GitLab

Suppose we have

data S = MkS { foo, bar, baz :: Int }
data T = MkT { foo, bar, baz :: Int }

blah x = x { foo = 1, bar = 2, baz = 3 }

GHC rightly errors, with

Scratch.hs:35:10: error:
    * Record update is ambiguous, and requires a type signature
    * In the expression: x {foo = 1, bar = 2, baz = 3}
      In an equation for `blah': blah x = x {foo = 1, bar = 2, baz = 3}
   |
35 | blah x = x { foo = 1, bar = 2, baz = 3 }
   |          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

But, with the reduction in power of -XDuplicateRecordFields coming soon, a type signature will be insufficient for disambiguating. Indeed, there is no good way to disambiguate this record update; the user will have to rewrite their code, likely to use a manual case or something like it. (If the records were in different modules, module prefixes could also help.)

In https://github.com/haskell/error-messages/issues/57, @adamgundry and I came up with

Scratch.hs:35:10: error:
    • Record update for fields 'foo', 'bar' and 'baz' is ambiguous,
      because they all appear in types 'S' and 'T'.
    * In the expression: x {foo = 1, bar = 2, baz = 3}
      In an equation for `blah': blah x = x {foo = 1, bar = 2, baz = 3}
   |
35 | blah x = x { foo = 1, bar = 2, baz = 3 }
   |          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

That would seem to be a nice little improvement, giving the user more information and not suggesting an ineffective fix.