Wildcard is inferred for partial function

[JordanMartinez]

Description

See purescript/purescript-lists#203 (comment)

To Reproduce

module Main where

import Prelude

data List a = Nil | Cons a (List a)

last :: forall a. Partial => List a -> a
last (Cons a Nil) = a
last (Cons _ xs) = last xs

Compiling the above produces this warning:

[1/1 WildcardInferredType] src/Main.purs:10:1

  10  last :: forall a. Partial => List a -> a
      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  
  Wildcard type definition has the inferred type
  
    a0
  
  in binding group last
  
  where a0 is a rigid type variable
          bound at (line 0, column 0 - line 0, column 0)

Expected behavior

The warning should not be produced.

PureScript version

This doesn't occur on v0.14.5 but does on v0.14.7.

[JordanMartinez]

Is this caused by #4208 here?

[rhendric]

Sure looks like it.

[JordanMartinez]

Looks like this warning was emitted in CI for that PR, too. See https://github.com/purescript/purescript/runs/4601156598?check_suite_focus=true#step:8:5016

However, we missed it because there's so many other warnings in the package set.

[rhendric]

@natefaubion, quick (?) question for you:

I'm looking at this change in #3500, and I notice that there isn't a corresponding censor in the BindingGroupDeclaration case. Is that intentional?

The regression I introduced seems to be prevented by the censorUnnamedLocalWildcards in the ValueDecl case, which is why out of the four Partial declarations in Data.List.Partial, only the two recursive ones get the warning. I'm wondering if I should attempt to adapt censorUnnamedLocalWildcards to also work on binding groups or if I should take an entirely different approach.

[natefaubion]

I can't think of an obvious reason.

Does the approach of using typeWildcard mean that if you have a partial function, and did not provide a type signature, you would see the wildcard warning along with the partial error?

[rhendric]

Not in any cases I've tested; I'm under the impression that in general no warnings are ever emitted if any errors are.

[rhendric]

I'm having trouble wrapping my head around the rules of wildcard warnings. Could you look at these examples and tell me which of them ought to be issuing warnings? None of them involve Partial.

[natefaubion]

I think your expectations are correct. If there's a top-level signature, then there should be no wildcard warnings ever. If there isn't a top-level signature, all bets are off. I don't remember the exact reason why that made sense at the time, but I'm sure I could be convinced it's nonsense.