When clearing plugin data at shutdown, wait for it to finish.

When clearing plugin data at shutdown, wait for it to finish.

Also, Add some browser tests for clearing plugin data and enable npapi_test_plugin on 64-bit Linux; apparently it's been fixed. 

BUG=58235, 18337, 26625
TEST=PluginDataRemoverTest.*

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=72123

[Bernhard Bauer, 2011-01-13 17:27:08]

Please review. This is the patch we discussed in December that avoids spinning
up the MessageLoop after the main MessageLoop has quit. I also added some UMA
stats to find out how long we wait for the plugin process.

[jam, 2011-01-13 19:52:43]

http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
File chrome/browser/plugin_data_remover.h (right):

http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
chrome/browser/plugin_data_remover.h:77: bool has_browser_ref_;
comment?

[Bernhard Bauer, 2011-01-14 15:20:39]

On 2011/01/13 19:52:43, John Abd-El-Malek wrote:
>
http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
> File chrome/browser/plugin_data_remover.h (right):
> 
>
http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
> chrome/browser/plugin_data_remover.h:77: bool has_browser_ref_;
> comment?

It's gone now :-) Turns out I don't really need to keep the UI message loop from
quitting, as nothing happens there anyway.

[Bernhard Bauer, 2011-01-14 16:32:14]

Now, just to make sure I'm not doing something stupid here: I still
keep doing stuff on the IO thread (opening a connection to the plugin
process, sending the ClearSiteData message) while I'm on the UI thread
outside of the MessageLoop and waiting, but that's okay, right? (As an
aside, I guess I could just do the waiting with a WaitableEvent
instead…)

On Fri, Jan 14, 2011 at 16:20,  <[email protected]> wrote:
> On 2011/01/13 19:52:43, John Abd-El-Malek wrote:
>
>
http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
>>
>> File chrome/browser/plugin_data_remover.h (right):
>
>
>
http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
>>
>> chrome/browser/plugin_data_remover.h:77: bool has_browser_ref_;
>> comment?
>
> It's gone now :-) Turns out I don't really need to keep the UI message loop
> from
> quitting, as nothing happens there anyway.
>
> http://codereview.chromium.org/6308001/
>

[jam, 2011-01-14 18:03:00]

Hmm, the browser code does assume that the UI thread outlives the IO thread
(see browser_thread.h).  Is the MessageLoop of the UI thread still alive,
even if it's not pumping?  If not, that would probably cause crashes because
it would violate the assumptions of BrowserThread.

On Fri, Jan 14, 2011 at 8:31 AM, Bernhard Bauer <[email protected]> wrote:

> Now, just to make sure I'm not doing something stupid here: I still
> keep doing stuff on the IO thread (opening a connection to the plugin
> process, sending the ClearSiteData message) while I'm on the UI thread
> outside of the MessageLoop and waiting, but that's okay, right? (As an
> aside, I guess I could just do the waiting with a WaitableEvent
> instead…)
>
> On Fri, Jan 14, 2011 at 16:20,  <[email protected]> wrote:
> > On 2011/01/13 19:52:43, John Abd-El-Malek wrote:
> >
> >
>
http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
> >>
> >> File chrome/browser/plugin_data_remover.h (right):
> >
> >
> >
>
http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
> >>
> >> chrome/browser/plugin_data_remover.h:77: bool has_browser_ref_;
> >> comment?
> >
> > It's gone now :-) Turns out I don't really need to keep the UI message
> loop
> > from
> > quitting, as nothing happens there anyway.
> >
> > http://codereview.chromium.org/6308001/
> >
>

[Bernhard Bauer, 2011-01-14 18:10:57]

On Friday, January 14, 2011, John Abd-El-Malek <[email protected]> wrote:
> Hmm, the browser code does assume that the UI thread outlives the IO thread
(see browser_thread.h).  Is the MessageLoop of the UI thread still alive, even
if it's not pumping?  If not, that would probably cause crashes because it would
violate the assumptions of BrowserThread.

The UI thread does outlive the IO thread, it's just its MessageLoop
that quits (but stays alive, so it basically eats tasks).


>
> On Fri, Jan 14, 2011 at 8:31 AM, Bernhard Bauer <[email protected]> wrote:
>
> Now, just to make sure I'm not doing something stupid here: I still
> keep doing stuff on the IO thread (opening a connection to the plugin
> process, sending the ClearSiteData message) while I'm on the UI thread
> outside of the MessageLoop and waiting, but that's okay, right? (As an
> aside, I guess I could just do the waiting with a WaitableEvent
> instead…)
>
> On Fri, Jan 14, 2011 at 16:20,  <[email protected]> wrote:
>> On 2011/01/13 19:52:43, John Abd-El-Malek wrote:
>>
>>
http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
>>>
>>> File chrome/browser/plugin_data_remover.h (right):
>>
>>
>>
http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
>>>
>>> chrome/browser/plugin_data_remover.h:77: bool has_browser_ref_;
>>> comment?
>>
>> It's gone now :-) Turns out I don't really need to keep the UI message loop
>> from
>> quitting, as nothing happens there anyway.
>>
>> http://codereview.chromium.org/6308001/
>>
>
>

[Bernhard Bauer, 2011-01-18 21:17:26]

Ping?

