Error messages suffer when applying a visible type application to a typoed/out-of-scope function

Summary

When working with the -XTypeApplications extension, making a typo in the function name to which the type is being applied can lead to cryptic error messages. My team has observed this in production; new Haskellers can find it very confusing, since it’s not obvious that the message originates from a typo rather than a more fundamental type error.

I’m not sure what can be done about this, but Stephanie encouraged me to file this bug when I saw her at ICFP, so I figured it deserves a place in the tracker.

Steps to reproduce

The following simple program compiles correctly:

{-# LANGUAGE TypeApplications #-}

foo :: Num a => a
foo = 5

main :: IO ()
main = print (id @Int foo)

If we introduce an out-of-scope function into the id @Int foo invocation, such as idd @Int foo, we get the following message, which doesn’t indicate that the idd function is out of scope:

Main.hs:12:15: error:
    • Cannot apply expression of type ‘t1’
      to a visible type argument ‘Int’
    • In the first argument of ‘print’, namely ‘(idd @Int foo)’
      In the expression: print (idd @Int foo)
      In an equation for ‘main’: main = print (idd @Int foo)
   |
12 | main = print (idd @Int foo)
   |               ^^^^^^^^^^^^

By contrast, if we insert a typo into the @Int invocation, we get a good error message:

Main.hs:12:19: error:
    Not in scope: type constructor or class ‘In’
    Perhaps you meant one of these:
      ‘IO’ (imported from Prelude), ‘Int’ (imported from Prelude)

If we make a typo in the foo parameter, we also get a good message:

Main.hs:12:23: error:
    • Variable not in scope: oo :: Int
    • Perhaps you meant one of these:
        ‘foo’ (line 9), ‘or’ (imported from Prelude)
   |
12 | main = print (id @Int oo)
   |                       ^^

Expected behavior

Given an invocation like idd @Int foo, I would expect to see a message that warned Variable not in scope: idd, rather than the somewhat-cryptic Cannot apply expression of type ‘t1’ message. As I mentioned above, I’m sure that this involves some sort of phasing restriction, so I understand if this is a nontrivial fix.

Environment

• GHC version used: 8.6.5

Optional:

• Operating System:
• System Architecture: