Add the ability to unload the HistoryBackend....

Add the ability to unload the HistoryBackend.

A small number of places used accessors like in_memory_url_database() or backend_loaded() with the expectation that if they weren't already functional, the history system was in the process of making them so.  I elected to make both of these functions that triggered lazy backend initialization.

BUG=23400
TEST=none

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=28461

[brettw, 2009-10-08 00:47:51]

http://codereview.chromium.org/267019/diff/1/10
File chrome/browser/history/history.cc (right):

http://codereview.chromium.org/267019/diff/1/10#newcode222
Line 222: LoadBackendIfNecessary();
I'm not clear what this call has to do with getting the in memory database.

http://codereview.chromium.org/267019/diff/1/10#newcode311
Line 311: LoadBackendIfNecessary();
Isn't this covered by ScheduleAndFOrget?

http://codereview.chromium.org/267019/diff/1/9
File chrome/browser/history/history.h (right):

http://codereview.chromium.org/267019/diff/1/9#newcode108
Line 108: bool BackendLoaded() {
Can you put the impl of this in the .cc file?

http://codereview.chromium.org/267019/diff/1/3
File chrome/browser/profile.h (right):

http://codereview.chromium.org/267019/diff/1/3#newcode200
Line 200: virtual void UnloadHistoryBackend() = 0;
This seems like an odd function to have on the Profile? Can your caller just do
what the implementation of this function does?

[Peter Kasting, 2009-10-08 00:59:35]

Take another look.

http://codereview.chromium.org/267019/diff/1/10
File chrome/browser/history/history.cc (right):

http://codereview.chromium.org/267019/diff/1/10#newcode222
Line 222: LoadBackendIfNecessary();
On 2009/10/08 00:47:51, brettw wrote:
> I'm not clear what this call has to do with getting the in memory database.

The Backend is what actually sets up the in_memory_backend_.

http://codereview.chromium.org/267019/diff/1/9
File chrome/browser/history/history.h (right):

http://codereview.chromium.org/267019/diff/1/9#newcode108
Line 108: bool BackendLoaded() {
On 2009/10/08 00:47:51, brettw wrote:
> Can you put the impl of this in the .cc file?

Yes, good catch

http://codereview.chromium.org/267019/diff/1/3
File chrome/browser/profile.h (right):

http://codereview.chromium.org/267019/diff/1/3#newcode200
Line 200: virtual void UnloadHistoryBackend() = 0;
On 2009/10/08 00:47:51, brettw wrote:
> This seems like an odd function to have on the Profile? Can your caller just
do
> what the implementation of this function does?

Yes; the implementation changed as I was writing the patch and it makes more
sense to do it that way now.

[brettw, 2009-10-08 01:22:01]

http://codereview.chromium.org/267019/diff/1/10
File chrome/browser/history/history.cc (right):

http://codereview.chromium.org/267019/diff/1/10#newcode222
Line 222: LoadBackendIfNecessary();
On 2009/10/08 00:59:35, Peter Kasting wrote:
> On 2009/10/08 00:47:51, brettw wrote:
> > I'm not clear what this call has to do with getting the in memory database.
> 
> The Backend is what actually sets up the in_memory_backend_.

But that happens asynchronously on another thread and asynchronously comes back
to this thread. It's guaranteed to have no effect on the return value of this
function.

[Peter Kasting, 2009-10-08 17:06:46]

http://codereview.chromium.org/267019/diff/1/10
File chrome/browser/history/history.cc (right):

http://codereview.chromium.org/267019/diff/1/10#newcode222
Line 222: LoadBackendIfNecessary();
On 2009/10/08 01:22:01, brettw wrote:
> On 2009/10/08 00:59:35, Peter Kasting wrote:
> > The Backend is what actually sets up the in_memory_backend_.
> 
> But that happens asynchronously on another thread and asynchronously comes
back
> to this thread. It's guaranteed to have no effect on the return value of this
> function.

I know.  Similarly, BackendLoaded() starts the backend setting up even though
that also cannot affect the return value.  In both cases, the reason for this is
that callers of these expect that the history system is in the process of
setting up the backend if it hasn't finished doing that already, so they call
these functions and, when they return NULL, wait around hoping that the result
will change later.  Since I broke that expectation by unloading the backend, I
need to make these calls trigger a "hey, someone is interested in the backend,
go start setting it up" behavior.

[brettw, 2009-10-08 21:17:14]

I guess this is OK. Can you add a comment to the in memory getter with a brief
version of what you wrote here. It's the kind of thing I would delete if I came
across it later since it doesn't do anything.

LGTM