Widen parLiftN's constraint to NonEmptyParallel

[kubukoz]

Follow-up to #4340. The current state doesn't allow for using those functions on types that don't have a pure, such as ciris's ConfigValue (which is something where people should almost always use the parallel variant...)

[armanbilge]

Unfortunately we have no MiMa support here, because syntax classes are private (which MiMa correctly ignores).

It's arguably a bug that methods on private classes are publicly usable.

• Can we please reach some kind of decision on access scala/bug#6794

[kubukoz]

okay, I was able to follow your advice from Discord and just created a new copy of the syntax class, except the old implicit conversion has been made non-implicit.

I tested it with the following:

main.scala:

//> using dep org.typelevel::cats-core:2.13.0
import cats.syntax.all.*
case class Foo(a: Int, b: String)

object demo extends App {
  def foo(a: Either[String, Int], b: Either[String, String]) = (Foo.apply _).parLiftN(a, b)

  println(foo(Left("error"), Right("hello")))
}

run.sh:

#!/usr/bin/env bash

set -e
set -x

for sv in 2.12 2.13 3
do
  echo "compiling $sv"

  if [ "$sv" == "2.12" ]; then
    EXTRA_FLAGS="-Ypartial-unification"
  else
    EXTRA_FLAGS=""
  fi

  scala-cli compile main.scala --output-directory out --scala "$sv" -Xsource:3 $EXTRA_FLAGS
  DEPS=$(cs fetch org.typelevel:cats-core_$sv:2.13.0-4-2d82b53-SNAPSHOT --classpath)

  echo "running: "
  java -cp "$DEPS:out" 'demo'
done

It succeeds in all versions. Does that sound about right?

[satorg]

I mean... parLiftN and friends just have been released... Can we consider risking it and ignore the binary incompatibilities there? We could quickly arrange 2.13.1 and recommend to hold on upgrading to 2.13.0 😅

Maybe no one even notice :)

[kubukoz]

Your call 😅

[satorg]

fyi: #4704

[armanbilge]

Actually, this PR should bump to 2.14. It's making source-breaking changes and forwards binary-breaking changes, with respect to Cats v2.13.x so we need to bump the minor to reflect that.

[kubukoz]

How are the changes source-breaking?

[satorg]

@kubukoz , I'm wondering too: #4704 (comment)
Perhaps there's some mystery to be unraveled :)

[armanbilge]

How are the changes source-breaking?

This code compiles in 2.13, but won't after these changes.

object syntax extends cats.syntax.FunctionApplySyntax
import syntax.*
fn.liftN(...)

[kubukoz]

hmm I wonder if we can make the old syntax trait extend the new. Still, if someone hand-picks their imports they'll be source-broken. Do we care about this use case?

I don't suppose we'll release a 0.14 soon...

[armanbilge]

Still, if someone hand-picks their imports they'll be source-broken. Do we care about this use case?

Yes, that's the point. Whenever we make any change like this, it is technically source-breaking.

To clarify, I'm not saying we should avoid source-breaking changes: this makes it impossible to add new features. What I'm saying is that, because these typical changes are source-breaking, we have to bump the base version.

[kubukoz]

16 months later, might be a good time for 2.14 😆 bumped baseline to 0.14, let's see if this still builds.

[kubukoz]

Let's get this one done