On Tue, Nov 24, 2009 at 3:14 AM, Christophe Grand <christo...@cgrand.net> wrote:
> On Mon, Nov 23, 2009 at 11:56 PM, John Harrop <jharrop...@gmail.com> wrote:
>>
>> I'm starting to think that for some tasks Clojure could use a concept of
>> "row locking" with maps.  It would mean having a map-of-refs type that was
>> integrated with the STM, so multiple updates whose keys didn't collide could
>> occur concurrently.
>> It *might* be possible to do this using commute and update-in, with a ref
>> wrapping the whole map. The tricky thing is you want the update-ins to
>> commute if the keys are not the same, but not if they are. Perhaps we need a
>> "conditional commute" that takes two extra arguments, a value to test and a
>> binary predicate. Then it could be done with (conditional-commute key = map
>> update-in [key] val-transform-fn).
>> The idea here being, in this case, that if two of these were done in
>> overlapping transactions, the first arguments would be compared using the
>> second argument. If the result was true one transaction would be retried, if
>> false the operations would commute. (If the second arguments to the two
>> conditional-commutes differed the transaction would be retried.)
>
> I had a similar idea but more general: being able to specify invariants
> inside a transaction. Commit will procede only if the invariant still holds.
> Your proposed conditional-commute could be rewritten:
>
> ;; (conditional-commute key = map update-in [key] val-transform-fn)
> (invariant (@map key))
> (commute map update-in [key] val-transform-fn)
>
> See the attached patch for a prototype.
>

I had forgotten about this, could you please make an issue for it? I'd
like to look into it for a future feature.

Thanks,

Rich

-- 
You received this message because you are subscribed to the Google
Groups "Clojure" group.
To post to this group, send email to clojure@googlegroups.com
Note that posts from new members are moderated - please be patient with your 
first post.
To unsubscribe from this group, send email to
clojure+unsubscr...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/clojure?hl=en

Reply via email to