Report the OSATE and language-server build in osate-cli help

[lwrage]

Closes #5.

$ osate-cli help
osate-cli 0.1.0
  language server 0.1.0.v20260902-1320 (7fbfec9)
  OSATE           2.19.0.vfinal (4256148)

Why it is baked in rather than discovered

None of this is available at runtime, which is what shaped the design. OSATE bundles carry independent versions — org.osate.aadl2 is 6.1.1, not 2.19.0 — and no bundle manifest records a commit. scripts/build-test-release is the only place that knows both the osate2 gitlink and the built language-server bundle version, so it supplies all four values and Maven filters them into version.properties beside the existing version.

osate.commit is the gitlink, not the submodule's HEAD: it is the reviewed pin, and the build already refuses to proceed when the two disagree.

--version is deliberately untouched

Its output is compared for exact equality by CliEndToEndIT and parsed by the packaging scripts' --expect-version check. So versionLine() stays one line and a new versionDetail() carries the provenance into help() only. A test pins that, because appending to the version line is an obvious temptation and the breakage would show up somewhere unrelated.

-dirty, and why it warns rather than fails

ls.commit gains a -dirty suffix when the tree had uncommitted changes — without it the commit would appear to identify code that is not what was compiled. The superproject is allowed to be dirty during development, unlike the submodule, so marking beats refusing. Both the suffix and the commit are captured before any build runs, so they describe the sources that went in. build-provenance.properties gets the same treatment for tooling.commit, which had the identical problem.

abbreviate() shortens only the leading hex run, so <sha>-dirty displays as 7fbfec9-dirty rather than dropping the marker that says the value cannot be trusted, and unknown is returned untouched instead of truncated into something that reads like a commit.

Release guards

The values only appear when the build goes through the script, so:

• build-release-artifacts.sh refuses to package a distribution reporting unknown — a hand-built CLI cannot be released claiming a provenance it does not have.
• It also refuses one whose recorded osate.commit disagrees with the current gitlink, extending to the artifact an invariant the source build already enforces.

A dirty tree only warns there: the suffix is already visible in the jar and in help, so the artifact is not misleading, and failing would block the local packaging smoke tests packaging/README.md documents.

Verified

• --version unchanged; help shows all three lines.
• Guards exercised all three ways: accepts a correct dist (with the dirty warning), rejects an unknown-provenance jar, rejects a gitlink mismatch.
• VersionTest covers both provenance paths — it skips the "reports unknown" case when the build supplied values and runs it when it did not, and both were exercised across a scripted and a standalone build.
• Documented in OSATE-CLI.md, with four cases added to the CLI manual test plan.

Incidentally removes a Maven invocation: osate.version is now read from the pinned submodule's pom, which the --skip-osate precondition check was already doing separately.

🤖 Generated with Claude Code