0.19: turn `.transform` (on services) into an extension method

[kubukoz]

The rationale is: sometimes you have multiple algebras that you want to combine into one:

val f: Alg1[IO] = ???
val f2: Alg2[IO] = ???

val fCombined: Alg[IO] & Alg2[IO] = new {
  export f.*
  export f2.*
}

For example, when using BSP, you might want to have a single client for BuildServer and ScalaBuildServer. This doesn't work, because both algs have a different signature of def transform: Scala refuses to export it, and it might not even be possible to implement yourself.

If we make it an extension method, or have users call Transformation.of(...) themselves, we can support this usecase - the only problem is that you can't transform such a combined client, but I doubt users of combined algebras will care.

[kubukoz]

Implementation notes

You should start by creating a new branch forking from series/0.19 - this is a change that'll break compiled code in a very binary-incompatible way, so we shouldn't do it in a patch release.

This is a change that will have to happen in the code generator.

This part currently renders the def transform:

smithy4s/modules/codegen/src/smithy4s/codegen/internals/Renderer.scala

Line 361 in 618b31d

line"final def $transform_: $Transformation.PartiallyApplied[$genName[F]] = $Transformation.of[$genName[F]](this)"

so the function will have to be rendered in an implicit class in the companion object.

Testing (manual)

Run the sbt task bootstrapped/managedSources - the files generated by this module are tracked by Git, so you'll see the diff immediately after the task completes.

Compile these sources with bootstrapped/compile to make sure they're still valid Scala.

Testing (automated)

Run bootstrapped/test (or bootstrapped/testOnly smithy4s.TransformationSpec) - TransformationSpec directly tests the method in question.

Extra notes

Here's the documentation for the feature: https://disneystreaming.github.io/smithy4s/docs/guides/smithy4s-transformations/

[Jay-Lokhande]

Hey I tried this issue, here is my fork

[stewSquared]

@Jay-Lokhande Is this work still active? Would you like to open a PR?

[kubukoz]

There is one - #1735 - we just need to make a decision on whether an extension method or a "static" (companion member) method is better. Both have the same type inference problems that the current state does not.

[kubukoz]

also, update: it actually is impossible to implement transform - it's a final method since #1588 (by yours truly). So yeah, this is still solving a real problem - the inability to compose algs with inheritance.