FWIW, I have another version of this CL that just uses a
WaitableEvent, which seems to work as well, but I'm not sure if there
was a specific reason why we chose to prefer waiting for the process
instead.

On Friday, January 14, 2011, Bernhard Bauer <[email protected]> wrote:
> On Friday, January 14, 2011, John Abd-El-Malek <[email protected]> wrote:
>> Hmm, the browser code does assume that the UI thread outlives the IO thread
(see browser_thread.h).  Is the MessageLoop of the UI thread still alive, even
if it's not pumping?  If not, that would probably cause crashes because it would
violate the assumptions of BrowserThread.
>
> The UI thread does outlive the IO thread, it's just its MessageLoop
> that quits (but stays alive, so it basically eats tasks).
>
>
>>
>> On Fri, Jan 14, 2011 at 8:31 AM, Bernhard Bauer <[email protected]> wrote:
>>
>> Now, just to make sure I'm not doing something stupid here: I still
>> keep doing stuff on the IO thread (opening a connection to the plugin
>> process, sending the ClearSiteData message) while I'm on the UI thread
>> outside of the MessageLoop and waiting, but that's okay, right? (As an
>> aside, I guess I could just do the waiting with a WaitableEvent
>> instead…)
>>
>> On Fri, Jan 14, 2011 at 16:20,  <[email protected]> wrote:
>>> On 2011/01/13 19:52:43, John Abd-El-Malek wrote:
>>>
>>>
http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
>>>>
>>>> File chrome/browser/plugin_data_remover.h (right):
>>>
>>>
>>>
http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
>>>>
>>>> chrome/browser/plugin_data_remover.h:77: bool has_browser_ref_;
>>>> comment?
>>>
>>> It's gone now :-) Turns out I don't really need to keep the UI message loop
>>> from
>>> quitting, as nothing happens there anyway.
>>>
>>> http://codereview.chromium.org/6308001/
>>>
>>
>>
>

[Bernhard Bauer, 2011-01-18 21:19:14]

Ping?

FWIW, I have another version of this CL that just uses a
WaitableEvent, which seems to work as well, but I'm not sure if there
was a specific reason why we chose to prefer waiting for the process
instead.

On Friday, January 14, 2011, Bernhard Bauer <[email protected]> wrote:
> On Friday, January 14, 2011, John Abd-El-Malek <[email protected]> wrote:
>> Hmm, the browser code does assume that the UI thread outlives the IO thread
(see browser_thread.h).  Is the MessageLoop of the UI thread still alive, even
if it's not pumping?  If not, that would probably cause crashes because it would
violate the assumptions of BrowserThread.
>
> The UI thread does outlive the IO thread, it's just its MessageLoop
> that quits (but stays alive, so it basically eats tasks).
>
>
>>
>> On Fri, Jan 14, 2011 at 8:31 AM, Bernhard Bauer <[email protected]> wrote:
>>
>> Now, just to make sure I'm not doing something stupid here: I still
>> keep doing stuff on the IO thread (opening a connection to the plugin
>> process, sending the ClearSiteData message) while I'm on the UI thread
>> outside of the MessageLoop and waiting, but that's okay, right? (As an
>> aside, I guess I could just do the waiting with a WaitableEvent
>> instead…)
>>
>> On Fri, Jan 14, 2011 at 16:20,  <[email protected]> wrote:
>>> On 2011/01/13 19:52:43, John Abd-El-Malek wrote:
>>>
>>>
http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
>>>>
>>>> File chrome/browser/plugin_data_remover.h (right):
>>>
>>>
>>>
http://codereview.chromium.org/6308001/diff/4001/chrome/browser/plugin_data_r...
>>>>
>>>> chrome/browser/plugin_data_remover.h:77: bool has_browser_ref_;
>>>> comment?
>>>
>>> It's gone now :-) Turns out I don't really need to keep the UI message loop
>>> from
>>> quitting, as nothing happens there anyway.
>>>
>>> http://codereview.chromium.org/6308001/
>>>
>>
>>
>

[Bernhard Bauer, 2011-01-20 18:01:56]

After adding some browser tests, there came up timing issues with waiting for
the plugin process to exit, so I changed it to use WaitableEvent instead. 

Could you take another look? Thanks!

[jam, 2011-01-20 18:34:15]

lgtm

sorry for the delay, I didn't see the old emails (in the future feel free to
ping me over IM when I miss them :) )

http://codereview.chromium.org/6308001/diff/27001/chrome/browser/browser_proc...
File chrome/browser/browser_process.h (right):

http://codereview.chromium.org/6308001/diff/27001/chrome/browser/browser_proc...
chrome/browser/browser_process.h:203: std::string
plugin_data_remover_mime_type_;
this needs a comment.  by convention we don't put the comment beside the simple
setter/getters, and put them by the variable name