Avoiding BlockedIndefinitelyOnSTM exceptions

[wren romano]

On Sat, Jul 19, 2014 at 11:24 PM, wren romano <winterkoninkje at gmail.com> wrote:
>     -- | The second argument allows handling 'BlockedIndefinitelyOnSTM' etc.
>     runSTSTM :: (forall s. STSTM s a) -> (STMError -> b) -> b

That should've been something more sensible, like:

    atomicallySTSTM :: (forall s. STSTM s a) -> (STMError -> b) -> IO
(Either a b)

> The idea being that, if the problem with catching
> BlockedIndefinitelyOnSTM has to do with the fact that all STM
> variables have global scope and so even if we could catch the
> exception itself we'd still have problems with cleaning up the
> collateral damage[1], then that's why it doesn't make sense to allow
> BlockedIndefinitelyOnSTM to be caught.
>
> [1] I don't know if this is actually the reason why why
> BlockedIndefinitelyOnSTM is uncatchable, rather it sounded like this
> is what Neil Davies was suggesting to be the reason. Also, I do seem
> to recall something like this actually being the case; though it's
> unclear whether the STSTM approach would actually be able to solve the
> problem.

Fwiw, after (re)reading ezyang's blog post about
BlockedIndefinitelyOnSTM, it seems clear that this is not actually
what the problem is. Though he also mentioned that the exact details
of BlockedIndefinitelyOnSTM shouldn't be relied upon, since the
exception is just a trick to kill off useless threads in order to
collect more garbage.

...

In terms of solving your actual problem in pipes, it seems like what
you really want are weak-references for STM. That way you don't rely
on the behavior of BlockedIndefinitelyOnSTM, and you can still get
garbage collection to send you signals whenever something becomes
inaccessible, allowing you to handle things however you like. Seems
like it shouldn't be too hard to do, since weak-references for IO are
already supported; though, of course, it'll still take a fair deal of
hacking to implement it.

-- 
Live well,
~wren