Re: [APC-DEV] Re: [PECL-CVS] svn: /pecl/apc/trunk/ apc_cache.c apc_cache.h php_apc.c php_apc.h

Rasmus Lerdorf wrote:

Gopal V wrote:

Rasmus Lerdorf wrote:

shire wrote:

hrm, yeah I don't think we can get away without the lock here, simply because of the (*slot)->next/key type pointer usage. I believe the structure would allow that to be an invalid pointer if it's still in the process of being updated, inserted, removed etc.

I guess that is true. I was thinking even if a slot is deleted, it would be on the pending list and the *next would still be valid and take us back into the cache entries, but that is only true if the ref_count is non-zero, otherwise the entry is destroyed right away. With the lock though, apc_exists is only going to be faster than apc_fetch() by a memcpy (or many memcpys for an array).

The issue with such functions is that it sets up an easy race condition in user land. if(!apc_exists(...)) {

     apc_store();

} Which is nearly identical to the user cache slam I keep talking about. They're nice in some api sense of the word, but I fail to see how someone is going to use this one in any better way than the old fetch or store/add.

No, that wouldn't be the use case because that is exactly what apc_add() is for. apc_exists() is a slightly faster way to check if something is in the cache and it doesn't suffer from the value=false problem using apc_fetch() has when used for this.

As I understood this I thought the user wanted this to simplify his API approach when just checking if something was in the cache. So I think that I'm fine with this function being added to APC, regardless of it's speed difference. In some cases this could be faster, as I think we do spend a good bit of time copying and setting up zval structures if the data is an array, longer string, or that we are just fetching a large number of items. -shire