImpl.

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
File chrome/browser/extensions/extension_message_service.h (right):

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
chrome/browser/extensions/extension_message_service.h:128: void
PendingCloseChannel(int port_id, ExtensionHost*) {
On 2012/03/16 00:39:30, Matt Perry wrote:
> On 2012/03/15 23:56:03, Yoyo Zhou wrote:
> > Why is there an extra unused argument here? It doesn't appear to be passed
in
> > where it's Bound. Same in PendingPostMessage.
> 
> These are curried callback functions. The lazy background task queue passes in
> the ExtensionHost. PendingCloseChannel and PendingPostMessage don't use the
host
> param, though. I'll add a comment.

Hey, that makes sense; I hadn't seen it actually being used in Chromium yet.

http://codereview.chromium.org/9704031/diff/7005/chrome/browser/extensions/la...
File chrome/browser/extensions/lazy_background_task_queue.h (right):

http://codereview.chromium.org/9704031/diff/7005/chrome/browser/extensions/la...
chrome/browser/extensions/lazy_background_task_queue.h:51: virtual void
Observe(int type,
This is public in NotificationObserver, so it should be public here too.

[Matt Perry, 2012-03-16 01:10:31]

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
File chrome/browser/extensions/extension_message_service.cc (right):

http://codereview.chromium.org/9704031/diff/5006/chrome/browser/extensions/ex...
chrome/browser/extensions/extension_message_service.cc:236: return;
On 2012/03/16 01:04:55, Yoyo Zhou wrote:
> On 2012/03/16 00:39:30, Matt Perry wrote:
> > On 2012/03/15 23:56:03, Yoyo Zhou wrote:
> > > Can this ever happen?
> > 
> > Yes. This code starts the background page in response to a message from a
> > content script. (See the api test.)
> 
> Hmm, I think I meant to ask if MaybeAddPendingOpenChannelTask could ever
return
> false, i.e. the process is not there but it's not a lazy background page. But
it
> seems like this is handled in OpenChannelImpl.

Ah, yes. It returns false when the extension does not have a background page
(lazy or otherwise). This might happen, for example, if a content script is
trying to send a message to a popup. In that case, we want to call through to
OpenChannelImpl and send the disconnect back.

http://codereview.chromium.org/9704031/diff/7005/chrome/browser/extensions/la...
File chrome/browser/extensions/lazy_background_task_queue.h (right):

http://codereview.chromium.org/9704031/diff/7005/chrome/browser/extensions/la...
chrome/browser/extensions/lazy_background_task_queue.h:51: virtual void
Observe(int type,
On 2012/03/16 01:04:55, Yoyo Zhou wrote:
> This is public in NotificationObserver, so it should be public here too.

It's a matter of style, but I prefer putting interface implementations in the
private section when they should only be called via the interface. There is
precedence for this elsewhere in Chrome.