Re: [APC-DEV] 3.0.12 release

Gopal Vijayaraghavan wrote:

Rasmus Lerdorf wrote:

I am seeing lots of commits fly by from Gopal, Sara and Ilia. Cool! ;)

I guess this list is working then ;)

Is everyone at a stable point for a 3.0.12 release?

Not really, mainly because I know the optimizer is broken. Not that it wasn't any less broken in 3.0.11 :)

Yes, I am not too worried about the optimizer at this point.

I'd also like to explore having a way to optionally configure a local process-level quick cache that wouldn't require any locking.

I'm having trouble profiling a multi-process lock contention scenario with oprofile. I anybody's had success with oprofile, please advise.

fcntl-locking even without contention is just plain slow. Don't you think having a local copy would speed things up? It may not, but my guess is that especially with heavy user-cache usage, lock contention is an issue. Remember that we hold the lock for the duration of the memcpy on a user cache hit.

An malloc() cache of 20 entries doesn't sound too hard to do. Especially since there's none of the pending member hacks in a process (what about TSRM pthread folks ?).

Yeah, for TSRM I guess we don't gain anything, so I'd say disable it in TSRM mode. Nobody sane runs TSRM anyway. -Rasmus