Please publish for Scala 2.13.14

[SethTisue]

...which is now on Maven Central, as per https://contributors.scala-lang.org/t/scala-2-13-14-release-planning/6581/13

[rossabaker]

I tried with the back-publish script, but ran into java.lang.NoClassDefFoundError: Could not initialize class sbt.internal.parser.SbtParser$.

[armanbilge]

Worked fine for me 🤷

https://repo1.maven.org/maven2/org/typelevel/kind-projector_2.13.14/

[TonioGela]

I tried with the back-publish script, but ran into java.lang.NoClassDefFoundError: Could not initialize class sbt.internal.parser.SbtParser$.

FWIW this happens to me when using JDK21 with an old version of sbt. If you update to 1.10.x it should be good to go.

[SethTisue]

If you update to 1.10.x it should be good to go

PR: #281

[rossabaker]

That's helpful, but the old version of SBT is in the tag that we are backpublishing from. 0.13.3 caused a lot toil downstream, so there's a lot of value in avoiding a 0.13.4. Right now, the only impure thing about backpublishing is providing $SCALA_VERSION as an argument to the script. I don't see a way to upgrade SBT without a new tag or brazen irreproducibility.

Next time, we could tag a v0.13.3_2.13.15 or some such, and have that include #281. That would knock off the bitrot, not make filthy liars of us, and still produce a 0.13.3. My fear there is that tag format interfering with some other tooling.

Alternatively, we can be more careful to use jdk8. That's what the flake.nix is all about, and I must have overlooked direnv screeching at me that I was on a new machine. And then #281 is there on main for whatever future problem triggers 0.13.4.

[SethTisue]

fwiw, in my own repos elsewhere, I often backpublish from HEAD rather than from the actual tag if I look at the diffs since the tag and I can see that the only things that have been merged since the tag are immaterial-to-the-end-user things like the sbt version.

This may make me a "filthy liar" :-)

[rossabaker]

My problem with that is it's hard to nail down what's "immaterial". Which Scala 2.13 something was built with used to be immaterial. And then, before bidirectional compatibility is broken, a Scala patch version in a dependency may now manifest as a warning in the latest SBT. That was a good change in SBT, to insulate us against the more severe breakage that is to come. But it is also something that retroactively became material.

Since this plugin is fully cross-built, even that particular change is immaterial to this project, but it underscores why I'm squeamish any divergence from the tag.

[armanbilge]

Alternatively, we can be more careful to use jdk8.

Mm, yes. I was following the release instructions in the README which is why I didn't have that problem.