[cchlanger]

gerlib moves to Scala 3.9.0, the long-term support line, and to sbt 1.13.1. Its libraries and sbt plugins move to their newest releases within their major versions. The zarr module goes, with jzarr and Unidata's repository, which only it needed. CI runs the scalafmt and scalafix checks, which it had skipped until now, and tests on Java 21 only. looptrace's move to Scala 3.9 (gerlichlab/looptrace#583) needs this release.

Merge with a merge commit, not a squash or a rebase. The tag v0.7.0 is already pushed, on the version commit fb2edc6, and looptrace's submodule points at that commit. A merge commit puts it on main. A squash or a rebase would leave both on a commit that isn't on main.

Closes #30.

Breaking

• Scala 3.9. An older Scala 3 compiler can't read a library compiled with 3.9, so a project that depends on v0.7.0 needs Scala 3.9 too.
• No zarr module, and no readRoiData or getOriginZYX in roi. A project that uses them can stay on v0.6.0. The one other user is reparafil, which pins v0.3.0 and Scala 3.5.2.

Why the zarr module goes (#30)

• Unidata: jzarr 0.4.2 brings Unidata's netCDF library, edu.ucar:cdm-core, and jblosc. cdm-core isn't on Maven Central, so every build had to name Unidata's repository.
• Every user got it: roi depended on zarr only for readRoiData, so every user of roi or io brought all three. looptrace's JAR carried about 1,800 of their entries, and its build named Unidata's repository too.
• CI broke on it: on 7 November, Java on GitHub's macOS runner rejected that repository's certificate, and looptrace's CI failed while publishing gerlib.
• Nothing needs it: nothing else in gerlib used the module, and no test covered it. jzarr reads only format-2 stores, while looptrace's stores are format 3.

#30 proposed moving readRoiData into zarr. That would have kept Unidata's repository in gerlib's own build. It would also have stayed in looptrace's CI and image, which publish every module.

Other changes

• Scala 3.6.4 → 3.9.0.
  • Under -source:future, 3.9 refuses implicit val and wants explicit implicit arguments passed with using. Each test suite's PropertyCheckConfiguration is now a given of the suite, which the compiler prefers over the inherited one.
  • The stricter unused-symbol checks removed unused imports and parameters. Among them is the A: Order bound of max on coordinates, which now asks for less.
• sbt 1.12.13 → 1.13.1.
• Libraries: iron 3.4.1, upickle and ujson 4.4.3, pureconfig 0.17.11, mouse 1.4.0, scala-graph 2.0.3, fs2-data-csv 1.14.1, fs2-io 3.14.0 and os-lib 0.11.10. For the tests: scalatest 3.2.20 and scalacheck 1.20.0.
• iron's RefinedSubtype: NonnegativeInt, NonnegativeReal, PositiveInt, PositiveReal and Distance now extend it instead of RefinedType. They stay subtypes of what they refine, as they were before iron 3.1.0, so a PositiveInt is still an Int.
• Untraced reads: since 4.4.0, upickle traces read by default, and a traced error's message is only the JSON path. DistanceLike's codec reads untraced, so its error still describes the problem with the value. A new test pins this.
• Plugins:
  • sbt-assembly 2.5.0;
  • sbt-github-actions 0.32.2, which moves the workflow to actions/setup-java v6;
  • sbt-scalafmt 2.6.2, sbt-buildinfo 0.13.2 and sbt-scalafix 0.14.9;
  • typelevel's scalafix rules 0.6.0;
  • scalafmt 3.11.5.
• CI's scalafmt and scalafix checks run. Their condition was never true: it compared the runner's OS with the label ubuntu-latest, and tested runner.java-version, which doesn't exist. It now tests runner.os == 'Linux' and the matrix's Java entry.
• CI tests on Java 21 only, on Linux and macOS, the Java looptrace builds and runs gerlib with. The Ubuntu job on Java 17 is gone.
• CI installs the sbt launcher the build asks for, 1.13.1, rather than setup-sbt's default, the newest one, so a new sbt release can't change how CI starts the build.

These come after the tag, so the changelog lists them under "Unreleased".

Commits

• d4d1c0f sbt 1.13.1.
• 4b05638 Scala 3.9.0.
• 7e7bbec the libraries.
• dc00a2e the plugins and scalafix rules.
• e9bf4f8 scalafmt 3.11.5, which aligns six case arrows anew.
• b1884f4 the CI checks run.
• e73ed45 no zarr module.
• fb2edc6 version 0.7.0, tagged v0.7.0.
• 00f4c7b CI tests on Java 21 only.
• 98a7dae CI installs the sbt launcher the build asks for.
• a03c006 the changelog says why the launcher is pinned.

Checked

These checks ran on Java 21 (Zulu 21.0.11):

• The tests: 69 passed, 1 pending.
• scalafmt, scalafix and the workflow check are clean.
• From a fresh clone, with an empty dependency cache: sbt "assembly; publishLocal", the command looptrace's Dockerfile runs, publishes the 12 modules. Nothing is downloaded from Unidata.
• looptrace on this release (cl/jvm_updates, for gerlichlab/looptrace#583):
  • its Scala suite passes, with 160 tests passing and 47 pending;
  • its JAR holds no jzarr, cdm-core or jblosc.

Before merging

• CI green, including the first real run of the scalafmt and scalafix checks.

[valesori]

Ok for me to merge!