Detail UnapplyInvalidReturnType error message

[mbovel]

Investigated during today's issue spree with @natsukagami and @TheElectronWill. Fixes #17077.

The reference documentation on “Product Match” states:

• U <: Product
• N > 0 is the maximum number of consecutive (val or parameterless def) _1: P1 ... _N: PN members in U
• Pattern-matching on exactly N patterns with types P1, P2, ..., PN

From the second item:

• empty products as return types for unapply are not accepted (N > 0)
• a Product return type must have _1.._n members

Therefore, it seems that the current behavior is correct.

Side note: Product0 has been removed: #10398.

This PR adds 2 tests that confirm the current behavior, and adds details in the error message.

[Sporarum]

Looks good to me ^^

The only thing is the link, we should probably have a broader discussion in how to integrate general/spec information into error messages that are necessarily smaller than them

(I am not saying we should remove the link, just that we should make sure at a later time that links are really what we want to use)

[ckipp01]

The only thing is the link, we should probably have a broader discussion in how to integrate general/spec information into error messages that are necessarily smaller than them

(I am not saying we should remove the link, just that we should make sure at a later time that links are really what we want to use)

We're not utilizing them yet, but inside of DiagnosticCode there is an explanation field:

https://github.com/lampepfl/dotty/blob/274babf8cd6e1dcd5a14fed735570d20a6e2af3b/sbt-bridge/src/dotty/tools/xsbt/DiagnosticCode.java#L12

At least in LSP terms, this is typically a URI that leads to an error index where these types of things could also be explained. If we are going to start using links, it would be amazing to ensure they are structured in the places that they can be to start enabling using these types of features. However, if we were to do that, we'd need an error index first 😆

[mbovel]

If that sounds okay to you, I will merge this PR as-is for now as it has a much narrower scope.

We will hopefully tackle more structured diagnosis in the future as a separate issue.