[Haskell-cafe] ReaderT and concurrency

[Tomasz Zielonka]

On Wed, Nov 16, 2005 at 10:53:17AM -0800, Andrew Pimlott wrote:
> On Wed, Nov 16, 2005 at 07:20:48PM +0100, Tomasz Zielonka wrote:
> > I think it wouldn't be possible using only methods in MonadIO.
> 
> Here's what I had in mind for forkProcess.  Recall that the fork syscall
> returns 0 to the child and a pid to the parent.
> 
>     forkProcess :: MonadIO m => m () -> m ProcessID
>     forkProcess io = do pid <- liftIO forkSyscall
>                         if (pid == 0) then io >> exit
>                                       else return pid

Well, yes, I haven't thought about it. I don't know forkSyscall
and I can't say if it's a good idea or not.

> But maybe the primitives used by forkIO do not allow this approach.  I'm
> not sure.

I'm pretty sure they don't, but I may be wrong again. They could
probably be extended to allow this. However, I'm not sure I would like
this, because it would be too easy to break abstraction.

> > Besides,
> > what should be the semantics of forkIO for (StateT s IO)? I can't think
> > of anything reasonable.
> 
> I don't see the problem--they would each get a copy of the state.
> Although I could be missing something, as I use System.Posix.Process
> more than Control.Concurrent.

Yes, but the changes made in the child thread would be ignored
in the parent.

Best regards
Tomasz