Bug 116595 – REGRESSION(r150491): WebKit2.DOMWindowExtensionNoCache fails

• Status: NEW
• Component: WebKit2

[Ryosuke Niwa]

WebKit2.DOMWindowExtensionNoCache started failing after this revision:
http://build.webkit.org/builders/Apple%20MountainLion%20Debug%20WK2%20%28Tests%29/builds/10020

Note: Google Test filter = WebKit2.DOMWindowExtensionNoCache
[==========] Running 1 test from 1 test case.
[----------] Global test environment set-up.
[----------] 1 test from WebKit2
[ RUN      ] WebKit2.DOMWindowExtensionNoCache
LEAK: 29 WebCoreNode
LEAK: 5 CachedResource
Timeout

[Ryosuke Niwa]

Confirmed. It passes on r150490 but times out on r150491.

[Ryosuke Niwa]

I've tried to add add a call to WKTerminate but that broke the test. Now we wouldn't get the last 3 messages:

Extension states:
First page, main frame, standard world - Destroyed
First page, main frame, non-standard world - Destroyed
First page, subframe, standard world - Destroyed
First page, subframe, non-standard world - Destroyed
Second page, main frame, standard world - Destroyed
Second page, main frame, non-standard world - Destroyed
First page, main frame, standard world - Connected
First page, main frame, non-standard world - Connected
First page, subframe, standard world - Connected
First page, subframe, non-standard world - Connected

Extension states:
First page, main frame, standard world - Destroyed
First page, main frame, non-standard world - Destroyed
First page, subframe, standard world - Destroyed
First page, subframe, non-standard world - Destroyed
Second page, main frame, standard world - Destroyed
Second page, main frame, non-standard world - Destroyed
First page, main frame, standard world - Removed
First page, main frame, non-standard world - Removed
First page, subframe, standard world - Removed
First page, subframe, non-standard world - Removed

TestComplete

[Ryosuke Niwa]

Committed r150564: <http://trac.webkit.org/changeset/150564>

[Andreas Kling]

Created attachment 202838 [details]
Tweak the test for new WKPageClose semantics

[Ryosuke Niwa]

Comment on attachment 202838 [details]
Tweak the test for new WKPageClose semantics

View in context: https://bugs.webkit.org/attachment.cgi?id=202838&action=review

> Tools/ChangeLog:10
> +        Since WKPageClose() will now disconnect from the web process if it's the last page
> +        being disconnected, stop expecting bundle messages after that point.
> +

Don't we still need to wait for WebProcess to terminate?

[Andreas Kling]

I'd like some advice on what to do here.

Now that WebPageProxy::close() may disconnect from the web process, we've removed the ability for an injected bundle to reliably communicate with the UI process in its willDestroyPage() callback.

Is this functionality important?

I suggested to Anders that we could switch the test to using WKPageTryClose() instead of WKPageClose(), and then let WebPage::tryClose() call out to the injected bundle's willDestroyPage() if it plans to indeed kill the page. Anders was not impressed.

[Sam Weinig]

(In reply to comment #6)
> I'd like some advice on what to do here.
> 
> Now that WebPageProxy::close() may disconnect from the web process, we've removed the ability for an injected bundle to reliably communicate with the UI process in its willDestroyPage() callback.
> 
> Is this functionality important?
> 
> I suggested to Anders that we could switch the test to using WKPageTryClose() instead of WKPageClose(), and then let WebPage::tryClose() call out to the injected bundle's willDestroyPage() if it plans to indeed kill the page. Anders was not impressed.

It seems valuable to preserve the ability for willDestroyPage() to always run, but I can also see arguments against it, since being able to shoot a process in the head when closing its last page could certainly be more efficient.  Perhaps we need new SPI to allow choosing which behavior is desirable, or maybe we can piggyback on existing sudden termination API.