How to download Ivy dependencies "explicitly" taking the Scala version into account

[hmf]

For

  override def scalaVersion = T{ "3.1.3" }

I have the following code:

  def mDocLibs = 
      T{
        import coursier._
        val files = Fetch().addDependencies(dep"org.scalameta::mdoc:2.3.3").run()
        val pathRefs = files.map(f => PathRef(os.Path(f)))
        Agg(pathRefs : _*)
      }

In the project's module I also have:

  override def ivyDeps = Agg(
      ivy"org.tensorflow:tensorflow-core-api:${tensorflowVersion}",
      ivy"org.tensorflow:tensorflow-core-api:${tensorflowVersion};classifier=${javaCPPPlatform()}",
      ivy"org.scalameta::mdoc:2.3.3"
  )

If I compare the output (just focusing on MDoc):

    val rp = mDocLibs().map(_.path)
    val cp = runClasspath().map(_.path)

    val s1 = Set(cp.toList:_*)
    val s2 = Set(rp.toList:_*)
    val s = s1 -- s2

I see that for the explicit download I have:

home/usercache/coursier/v1/https/repo1.maven.org/maven2/org/scalameta/mdoc_2.13/2.3.3/mdoc_2.13-2.3.3.jar

and its dependencies. But for the standard ivyDeps I get:

/home/user/.cache/coursier/v1/https/repo1.maven.org/maven2/org/scalameta/mdoc_3/2.3.3/mdoc_3-2.3.3.jar

and its dependencies. So what Mill API can I use to generate the ivyDeps equivalent taking the Scala version into account?

TIA

[lefou]

ivyDeps with a double-collon as the first separator will use the same Scala version that is configured for the module. You can use ivy"org.scalameta::mdoc:2.3.3".withDottyCompat(scalaVersion()) to use a compatible Scala 2.13 dependency.

[hmf]

@lefou thanks for the feedback. I am assuming that you are referring to the use of override def ivyDeps. However, I need to download these dependencies separately so that I can use them like this:

  override def mdoc : T[PathRef] = T {
  
    val rp = mDocLibs().map(_.path)
    val cp = runClasspath().map(_.path)

    val dir = T.dest.toIO.getAbsolutePath
    val dirParams = mdocSources().map(pr => Seq(
        "--classpath", toArgument(cp),
        "--in", pr.path.toIO.getAbsolutePath, 
        "--out",  dir)
      ).iterator.flatten.toSeq
  
    Jvm.runLocal("mdoc.Main", rp, dirParams)
  
    PathRef(T.dest)
  }

Note how the cp uses ivyDeps but rp does not. rp should just contain the dependencies of one specific library - MDoc. The use of Coursier's Fetch().addDependencies does not take the Scala version into account. I would like to use Mill to identify, download and report on the dependencies as it does using ivyDeps to generate the runClasspath.

Hope this makes sense.

Just as an aside. I can use the ivyDeps with the double colon and do not require to add .withDottyCompat(scalaVersion()). It works perfectly. 8-)

[hmf]

I am one step closer. If I use:

        val files = Fetch().addDependencies(dep"org.scalameta:mdoc_3:2.3.3").run()

So the question is, how does Mill to handle this Scala 2 versus 3 issue? Is it some Coursier parameter? I can see Coursier also handles the :: "interop" format but still have not a way to use it correctly.

[lefou]

In the end, it's just some convenience to add a _2.13 or _3 suffix to the artifact name.

If you look into how ivyDeps are resolved (hint: resolvedIvyDeps), you'll find CoursierModule.resolveDeps, which you also can use directly to implement mDocLibs.

[hmf]

Exactly what I was looking for. Having difficulty tracking this down via the Git's search. Worked like a charm. Thank you.