Adrian Hey wrote:
>
> On Mon 24 Jan, Michael Hobbs wrote:
> > (*) One place where the World -> (World, a) model breaks down is when
> > the IO function is a blocking function such as "getChar :: IO Char". If
> > this function was equivalent to World -> (World, a) then that would mean
> > that the result is completely determined by the input World value.
>
> As a world value sceptic myself, I am reluctant to say this, but in
> all honesty I don't think that this argument holds. If I understand
> you correctly your objection is that getChar should never block, it
> should get a character instantly because the information about which
> character will eventually be got should already be embedded in the
> input world value.
Actually, I never thought about whether or not getChar should ever
block. My intention is to point out that trying to wed referential
transparency with I/O using the concept of World values is filled with
philosophical ambushes and tiger traps. The particular trap I was trying
to point out here was that of determinism. You just pointed out a
different one along the same lines: If getChar really was referentially
transparent, with respect to the given World value, should it block?
There might exist a philosopher with enough skill to successfully
navigate past these traps and pitfalls and come up with a coherent
answer, but I know that I am not that philosopher. My question is this:
In the end, does it really make a difference whether or not we can
successfully model I/O as being somehow referentially transparent? Can
we pull out Occam's Razor[1] and simply cut referential transparency
away from I/O and come out with a simpler model?
- Michael Hobbs
[1] gratuitous philosophical reference to make me look smart
