Simon Peyton Jones pushed to branch wip/T20264 at Glasgow Haskell Compiler / GHC Commits: d561a318 by Simon Peyton Jones at 2026-03-06T00:20:24+00:00 Weaken Lint invariant Don't insist that an occurrence has syntactially the same type as the binder. This is hard to ensure in `mkPolyAbsLams`. I tried a shadowing x=x binding, but that didn't quite work (when compling cabal) and maybe it is not necessary anyway. - - - - - 2 changed files: - compiler/GHC/Core/Lint.hs - compiler/GHC/Core/Utils.hs Changes: ===================================== compiler/GHC/Core/Lint.hs ===================================== @@ -3775,9 +3775,10 @@ checkBndrOccCompatibility in_bndr v_occ -- Check that binder and occurrence have same type -- See Note [Occurrence should have the same type as binder] - ; checkL (pickyEqType occ_ty in_bndr_ty) $ -- Compares InTypes - hang (text "Mismatch in type between binder and occurrence") - 2 extra_info +-- ; checkL (pickyEqType occ_ty in_bndr_ty) $ -- Compares InTypes + ; ensureEqTys occ_ty in_bndr_ty $ + hang (text "Mismatch in type between binder and occurrence") + 2 extra_info ; checkLM (sameUnfolding in_bndr v_occ) $ hang (text "Mismatch in type-let unfolding between binder and occurrence") ===================================== compiler/GHC/Core/Utils.hs ===================================== @@ -3360,14 +3360,14 @@ So: * `AbsVars` is a list of free variables, in dependency order, to abstract. Some of them may be tyvars-with-unfoldings. For example - vs = [a, b=[a], x:b] + vs = [a, b{=[a]}, x:b{=[a]}] * When we make an AbsVars list, we close over the free vars of the unfoldings - of any tyvars in it. So if `b{=Maybe a}` is in the list then so is `a` + of any tyvars in it. So if `b{=[a]}` is in the list then so is `a` * `mkCoreAbsLams` (more generally `mkPolyAbsLams`) forms a lambda abstraction pushing the tyvar bindings into the body: - mkCoreAbsLams [a, b=[a], x:b] body + mkCoreAbsLams [a, b{=[a]}, x:b{=[a]}] body = \a. \(x:[a]). let @b = [a] in let x:b = x in -- See (AFV1) body @@ -3389,7 +3389,7 @@ Wrinkles Why do we need that funny (non-recursive!) `x:b = x` binding? Answer: without it we'd have this: = \a. \(x:[a]). let @b = [a] in - reverse (x:b) + reverse (x:b{=[a]}) where the /occurrence/ Var (x:b) has a different type to the /binding/ x:[a]. -} @@ -3423,13 +3423,13 @@ mkPolyAbsLams (getter,setter) bndrs body | isTyVar var, change_ty , let binds' | isDeadBinder var = binds - | otherwise = (bndr, varToCoreExpr var1) : binds + | otherwise = binds -- (bndr, varToCoreExpr var1) : binds -- Why this let-binding? = Lam (setter var1 bndr) (go unf_tvs binds' bndrs) | isId var, change_ty || change_unf , let binds' | isDeadBinder var = binds - | otherwise = (bndr, varToCoreExpr id2) : binds + | otherwise = binds -- (bndr, varToCoreExpr id2) : binds = Lam (setter id2 bndr) (go unf_tvs binds' bndrs) | otherwise View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/d561a3188610076b56d6f817a9187352... -- View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/d561a3188610076b56d6f817a9187352... You're receiving this email because of your account on gitlab.haskell.org.