Excerpts from Simon Marlow's message of Tue Aug 31 05:02:13 -0400 2010:
I think the idea of annotating interruptible calls should be good enough. Simple blocking system calls like "read" can all be annotated as interruptible without any problems. Also, pthread_cancel() provides ways to control when cancellation can occur - a thread can say whether it allows cancels at any time or only at cancel points, so that will allow critical sections to be protected, and allow more complicated foreign calls to be made interruptible too.
Gotcha.
It seems to me that the obvious thing to do is only allow bound FFI calls to run on bound threads. What goes wrong with this approach? Is the waste of threads too severe?
Not sure what you mean here: what's a bound FFI call?
Good point: we don’t distinguish between FFI calls that require thread local state and which ones don’t: this might be a good thing to allow annotating. If we did know, then we could simply arrange for calls that use thread-local state to run on those threads, and we would still be able to farm out other FFI calls as necessary. A technical question about cleaning up task: when I run freeTask on the task, I get the following error: Foo: internal error: invalid closure, info=0xb76fb418 (GHC version 6.13.20100823 for i386_unknown_linux) freeTask is only used from freeTaskManager, so I suppose it’s not quite the right thing to do, however, as far as I can tell GHC doesn’t have a current story for freeing tasks. How should I proceed in figuring out the cause of this error? Cheers, Edward