Software Transactional Memory updated - Code Review Sought!

December 20, 2007 at 12:49 PM | categories: python, oldblog | View Comments

I've updated the STM code in subversion at present here:
I would really appreciate any code review from anyone who's capable of doing so - even if you've never done a code review before! Things I think need particular attention is the detail around locking. Specifically the following 4 methods are intended to be used by a user of the code:
class Store(object):
    def usevar(self, key):
    def set(self, key, value):
    def using(self, *keys):
    def set_values(self, D):
The above 4 functions entirely manage the locking. The following functions are internal and all assume that the store is locked appropriately before an update attempt:
    def __get(self, key):
    def __make(self, key):
    def __do_update(self, key, value):
    def __can_update(self,key, value):
Anyhow, why is this useful? Well, it turns out this should make resource management a lot simpler. If I'm right, the following code should be sufficient for the Dining Philosopher's problem:
take forks... (pseudo python, ignoring various details)
not_have_forks = True
while not_have_forks:
      S = <get the store from somewhere>
      forks = S.using("fork.1", "fork.2")
      if forks["1"].value == None and forks["2"].value == None:
          forks["1"].value = id(self)
          forks["2"].value = id(self)
          not_have_forks = False
   except ConcurrentUpdate:
       time.sleep(random()) # random backoff - doesn't guarantee success
   except BusyRetry:
       time.sleep(random()) # random backoff - doesn't guarantee success
# Put forks back - should always succeed. Should. :-)
    forks["1"].value = None
    forks["2"].value = None
except ConcurrentUpdate:
except BusyRetry:
... which I think is pretty neat :-)

What inspired me to this? Well, there's been a thread on the Python UK list recently about my STM implementation, which resulted in a discussion about a system called "MASCOT" being discussed. Essentially MASCOT looks incredibly similar to Kamaelia - at least superficially. From my perspective, the mere existence of MASCOT and its various subsystems which looks very similar I think validate a lot of the core ideas in Kamaelia. I want to write a post up about that separately though.... More later!

Anyhow, back to this, if anyone is willing to code review the STM code here:
I'd greatly appreciate it!

blog comments powered by Disqus