Re: Why is this not allowed?

From: Pierre Weis (Pierre.Weis@inria.fr)
Date: Fri Dec 03 1999 - 14:50:47 MET


From: Pierre Weis <Pierre.Weis@inria.fr>
Message-Id: <199912031350.OAA04967@pauillac.inria.fr>
Subject: Re: Why is this not allowed?
To: prevost@maya.com (John Prevost)
Date: Fri, 3 Dec 1999 14:50:47 +0100 (MET)
In-Reply-To: <87n1rtb2z7.fsf@isil.maya.com> from "John Prevost" at Dec 2, 99 01:40:44 pm

> Damien Doligez <Damien.Doligez@inria.fr> writes:
>
> > ># let rec id'' = id;;
> > >This kind of expression is not allowed as right-hand side of `let rec'
> >
> > Because we don't know how to compile "let rec x = x" or
> > let rec x = y
> > and y = x
> >
> > Moreover, you can just remove the "rec" and it works. It is possible
> > to implement "let rec var1 = var2", but the usefulness is quite small
> > compared to the amount of code we would need to write.
>
> Well, you can't quite. Because of the value restriction, id'' defined
> like this:
>
> let id'' = id;;
>
> has type:
>
> id'' : '_a -> '_a
>
> (That is, it'll only work for a single type.)
>
> John.

There is some confusion here: since «id» is an expression that is a
mere variable, it is not expansive; hence it can be safely generalized;
hence it can be used with different uncompatible types:

# let id x = x;;
val id : 'a -> 'a = <fun>
# let id'' = id;;
val id'' : 'a -> 'a = <fun>
# id'' 1, id'' true;;
- : int * bool = 1, true

Rule of thumb concerning the polymorphism generalisation restriction:

the restriction is performed by matching the right hand side expression e
of a let definition:

1 if e is a constant, a variable expression, a function (that is a
syntactically explicit function, introduced by the keyword fun or
function), or a constructor application, then its type can safely be
generalised.

2 otherwise the type of e is not generalized.

Condition 1 is a rough approximation of the property ``the execution
of e cannot create a polymorphic mutable value''; in practice,
condition 2 means that if e is a function application then its type is
not generalized.

Best regards,

Pierre Weis

INRIA, Projet Cristal, Pierre.Weis@inria.fr, http://cristal.inria.fr/~weis/



This archive was generated by hypermail 2b29 : Sun Jan 02 2000 - 11:58:29 MET