Bug 93913 – Removing a plug-in element from a page opened in a background tab in Safari crashes

• Status: RESOLVED
• Resolution: FIXED
• Component: WebKit2

[Brady Eidson]

Removing a plug-in element from a page opened in a background tab in Safari crashes

In Safari background-tab opened plug-ins aren't initialized until the tab is brought to the foreground.  When such a page removes a plug-in element, we crash:

#0	0x0000000101a1587c in WTF::RefPtr<CoreIPC::Connection>::get at /Volumes/SSD-Data/ZinUser/build/Debug/usr/local/include/wtf/RefPtr.h:58
#1	0x0000000101b18f4c in WebKit::PluginProcessConnection::connection at /Volumes/SSD-Data/git/OpenSource/Source/WebKit2/WebProcess/Plugins/PluginProcessConnection.h:54
#2	0x0000000101b2ccdd in WebKit::PluginProxy::destroy at /Volumes/SSD-Data/git/OpenSource/Source/WebKit2/WebProcess/Plugins/PluginProxy.cpp:202
#3	0x0000000101b012aa in WebKit::Plugin::destroyPlugin() ()
#4	0x0000000101b37c24 in WebKit::PluginView::~PluginView() ()

In this case the PluginProxy has a null m_connection.  A simple null check should suffice.

In radar as <rdar://problem/12057991>

[Brady Eidson]

Created attachment 158157 [details]
Patch v1 - Fix + regression test

[Brady Eidson]

Hmmmm I see there's a GTK failure already...

./.libs/libWebCoreInternals.a(./.libs/../Source/WebCore/testing/.libs/libWebCoreInternals_la-InternalSettings.o):InternalSettings.cpp:function WebCore::InternalSettings::Backup::restoreTo(WebCore::Page*, WebCore::Settings*): error: undefined reference to 'WebCore::Page::setCanStartMedia(bool)'
./.libs/libWebCoreInternals.a(./.libs/../Source/WebCore/testing/.libs/libWebCoreInternals_la-InternalSettings.o):InternalSettings.cpp:function WebCore::InternalSettings::setCanStartMedia(bool, int&): error: undefined reference to 'WebCore::Page::setCanStartMedia(bool)'
collect2: error: ld returned 1 exit status

As far as I can tell Page::setCanStartMedia() is defined in all ports.  Dunno what this is about yet.

[Build Bot]

Comment on attachment 158157 [details]
Patch v1 - Fix + regression test

Attachment 158157 [details] did not pass win-ews (win):
Output: http://queues.webkit.org/results/13486729

[Tim Horton]

(In reply to comment #2)
> As far as I can tell Page::setCanStartMedia() is defined in all ports.  Dunno what this is about yet.

The Windows error makes me wonder if it's not something to do with exports or something?

[Brady Eidson]

(In reply to comment #4)
> (In reply to comment #2)
> > As far as I can tell Page::setCanStartMedia() is defined in all ports.  Dunno what this is about yet.
> 
> The Windows error makes me wonder if it's not something to do with exports or something?

Indeed the win and gtk failures are uncannily the same error...  I was under the (perhaps quite mistaken) impression that Mac was the only platform with explicit symbol exporting (WebCore.exp.in).

Grumble!

[Brady Eidson]

Created attachment 158578 [details]
Patch v2 - Shot at fixing GTK

[Build Bot]

Comment on attachment 158578 [details]
Patch v2 - Shot at fixing GTK

Attachment 158578 [details] did not pass win-ews (win):
Output: http://queues.webkit.org/results/13516028

[Brady Eidson]

(In reply to comment #7)
> (From update of attachment 158578 [details])
> Attachment 158578 [details] did not pass win-ews (win):
> Output: http://queues.webkit.org/results/13516028

I think I have a response to the Win failure (WebKit2.def for exports) but will let the gtk ews play out on patch v2 first.

[Brady Eidson]

Created attachment 158593 [details]
Patch v3 - All the right exports in all the right places

[Brady Eidson]

Will land after win ews confirms it builds.

[Brady Eidson]

http://trac.webkit.org/changeset/125695