http://build.chromium.org/p/client.v8/builders/V8%20Linux64%20ASAN

The bug:
The code assumed that a weak Persistent whose weak callback is being
called would still be weak. That isn't true since the persistent is
un-weakened by the garbage collector before calling the weak callback. [1]

Specifically, PersistentValueMap would funnel all 'remove' actions
through its Release method, which uses PersistentBase::ClearWeak to
obtain the callback data. [2] For 'removes' caused by the weak callback,
ClearWeak always returns a NULL-pointer since by that time the weak
persistent was already un-weakend. The result was a memory leak in
the test, since the code to delete the weak callback data would
delete NULL.

The fix:
I explicity call Traits::DisposeCallbackData from the weak callback
with the data obtained from the v8::WeakCallbackData. To avoid invalid
calls to DisposeCallbackData, I also check whether this instance is
(still) weak before calling it. (That check could easily be elided
if it's expensive, for the price of having two 'remove' code paths.)

Severety:
Probably low. At least in Chromium, noone uses the API in a way to
trigger this; only the test does.

[1] https://code.google.com/p/chromium/codesearch#chromium/src/v8/src/global-handles.cc&q=global-handles.cc&sq=package:chromium&type=cs&l=231
[2] https://code.google.com/p/chromium/codesearch#chromium/src/v8/include/v8-util.h&sq=package:chromium&l=332-345

R=dcarney@chromium.org, dcarney

BUG=

Review URL: https://codereview.chromium.org/297193004

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@21514 ce2b1a6d-e550-0410-aec6-3dcde31c8c00