Improve GetWeakPtr() to better protect against unsafe usage patterns.

base/memory/weak_ptr.h

Index: base/memory/weak_ptr.h
diff --git a/base/memory/weak_ptr.h b/base/memory/weak_ptr.h
index 33d1e473624583d9c7ec7e7c84afbe844608187f..25de4ed718df3891cc33f06a48cd0337e852d830 100644
--- a/base/memory/weak_ptr.h
+++ b/base/memory/weak_ptr.h
@@ -272,6 +272,9 @@ class WeakPtrFactory {
     ptr_ = NULL;
   }
 
+  // Returns a new WeakPtr. This is not thread-safe - in general callers should

[dcheng, 2016/02/19 20:45:00]

"in general" is only true for code that wants a weak pointer to (presumably)
bind on multiple threads, right? This recommendation feels a bit too strong, and
it doesn't seem to be consistent with how it's typically used today.

[Wez, 2016/03/01 23:20:28]

I've reworded the comment to be more explicit about concurrency, but without
making it even more verbose it is non-trivial to see why it would _ever_ be safe
to call GetWeakPtr() directly on the factory from another thread, since that
means the caller is making sure the WeakPtrFactory is still live at the time, in
which case it's questionable why they need the WeakPtr...

I'm thinking that replacing the wall-of-words under THREAD SAFETY with some
simple SAFE and UNSAFE threading examples, with a brief comment here, would be
better - WDYT?

+  // take and store a WeakPtr to pass to code that needs to post tasks rather
+  // than call GetWeakPtr() from different threads.
   WeakPtr<T> GetWeakPtr() {
     DCHECK(ptr_);
     return WeakPtr<T>(weak_reference_owner_.GetRef(), ptr_);