My understanding of the MR is heavily influenced by the work I did on
Hatchet, which is based
directly on Mark Jones' paper (and code) "Typing Haskell in Haskell".
I thought I would go back to that paper and see how he defines
"simple" pattern bindings
and the MR.
I now quote directly from the relevent sections of the paper:
He represents function bindings with the Alt type, which is defined
on page 9:
"The representation of function bindings in following sections
uses alternatives, represented by values of type Alt :
type Alt = ([Pat], Expr)
An Alt specifies the left and right hand sides of a function
definition."
On page 12 he defines what simple means with repsect to the Alt type
and the MR:
"A single implicitly typed binding is described by a pair con-
taining the name of the variable and a list of alternatives:
type Impl = (Id , [Alt])
The monomorphism restriction is invoked when one or more
of the entries in a list of implicitly typed bindings is simple,
meaning that it has an alternative with no left-hand side
patterns.
restricted :: [Impl] -> Bool
retricted bs
= any simple bs
where
simple (i, alts) = any (null . fst) alts
"
Curiously, his Alt type does not offer any way to represent pattern
bindings where
the lhs is a non-variable pattern. But he addresses this point later
on page 13:
"In addition to the function bindings that we have seen al-
ready, Haskell allows variables to be defined using pattern
bindings of the form pat = expr . We do not need to deal di-
rectly with such bindings because they are easily translated
into the simpler framework used in this paper. For example,
a binding:
(x,y) = expr
can be rewritten as:
nv = expr
x = fst nv
y = snd nv
where nv is a new variable. The precise definition of the
monomorphism restriction in Haskell makes specific refer-
ence to pattern bindings, treating any binding group that
includes one as restricted. So, at first glance, it may seem
that the definition of restricted binding groups in this pa-
per is not quite accurate. However, if we use translations as
suggested here, then it turns out to be equivalent: even if
the programmer supplies explicit type signatures for x and
y in the original program, the translation will still contain
an implicitly typed binding for the new variable nv."
It is not obvious that this is consistent with the Report; I'll have to
think about it more.
Cheers,
Bernie.
_______________________________________________
Haskell-Cafe mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/haskell-cafe
