Update TypeMismatch for multiple possible types.

[bsermons]

Per comment in paf31/purescript-foreign-generic#11

[paf31]

I think we need to use a non-empty array, for this reason.

[garyb]

I agree, but just to say, purescript-nonempty is in contrib currently and this is core.

[paf31]

Thanks!

[paf31]

An alternative here is to move the non-empty array to the outer level:

type F = Either (NonEmpty ForeignError)

We probably don't want a dependency on any non core library for NonEmpty though, so we should think carefully about that.

[bsermons]

I can see the value of adding NonEmpty at the top level since it would make it good for use with validating the whole structure but is there a reason it couldn't exist at the TypeMismatch level also?
Seems like it would be hard to get a nice message like the 'expected one of [...], found ...' with it only at the top level.

What are the requirements for a package to be in core vs contrib?

[paf31]

The Show instance looks good thanks. I think we just need to come to some conclusion regarding NonEmpty.

[bsermons]

Yea I just did the easy thing first since I wasn't sure if NonEmpty being in contrib is a show stopper or if there was some other alternative than to using NonEmpty.

I'll continue ahead with it if you don't think it'll be a problem and see how far I can get with the suggestion above with moving the dependency to the outer level.

[bsermons]

Played around with using type F = Either (NonEmpty ForeignError). You can see what it would look like here (purescript-foreign) and here (purescript-foreign-generic).

The last commit on each is with TypeMismatch String String and the parent commit is with TypeMismatch (Array String) String.

Some notes:
End up using unsafePartial to go from SigProd's constructors array to NonEmpty
Wasn't sure what to use as the expected type when SigProd's constructors were empty, so just used the name.
Lose the nicer message when using the simple String String constructor.