Pattern matching DSL

[serras]

This PR proposes a small DSL to describe pattern matching. This is based on three ideas:

1. Pattern matching is essentially as attempting to match several Optionals in a row,
2. We can build optionals that match on more than one value (that is, Optional<S, A> + Optional<S, B> = Optional<S, Pair<A, B>>
3. The usual way to define patterns looks a lot like function application, so we can reuse invoke to simulate them.

This is an example from the code. Note that then separates an optional that matches and the code that uses the information being matched as result.

val User.name: String get() = this.match {
  // Person(Name(firstName = fn), age if it < 18)
  User.person(Person.name(Name.firstName), Person.age.suchThat { it < 18 }) then { (fn, _) -> fn }
  // Person(Name(firstName = fn, lastName = ln)) -> "Sir/Madam $fn $ln"
  User.person(Person.name(Name.firstName, Name.lastName)) then { (fn, ln) -> "Sir/Madam $fn $ln" }
  // Company(name = nm, director = Name(lastName = d))
  User.company(Company.name, Company.director(Name.lastName)) then { (nm, d) -> "$nm, att. $d" }
}

[github-actions]

Kover Report

File	Coverage [40.00%]
arrow-libs/optics/arrow-optics/src/commonMain/kotlin/arrow/optics/match/Combinators.kt	32.50%
arrow-libs/optics/arrow-optics/src/commonMain/kotlin/arrow/optics/match/Match.kt	60.00%

Total Project Coverage	60.80%

[nomisRev]

Cool stuff! Never actually used Optics for pattern matching, knew it was theoretically possible. Can we also supports sealed classes through Prism?

[serras]

Can we also supports sealed classes through Prism?

We do! It's there in the example :P

[nomisRev]

Oh yes, of course! It's relying on Prism : Optional 😁
Wow, that is so cool! 👏 👏 👏

[nomisRev]

SHIP IT!!

[nomisRev]

Should we force a default lambda, and provide a second function matchOrThrow without the default lambda?

This way it's less likely to forget to catch IllegalArgumentException, and if you have no default you are forced to be more explicit.

[kyay10]

Haven't taken a full look at the PR yet, but this should definitely have an integration with SingletonRaise so that it fails similarly to Monad.fail from Haskell

[serras]

I've just pushed a new commit which I think covers both of your comments:

• The main function is now matchOrElse, which gets an additional argument telling what to do if no match is found;
• From those matchOrThrow, matchOrRaise, and matchUnit are derived.

I like the possibility of having a catch-all case within the DSL, since it mirrors how one usually defines pattern matching.

value.match {
  User.company.address(Address.street) then { s -> ... }
  default { "Unknown address " }
}

[nomisRev]

I like the possibility of having a catch-all case within the DSL, since it mirrors how one usually defines pattern matching.

Yes, that looks great, but I felt that the strategy should be explicit in the function names. Thank you, great improvement!

[raulraja]

Late to the party but wanted to say this looks great 👏