shire wrote:

> I was under the impression from our conversation that I was going to 
> implement these functions so I was a bit surprised when you committed 
> them instead.

Sorry, sort of all fell into place when I was doing the _apc_update()
call for upload progress fragmentation.

Feel free to clobber it with another impl, it was hardly few minutes on
top of the update call.

> My only minor comment so far is that we're making callback functions 
> under lock, which could be confusing.  I was going to do this for 
> Iterator but decided against it to avoid any ambiguity of when we 
> where/where not under lock.

I thought about that too, but it was a bit iffy to use asm cmpxchg
macros and hope shm writes work right. Also, I had images of an expunge
happening between the _find and the update in place.

I see that you've checked in the new allocator, I'll test it out
tonight. Also, going to build apc with the new valgrind macros[1]
to test out for race conditions.

Cheers,
Gopal
[1] - http://t3.dotgnu.info/blog/hacks/watching-writes-with-walgrind
-- 
The difference between insanity and genius is measured by success