Commit 85731000 authored by Ryan Scott's avatar Ryan Scott

Look through type synonyms in existential contexts when deriving Functor

This amounts to using `exactTyCoVarsOfType` instead of
`tyCoVarsOfType` in the right place. I also fixed a similar issue for
`-XDatatypeContexts` while I was in town (but couldn't be bothered to add a
test for it).

Test Plan: make test TEST=T13813

Reviewers: austin, bgamari, simonpj

Reviewed By: simonpj

Subscribers: simonpj, rwbarton, thomie

GHC Trac Issues: #13813

Differential Revision:
parent 559a0c5d
......@@ -512,7 +512,8 @@ cond_functorOK allowFunctions allowExQuantifiedLastTyVar _ rep_tc
tc_tvs = tyConTyVars rep_tc
Just (_, last_tv) = snocView tc_tvs
bad_stupid_theta = filter is_bad (tyConStupidTheta rep_tc)
is_bad pred = last_tv `elemVarSet` tyCoVarsOfType pred
is_bad pred = last_tv `elemVarSet` exactTyCoVarsOfType pred
-- See Note [Check that the type variable is truly universal]
data_cons = tyConDataCons rep_tc
check_con con = allValid (check_universal con : foldDataConArgs (ft_check con) con)
......@@ -524,7 +525,7 @@ cond_functorOK allowFunctions allowExQuantifiedLastTyVar _ rep_tc
-- in TcGenFunctor
| Just tv <- getTyVar_maybe (last (tyConAppArgs (dataConOrigResTy con)))
, tv `elem` dataConUnivTyVars con
, not (tv `elemVarSet` tyCoVarsOfTypes (dataConTheta con))
, not (tv `elemVarSet` exactTyCoVarsOfTypes (dataConTheta con))
= IsValid -- See Note [Check that the type variable is truly universal]
| otherwise
= NotValid (badCon con existential)
......@@ -666,4 +667,17 @@ As a result, T can have a derived Foldable instance:
foldr _ z T6 = z
See Note [DeriveFoldable with ExistentialQuantification] in TcGenFunctor.
For Functor and Traversable, we must take care not to let type synonyms
unfairly reject a type for not being truly universally quantified. An
example of this is:
type C (a :: Constraint) b = a
data T a b = C (Show a) b => MkT b
Here, the existential context (C (Show a) b) does technically mention the last
type variable b. But this is OK, because expanding the type synonym C would
give us the context (Show a), which doesn't mention b. Therefore, we must make
sure to expand type synonyms before performing this check. Not doing so led to
Trac #13813.
{-# LANGUAGE ConstraintKinds #-}
{-# LANGUAGE DeriveFunctor #-}
{-# LANGUAGE ExistentialQuantification #-}
{-# LANGUAGE KindSignatures #-}
{-# LANGUAGE StandaloneDeriving #-}
module T13813 where
import GHC.Exts (Constraint)
type C (a :: Constraint) b = a
data T a b = C (Show a) b => MkT b
deriving instance Functor (T a)
......@@ -91,3 +91,4 @@ test('T13297', normal, compile, [''])
test('T13758', normal, compile, [''])
test('drv-empty-data', [normalise_errmsg_fun(just_the_deriving)],compile, ['-ddump-deriv -dsuppress-uniques'])
test('drv-phantom', [normalise_errmsg_fun(just_the_deriving)],compile, ['-ddump-deriv -dsuppress-uniques'])
test('T13813', normal, compile, [''])
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment