Close out #322 residue: renderInstallCommand pin handling + planemo 0.75.45

[jmchilton]

Posted by Claude (AI assistant) on behalf of @jmchilton — not authored by them personally.

Closes the remaining residue from #322. Investigating it turned up that the issue's blocking premise was wrong, so items 1–3 were already resolved and only item 4 needed work.

The premise correction

#322 said the fork pin couldn't move to base planemo until a release carried galaxyproject/planemo#1636 (cli_metadata, output_schema). That PR merged 2026-05-14 and first shipped in 0.75.42 — two releases before the 0.75.44 the repo was already pinned to. Verified:

$ uvx --from planemo==0.75.45 planemo --help
  cli_metadata     Export structured metadata for Planemo CLI...
  output_schema    Export JSON Schemas for Planemo...

Items 1–3 (stale convert-nfcore cast, fork planemo.md in its bundle, fork SHA across vendored artifacts) were already done in earlier work; there are now zero references to the fork SHA or jmchilton/planemo anywhere in the tree. The pin note is corrected to say 0.75.42 rather than implying 0.75.44 was the first release to carry it.

Item 4 — renderInstallCommand (ef20f36)

renderInstallCommand unconditionally emitted <pkg>==<package_version>, which mangles any non-bare pin into invalid syntax. Base-planemo dodged it; the next VCS or range pin would not have.

Split out renderPackageSpec(pkg, version, origin), which classifies the pin first:

package_version	pypi	npm
1.2.3	pkg==1.2.3	pkg@1.2.3
>=1.2	pkg>=1.2	pkg@>=1.2
git+https://…	pkg @ git+https://… (PEP 508)	pkg@git+https://…

Specs are shell-quoted when needed, so a range pin isn't read as a redirect.

build-cli had no test harness at all — added vitest matching planemo-cli-meta's setup (tsc --noEmit && vitest run), with 9 tests covering the matrix above. The test earned its keep immediately: the first shell-safe charset omitted =, so bare pins came out quoted, which would have churned every existing cast bundle. Caught red, fixed, green.

Planemo rev to 0.75.45 (c3ad839)

Bumped content/cli/planemo/index.md and content/schemas/planemo-test-report.md (the latter's upstream: URL and Pin prose are live pins), regenerated vendored artifacts and the six CLI manual pages, recast the three planemo-pinned Molds, reassembled the pipeline harnesses, regenerated Dashboard.md.

Real deltas in 0.75.45: planemo test --failed gains an --lf alias and reads a new --failed_json; adds --shed_tool_data_table_config, --shed_data_manager_config, --shed_data_dir. The test-report JSON schema is byte-identical to 0.75.44.

Verification

make check exits 0 (validate 0 errors, 150/150 tests, generated + pipeline-assembly drift gates clean). packages-test, packages-lint, packages-format all clean.

Follow-ups (not in this PR)

• foundry-build cast --check creates its destination directory despite being read-only — a repo-wide drift scan leaves empty dirs behind and breaks tests/plugin-packaging.test.ts. Filed separately.
• The planemo pin is hardcoded in three sync scripts whose own comments name content/cli/planemo/index.md as the source of truth, and they stamp PIN.release into provenance rather than the version actually invoked — so a sync run against a mismatched PATH planemo produces artifacts that misreport their own provenance. Hit this during the rev. Worth fixing next.

🤖 Generated with Claude Code