Close the created popup widget on blurring a WebContents regardless of whether the renderer process…

Close the created popup widget on blurring a WebContents regardless of whether the renderer process is responsive

BUG=Go to https://www.alternativ-testing.fr/operaclickjackselect/index1.htm,
drag the link to a new tab, then go to the second tab and open HTML Select.
After a while, the first page will take focus and clicking on what remains
of the HTML select popup will crash the application.

[Mikus, 2014-04-22 12:11:05]

Ben, are you there?

[Ben Goodger (Google), 2014-04-22 23:55:56]

Did you see if it was possible to write a test for this? How are our js modal
dialogs tested?

[Mikus, 2014-04-23 11:46:54]

On 2014/04/22 23:55:56, Ben Goodger (Google) wrote:
> Did you see if it was possible to write a test for this? How are our js modal
> dialogs tested?

I guess it is, will do that.

[Mikus, 2014-04-23 11:46:56]

On 2014/04/22 23:55:56, Ben Goodger (Google) wrote:
> Did you see if it was possible to write a test for this? How are our js modal
> dialogs tested?

I guess it is, will do that.

[Mikus, 2014-04-24 07:10:52]

On 2014/04/23 11:46:56, Mikus wrote:
> On 2014/04/22 23:55:56, Ben Goodger (Google) wrote:
> > Did you see if it was possible to write a test for this? How are our js
modal
> > dialogs tested?
> 
> I guess it is, will do that.

I've been looking for someplace to test this and I cannot think of any viable
solution. I've tried with web_contents_impl_browsertest.cc but trying to create
a widget (web_contents_impl->CreateNewWidget) DCHECKs on surface_id_ check
(render_widget_host_impl.cc:215), so possibly some completely new test
infrastructure, including backing renderer, would need to be created to even
make this testable. Any suggestions / comments?

[Ben Goodger (Google), 2014-04-24 16:29:57]

On 2014/04/24 07:10:52, Mikus wrote:
> On 2014/04/23 11:46:56, Mikus wrote:
> > On 2014/04/22 23:55:56, Ben Goodger (Google) wrote:
> > > Did you see if it was possible to write a test for this? How are our js
> modal
> > > dialogs tested?
> > 
> > I guess it is, will do that.
> 
> I've been looking for someplace to test this and I cannot think of any viable
> solution. I've tried with web_contents_impl_browsertest.cc but trying to
create
> a widget (web_contents_impl->CreateNewWidget) DCHECKs on surface_id_ check
> (render_widget_host_impl.cc:215), so possibly some completely new test
> infrastructure, including backing renderer, would need to be created to even
> make this testable. Any suggestions / comments?

Convoluted, but you could possibly construct a content browsertest for this
that:

- creates two WebContents
- has a test page in both similar to the test case
- listens for the select to be opened in the second (probably from C++?)
- when that happens, calls a js function in the former that pops the js alert. I
think WC has a function to execute script in the page. In the first tab you'll
define an entrypoint function ShowDialog() that you can call from C++.
- targets a click on the popup at this point

Can you explore this a bit more? This is a really subtle behavior/crash so
having a regression test seems desirable.

[Mikus, 2014-04-25 07:34:12]

On 2014/04/24 16:29:57, Ben Goodger (Google) wrote:
> On 2014/04/24 07:10:52, Mikus wrote:
> > On 2014/04/23 11:46:56, Mikus wrote:
> > > On 2014/04/22 23:55:56, Ben Goodger (Google) wrote:
> > > > Did you see if it was possible to write a test for this? How are our js
> > modal
> > > > dialogs tested?
> > > 
> > > I guess it is, will do that.
> > 
> > I've been looking for someplace to test this and I cannot think of any
viable
> > solution. I've tried with web_contents_impl_browsertest.cc but trying to
> create
> > a widget (web_contents_impl->CreateNewWidget) DCHECKs on surface_id_ check
> > (render_widget_host_impl.cc:215), so possibly some completely new test
> > infrastructure, including backing renderer, would need to be created to even
> > make this testable. Any suggestions / comments?
> 
> Convoluted, but you could possibly construct a content browsertest for this
> that:
> 
> - creates two WebContents
> - has a test page in both similar to the test case
> - listens for the select to be opened in the second (probably from C++?)
> - when that happens, calls a js function in the former that pops the js alert.
I
> think WC has a function to execute script in the page. In the first tab you'll
> define an entrypoint function ShowDialog() that you can call from C++.
> - targets a click on the popup at this point
> 
> Can you explore this a bit more? This is a really subtle behavior/crash so
> having a regression test seems desirable.

Sure.