[Haskell] Re: Global Variables and IO initializers

[Jules Bean]

On 23 Nov 2004, at 11:53, George Russell wrote:

> I wrote (snipped):
> > 3) It needs no extensions to the Haskell language, and only fairly
> > standard hierarchical libraries like Data.IORef.
>
> Lennart Augustsson wrote (snipped):
> > It uses unsafePerformIO which is very much an extension to Haskell. 
> :)
>
> Ben Rudiak-Gould wrote (snipped):
> > I think by Haskell he means the common language currently 
> implemented by
> > GHC and Hugs.
>
> No.  I mean by the "Haskell language" what is described in
> the Haskell 98 Report.   unsafePerformIO is not part of the language,
> it is a value defined by one of the standard hierarchical libraries.
>

But it isn't a value, is it? Since it fails referential transparency.

'unsafePerformIO' is not, in the ordinary sense a 'library routine', 
since library routines might ordinarily be expected to preserve haskell 
semantics and program equivalence etc etc. It is a genuine extension to 
the language (and genuinely breaks the language's semantics unless you 
meet the proof obligation, which is not checkable by the compiler).

Jules