WIP: skip -SNAPSHOT suffix if publishing to new portal

[kubukoz]

Just a draft for #735 to kickstart the discussion.

We don't have any scripted tests here, do we?

[armanbilge]

Sonatype Central does not accept -SNAPSHOT versions.

The Central Publisher Portal does not support -SNAPSHOT releases, and deployments of -SNAPSHOT releases will have a version cannot be a SNAPSHOT error in their validation results. Versions should only be published to the Portal if they are intended to reach Maven Central.

[armanbilge]

Oh sorry, I missed the ! :)

Anyway, I feel it's disingenous for isSnapshot to be true but result in a version lacking the -SNAPSHOT.

[armanbilge]

Looking at how the setting is defined ...

https://github.com/sbt/sbt/blob/2f936295f158117baf9397ce8e9487bf9be8c87e/main/src/main/scala/sbt/Keys.scala#L520

[armanbilge]

Mmm, also disingenuous if isSnapshot is true but it's not published to the snapshot repo?

https://s01.oss.sonatype.org/content/repositories/snapshots/

[armanbilge]

Here was my idea:

• Disable CI snapshots if publishing to Central host #739

I guess the question I have is what does isSnapshot mean? Can something be a snapshot even if the version doesn't end in -SNAPSHOT and it's not published to a "snapshot" repository? Can you publish to Maven Central and call the artifacts snapshots?

[armanbilge]

Pinging @majk-p @bpholt @j-mie6 who participated in the Discord discussion. Thoughts?

[j-mie6]

My main thought is we should be careful about publishing anything that build tools may consider as a legitimate release. I could only imagine how annoying it would be to get warnings about updating to snapshots...

[majk-p]

I guess the question I have is what does isSnapshot mean?

I have the same question. When using sbt-typelevel for my projects, with regards to publishing, I always valued that the setup works out of the box. By "works" I mean two main use cases:

1. When merged to main, CI pipeline produces a "snapshot" - something I can test, use where latest features are needed
2. When tagged on main, CI produces a proper version, such that will be picked by tools like Scala Steward

In that sense isSnapshot is something that's not picked up by Scala Steward.

Following that for a second, let's have a look at isSnapshot defined for Scala Steward:

  private def isSnapshot: Boolean =
    components.exists {
      case a: Version.Component.Alpha => a.isSnapshotIdent
      case _: Version.Component.Hash  => true
      case _                          => false
    }

It defines snapshot as anything that has a hash or is a special case of Alpha release, so:

    final case class Alpha(value: String) extends Component {
      def isPreReleaseIdent: Boolean = order < 0
      def isSnapshotIdent: Boolean = order <= -7
      def order: Int =
        value.toUpperCase match {
          case "SNAP" | "SNAPSHOT" | "NIGHTLY" => -7
          case "FEAT" | "FEATURE"              => -6
          case "ALPHA" | "PREVIEW"             => -5
          case "BETA" | "B"                    => -4
          case "EA" /* early access */         => -3
          case "M" | "MILESTONE" | "AM"        => -2
          case "RC"                            => -1
          case _                               => 0
        }
    }

Back to sbt-typelevel, the current workaround with setting tlUntaggedAreSnapshots := false works fine because non-snapshots from mail look like: 0.1.2-1-4bce857 so they are snapshots as far as Steward is concerned.

IMO we have two main options for handling publishing to Central Portal

1. Agree that snapshots don't exist there and thus tlUntaggedAreSnapshots := true makes no sense. Instead set it to false as a default if central portal is configured. If user explicitly sets it to true, throw an error explaining what's going on
2. Overcome the fact that -SNAPSHOT suffix is not accepted by Central and in such case go with -SNAP or -NIGHTLY

I think I like the first option slightly better, but no hard opinion.

[aartigao]

If it may be useful as a feedback, in my opinion, being a maintainer that came late to the party and is only able to publish to Central Portal, I must accept that the rules have changed and I have no option for publishing SNAPSHOTs anymore.

To me, publishing SNAPSHOT stuff in Maven Central is counterintuitive. At least in my case, if I want users to test something before final release, I'll just build an RC. So, from the options enumerated by @majk-p, I'd choose 👇🏽

Agree that snapshots don't exist there and thus tlUntaggedAreSnapshots := true makes no sense. Instead set it to false as a default if central portal is configured. If user explicitly sets it to true, throw an error explaining what's going on

[armanbilge]

1. Agree that snapshots don't exist there and thus tlUntaggedAreSnapshots := true makes no sense.

SNAPSHOTS do continue to exist, just not on Sonatype/Maven. For example, it's reasonable to publishLocal a -SNAPSHOT (I believe this is a concern for @j-mie6). So the setting tlUntaggedAreSnapshots := true can still be useful.

That's why in #739 instead of forcing it to false, we disable the CI publish workflow from running on untagged commits.

[armanbilge]

Btw, thanks everyone for your thoughtful comments, I appreciate it :)

[majk-p]

SNAPSHOTS do continue to exist, just not on Sonatype/Maven

You're right, I meant that only in the context of newly created Scala projects that will likely go with Sonatype.

That's why in #739 instead of forcing it to false, we disable the CI publish workflow from running on untagged commits.

That's something I'd like to avoid, since (if I get it right) eliminates the first use-case I mentioned above. Is that correct?

[armanbilge]

@majk-p

That's something I'd like to avoid, since (if I get it right) eliminates the first use-case I mentioned above. Is that correct?

If tlUntaggedAreSnapshots := true, then we disable the CI publish workflow from running on untagged commits.

However, if tlUntaggedAreSnapshots := false, then we continue publishing every commit as usual. Is this the use-case you had in mind?

[majk-p]

Is this the use-case you had in mind?

Yes, that's right. I guess my real questions should be what's the default tlUntaggedAreSnapshots for new projects (using central). Asked the other way around, is sbt-typelevel publishing untagged artifacts out of the box for new projects?

[armanbilge]

Asked the other way around, is sbt-typelevel publishing untagged artifacts out of the box for new projects?

It's not. My concern is that it can confuse tooling so it has to be opt-in. FS2 used to publish like this and I remember having some problems. Still, it looks like Steward and Scaladex are able to handle it okay so maybe it's less of a problem now?

[armanbilge]

I think I want to move ahead with #739 but I'm curious if there are any more thoughts.

[kubukoz]

Like they say, the best way to find out the answer is sometimes to give the wrong answer first :D