593231 - Consider using a main thread refcount on strings

Status: RESOLVED INVALID

(Core :: XPCOM, defect)

Description

[Justin Lebar (not reading bugmail)]

Split off from bug 585661.

I'm going to play around with a simpler approach:

 * Keep a main-thread refcnt and a global refcnt.
 * In addref / release, check whether we're the main thread.  If so, modify
only the main-thread refcnt using a regular, thread-unsafe operation.  When the
main-thread refcnt goes from 1 to 0, atomically decrement the global refcnt
and, if the new global refcnt is 0, delete the object.

Comment 1

[Justin Lebar (not reading bugmail)]

I see no reason offhand we couldn't do this for more than just strings, but maybe strings are a good place to start.

Comment 2

[Justin Lebar (not reading bugmail)]

Hm...this might not work.

The assumption is that main-thread addrefs will be balanced by main-thread releases.  This does not seem to be the case.

Comment 3

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

This assumes not merely balance, but also that the object is destroyed on the main thread.

Comment 4

[Justin Lebar (not reading bugmail)]

(In reply to comment #3)
> This assumes not merely balance, but also that the object is destroyed on the
> main thread.

Why?  You destroy the object once the global refcnt goes to 0, regardless of whether that was caused by the main thread releasing the object or another thread releasing the object.

Comment 5

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Ah, you are right of course.

Comment 6

[Justin Lebar (not reading bugmail)]

This bug is going nowhere, fast.  Since main-thread addrefs aren't balanced by main-thread releases, I think the approach is INVALID.