Re: CLOS hash tables & threads

[Sam S. <sd...@gn...>]

Vladimir Tzankov wrote:
> On 2/18/10, Sam Steingold <sd...@gn...> wrote:
>> On a different note: maybe make-hash-table should accept a :shared argument
>> which would then add a mutex slot to it.
>> Then all destructive HT operations on such a table would check for this slot
>> and make sure that it has been locked.
>> (is mutex_lock on an already locked mutex expensive?)
>> e.g.:
>> (with-lock (ht)
>>    (incf (gethash x ht 0)))
> 
> Recursive mutexes are cheap.

in that case this :SYNC arg to make-hash-table looks reasonable.

> Current hash table implementation may need locking mutex on
> non-destructive lookups in order to guard against internal state
> changes that may cause segfaults (I am not 100% sure - will check it).

yes.