73495 - PluginArrayImpl is leaked and causing plugin host leak

Status: RESOLVED WORKSFORME

(Core :: DOM: Core & HTML, defect)

Description

[sean echevarria]

This just started happening within the last couple of days.  calling 
javascript:document.plugins.length causes PluginArrayImpl to be leaked.  
Because it's dtor does not run the plugin host (nsPluginHostImpl) does not get 
destroyed and that's a problem for plugins.

will attach testcase.

Comment 1

[sean echevarria]

Attachment: testcase

Comment 2

[Johnny Stenback (:jst)]

Av, could you have a look at this?

Comment 3

[sean echevarria]

the PluginArrayImpl that is getting leaked is owned by NavigatorImpl in 
nsGlobalWindow.cpp.  It too is leaked.

Comment 4

[Johnny Stenback (:jst)]

How do you check for the leak? You do know that the JS GC can and will hold on
to things that are referenced from a page sometimes longer than the lifetime of
the page, right? If things stay around past app shutdown then we have a real
problem.

Comment 5

[sean echevarria]

I've been setting breakpoints in NavigatorImpl::NavigatorImpl, 
NavigatorImpl::~NavigatorImpl, PluginArrayImpl::PluginArrayImpl, 
PluginArrayImpl::~PluginArrayImpl.

Just starting doing nsDocument::nsDocument and nsDocument::~nsDocument and 
confirm that when you run the testcase we leak nsDocuments.  If you start up 
with a blank page and then exit the ctor count matches the dtor count.  Go to 
the testcase and that's not the case.

Things living longer than a page is not the problem.  I'm seeing these things 
never getting destroyed - and the plugin host must get shutdown properly.

Comment 6

[Peter Lubczynski]

Sean,

Is this a dup of bug 73289?

Comment 7

[sean echevarria]

Peter: sort of, could be, not sure - a lot seems to have changed in the last 
week and I'm pretty sure I was experiencing 73289 before this one.  That's all 
I have right now to differentiate the two.  I opened 73289 for the specific 
case of the host getting leaked during the handling of a js exception.  This 
bug is broader and happens without the occurence of a js exception - 
nsPluginHostImpl, NavigatorImpl, PluginArrayImpl and nsDocument are getting 
leaked.

The patch for bug 73289 will prevent the leak of the plugin host from holding 
onto plugins past shutdown - but it doesn't fix the leak of the host itself.

Comment 8

[av (gone)]

What is the latest on this one?

Comment 9

[sean echevarria]

Last I checked, the testcase still showed a leak of the plugin host 
(the 'killing plugin host' printf has changed to 'nsPluginHostImpl dtor').  
It's no longer an issue from a plugin developer's view since we've assured that 
the plugin host now releases plugins at xpcom shutdown regardless of whether or 
not the host itself is released.  I've been experiencing cvs problems and can't 
update to see if the leak is still occurring.

Comment 10

[sean echevarria]

no more leak as of Apr 19!
closing as wfm