Bug 81504 – [chromium] DRT crashes at shutdown.

• Component: Tools / Tests
• Status: RESOLVED FIXED

[Hao Zheng]

In debug build of DRT, it crashes randomly at shutdown. I think it's due to a race condition. I can trigger the crash easily by this test:

out/Debug/DumpRenderTree /usr/local/code/webkit.git/LayoutTests/compositing/iframes/overlapped-nested-iframes.html

The stacktrace:

ASSERTION FAILED: !CCLayerTreeHost::anyLayerTreeHostInstanceExists()
Source/WebKit/chromium/src/WebCompositorImpl.cpp(78) : static void WebKit::WebCompositorImpl::shutdown()
1   0x48fd50
2   0x48fbbf
3   0x46cc36
4   0x4378d3
5   0x7fb676dcac4d __libc_start_main
6   0x422fd9
[30647:30647:19876206991:ERROR:process_util_posix.cc(142)] Received signal 11
        base::debug::StackTrace::StackTrace() [0x7315c6]
        base::(anonymous namespace)::StackDumpSignalHandler() [0x6ec18d]
        0x7fb676ddfaf0
        WebKit::WebCompositorImpl::shutdown() [0x48fd5a]
        WebKit::WebCompositor::shutdown() [0x48fbbf]
        TestShell::~TestShell() [0x46cc36]
        main [0x4378d3]
        0x7fb676dcac4d
        0x422fd9

[Hao Zheng]

Created attachment 132574 [details]
Patch

[Hao Zheng]

I'm wrong in the first comment. This is not random; it definitely happens after compositing.

[Tony Chang]

This looks sane to me, but I'm not familiar with this code.  Adding people who are.

[James Robinson]

I think we have to shut down WebKit::WebCompositor before we stop threading - when does that happen in DRT?

[Tony Chang]

(In reply to comment #4)
> I think we have to shut down WebKit::WebCompositor before we stop threading - when does that happen in DRT?

When exiting main(), we ~WebKitSupportTestEnvironment gets run, which calls webkit_support::TearDownTestEnvironment(), which calls WebKit::shutdown().  I believe that's when threading is stopped.  So, it's still after the new location of WebCompositor::shutdown().

[James Robinson]

Comment on attachment 132574 [details]
Patch

OK, that should be fine then!  We definitely should be shutting down after all WebViews are gone

[WebKit Review Bot]

Comment on attachment 132574 [details]
Patch

Clearing flags on attachment: 132574

Committed r111220: <http://trac.webkit.org/changeset/111220>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.

[Hao Zheng]

(In reply to comment #5)
> (In reply to comment #4)
> > I think we have to shut down WebKit::WebCompositor before we stop threading - when does that happen in DRT?
> 
> When exiting main(), we ~WebKitSupportTestEnvironment gets run, which calls webkit_support::TearDownTestEnvironment(), which calls WebKit::shutdown().  I believe that's when threading is stopped.  So, it's still after the new location of WebCompositor::shutdown().

Thanks for your explanation, Tony. I found it kind of tricky to load a blank page in ~WebViewHost, which is called by ~TestShell. Howerver, loading a page results in some function calls eventually going back into TestShell again, which is at the state of destruction then. This is kind of dangerous to me.