Top level mutable data structures problem

[Adrian Hey]

On Wednesday 20 Oct 2004 3:46 pm, Simon Marlow wrote:
> I liked the original idea.  I'm not sure if I agree with the argument
> that allowing fully-fledged IO actions in the initialisation of a module
> is unsafe.  I agree that it is a little opaque, in the sense that one
> can't easily tell whether a particular init action is going to run or
> not.

I'm not sure who was arguing that (not me anyway :-). I think the
argument was about whether or not there should be any such thing
as "module initialisation" at all, or at least that's what concerns
me most. (I don't think there should be).

> In any case, we're not going to rush to implement anything.  Discuss it
> some more ;-)

Good plan :-)

FWIW, at the moment the executive summary of MHO is that the laziness
and unpredictability of the unsafePerformIO hack is the one thing I like
about it and want to keep. What I don't like about it is the unsafety.

The unsafety I'm talking about is _not_ that that arises from allowing
arbitrary IO operations. I'm talking about the fact that the compiler
cannot be relied upon to generate code that accurately reflects the
programmers intentions without using NOINLINE pragma and -fno-cse
flag (the latter applying to an entire module).

There's also the (IMO) secondary issue of whether or not the IO actions
allowed during construction should be constrained in any way (e.g. using
type system tricks). I suggested a simple (perhaps naive) way to to this
(SafeIO monad), but I don't have any particularly strong views on this
either way. Just relying on programmers to use some common sense seems
fine to me also (this is what's currently done for finalisers for example).

Regards
--
Adrian Hey