Naming a cases branch variable the same as an ability constructor leads to empty ucm output

[samgqroberts]

My scratch file solely consists of:

repro : Text -> ()
repro = cases
  input -> ()

If I save, regardless of whether I change anything (except for the name input), ucm gives me empty output.
Eg. after 3 saves I see:

.>
.>
.>

If I change the name input to anything else, I get the expected "These new definitions are ok to add").

I believe this is somehow colliding with the following saved ability, and confusing ucm:

.> view CLI                                                                                                                                                                                                                    
  ability CLI where
    print : Text ->{CLI} ()
    input : {CLI} Text

[pchiusano]

Thanks. Suspect this is triggering #1578. The correct behavior should be either be a type error (because the LHS of a pattern is an ability operation) or it should just treat input as a wildcard variable and CLI.input as a pattern.

[pchiusano]

#1578 has been fixed, but the issue is still present. It looks like it's getting parsed as:

repro : Text -> ()
repro = cases
  CLI.input -> ()

See the branch fix/1731, and you can run the transcript with:

stack build && stack exec unison -- transcript unison-src/transcripts/fix1731.md

The parsing thing is easy enough to fix, just peep TermParser.parsePattern.

But even so, why is it just failing silently with no output? I'd at least expect a type error or something - the scrutinee is of type text and is being matched against CLI.input which is just an ill-formed pattern.

[pchiusano]

Suspect that checkPattern needs to be made robust to the case where the pattern indicates a data constructor, but it's actually referring to an ability constructor.

[dolio]

Debug tracing shows that this is indeed caused by ending up with a data constructor with an ability reference in it.