GitLab

Playing around with this file

module MethodRules where

class C1 a where
  m1 :: Maybe a

class C2 a where
  m2 :: Maybe a
  m2' :: Maybe a

foo :: Maybe a
foo = Nothing

{-# RULES "m1" m1 = foo #-}
{-# RULES "m2" m2 = foo #-}

I get

ghc -O2 -Winline-rule-shadowing MethodRules.hs  -ddump-rules -fforce-recomp 
[1 of 1] Compiling MethodRules      ( MethodRules.hs, MethodRules.o )

MethodRules.hs:13:11: warning: [-Winline-rule-shadowing]
    Rule "m1" may never fire because ‘m1’ might inline first
    Probable fix: add an INLINE[n] or NOINLINE[n] pragma for ‘m1’
   |
13 | {-# RULES "m1" m1 = foo #-}
   |           ^^^^^^^^^^^^^

==================== Tidy Core rules ====================
"m1" forall (@ a) ($dC1 :: C1 a). m1 @ a $dC1 = foo @ a
"m2" forall (@ a) ($dC2 :: C2 a). m2 @ a $dC2 = foo @ a

I find it odd that the inline-rule-shadowing warning behaves so different depending on whether the class is a single-method class or not.

Furthermore, I found no way to avoid the warning: I can’t add an {-# NOINLINE #-} pragma to a class method (see #10595). Worse, in terms of developer confusion, if the method happens to have a default implementation, I can add the {-# NOININE #-} pragma, but it would apply to the default method (I think), and will not prevent this warning.