GHCi uses the Show instance too eagerly

Learn
#1

Hi, I’ve been noticing that recent GHCis don’t pretty print type signatures like they used to (pre - 9?).

λ> readIORef

<interactive>:125:1-9: error: [GHC-39999]
    • No instance for ‘Show (IORef a0 -> IO a0)’

The old behaviour (which I prefer) is to instead call the :t macro for you:

λ> :t readIORef
readIORef :: IORef a -> IO a

In the GHC Error Index, GHC-39999 is “No instance arising”.

It also tries to specialize type parameters to () when no better alternative is in scope, which is debatable but a different matter I guess.

e.g. with functions:

λ> M.lookup

<interactive>:127:1-8: error: [GHC-39999]
    • No instance for ‘Show (() -> M.Map () a0 -> Maybe a0)’
#2

8.10 seems to be doing exactly the same for readIORef at least. Not sure it was ever smart enough to do what you say (interpret it as if you had :t): it’s a Haskell value and by default GHCi will try to print it… It may be a little surprising if it inserted :t on the fly I think, so I’m not sure it’s a good idea to add this feature.

2 Likes