Consider the following minimal example:
{-# LANGUAGE RankNTypes #-}
module Test where
class C w
data A = A (forall u. C u => u)
x :: forall u. C u => u
x = undefined
a = A x
This typechecks fine, as expected. However, if a
is refactored to use a let
statement:
{-# LANGUAGE RankNTypes #-}
module Test where
class C w
data A = A (forall u. C u => u)
x :: forall u. C u => u
x = undefined
a = let x' = x in A x'
It suddenly fails to typecheck with the following error:
test.hs:12:14: error:
* No instance for (C u0) arising from a use of `x'
* In the expression: x
In an equation for x': x' = x
In the expression: let x' = x in A x'
|
12 | a = let x' = x in A x'
| ^
test.hs:12:21: error:
* Couldn't match expected type `u' with actual type `u0'
because type variable `u' would escape its scope
This (rigid, skolem) type variable is bound by
a type expected by the context:
forall u. C u => u
at test.hs:12:19-22
* In the first argument of `A', namely x'
In the expression: A x'
In the expression: let x' = x in A x'
* Relevant bindings include x' :: u0 (bound at test.hs:12:9)
|
12 | a = let x' = x in A x'
Why is this happening? Doesn't this violate equational reasoning?
This is the result of the dreaded monomorphism restriction. Enabling
XNoMonomorphismRestriction
should cause this to compile.a = let x' = x in A x'
is not equivalent toa = A x
, because under monomorphism restrictionx'
inlet x' = ...
is monomorphic, but A requires a polymorphic argument.