Repository navigation
Release on a tag from the changelog, with the checks that were memory before - #15
Conversation
… before The release history lived in the README, the tag workflow created the GitHub release with generated notes only, and nothing checked that the tag named the version ida-plugin.json and config.py declare. The history moves into CHANGELOG.md unchanged, with a Keep a Changelog section above it; verify_metadata_sync.py reads the newest heading from there. Pushing a vX.Y.Z tag now refuses to continue unless the tag matches both version strings, the version has a changelog section, the commit is on main and CI passed on it; then runs the metadata, settings and quality checks, builds the plugin archive, lints it with hcli, and creates the GitHub release from the changelog section with the archive attached, through gh rather than a third-party action. Pre-release tags are marked as such. A PR that changes shipped plugin files has to touch CHANGELOG.md or carry the no-changelog label. Dependabot keeps the SHA-pinned actions current. CI runs on Python 3.12, the floor shared across the MCRIT ecosystem.
b7b1cfd to
ecfbb77
Compare
This branch moves the release history out of README.md and points verify_metadata_sync.py at CHANGELOG.md. The rebase onto main brought config.py and ida-plugin.json to 1.1.10, but 1.1.10's entry was written in README.md, which this branch deletes, so the newest changelog heading stayed at v1.1.9 and validate failed: Version mismatch: config.py VERSION does not match latest CHANGELOG.md release heading. Carries over the section already written on familiary#17, verbatim, so the two branches do not diverge, and updates the compare links to match. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
Pushed This branch moves the release history out of That's our doing, not yours - we released 1.1.10 straight to The commit carries over the Nice catch on the missing One thing that's ours to do before this merges: the |
The release history lives in the README, the tag workflow creates the GitHub release with generated notes only, and nothing checks that the tag names the version
ida-plugin.jsonandconfig.pydeclare —verify_metadata_sync.pychecks the three copies agree with each other, but av1.2.0tag on a tree that says1.1.9releases fine, under the wrong name.This is one of six pull requests giving the MCRIT ecosystem one release process — the same guard, the same tag-triggered workflow, the same changelog check, and a
RELEASING.mdin each repository stating only the values that differ. A maintainer learns it once.Changelog
The README "Version History" (11 releases) moves verbatim into
CHANGELOG.mdunder## Older releases, below a Keep a Changelog## [Unreleased]section; the README links to it.verify_metadata_sync.pyreads the newest release heading fromCHANGELOG.mdinstead of the README, accepting both the Keep a Changelog form and the carried-over### vX.Y.Z (date)form. Every PR that toucheshelpers/,widgets/,icons/,ida_mcrit.py,config.pyorida-plugin.jsonadds its own bullet or carries theno-changeloglabel; a newChangelogcheck enforces that (read-only token, no PR content executed).What pushing
vX.Y.Zdoes now.github/workflows/release.yml:ida-plugin.jsonandconfig.py,CHANGELOG.mdhas a## [X.Y.Z] - <date>section (written out as the notes), the commit is an ancestor ofmain, and thepytestworkflow concludedsuccesson that commitverify_metadata_sync.py --expected-version,verify_settings_sync.py,run_quality_checks.py,package_plugin.py,hcli plugin linton the tree and the ZIPgh release createwith the changelog section as body,--generate-notesappended,mcrit-ida-<version>.zipattached. Pre-release tags are marked--prereleaseand not--latestsoftprops/action-gh-releaseis replaced by the runner's ownghin this workflow: one fewer third party in the path that signs and announces a release. TheBuild Offline Dependenciesworkflow still runs from the published release, as before. The guard script has tests (tests/test_release_guard.py).Security
build-offline-dependencies.ymlexpandedgithub.event.release.tag_namedirectly inside PowerShell scripts in a job withcontents: write— a tag name is attacker-influenced text. The values now reach the scripts throughenv. The checkout there no longer keeps its credentials on the runner. This was the onezizmorerror in the repository; the remaining informational findings are inida-tests.yml, which is manual-dispatch only and out of scope here.ruff.ymlgainspermissions: contents: readandpersist-credentials: false; the new workflows follow the same pattern.dependabot.ymlkeeps the SHA-pinned actions current.zizmor --persona=regularon the workflows touched here: no findings;actionlintclean.Python
CI and the release workflow run on Python 3.12, and
ruff'starget-versionmoves frompy38(whichAGENTS.mdsaid was for formatting only) topy312. The ecosystem shares a 3.12 floor now;smda, which the plugin needs in IDA's interpreter, requires 3.12 from its next release, and IDA 9 bundles 3.12. Nothing in the plugin needed it.Tests
pytest;ruff checkandruff format --checkclean.verify_metadata_sync.py --expected-version 1.1.9,verify_settings_sync.pyandrun_quality_checks.pypass on the tree;package_plugin.pybuilds the ZIP andhcli plugin lint(0.23.0) passes on both the tree and the ZIP.What a maintainer has to configure once
no-changeloglabel. Nothing else.Compatibility
The plugin ZIP, its contents and the HCLI metadata are unchanged. The next release will be the first with a
## [X.Y.Z] - <date>section; until then a tag push fails at the guard with a message saying so.Related pull requests across the ecosystem
The same process, one PR per repository. None depends on another to merge; release order once merged is smda → purepdb → picblocks → MCRIT → MCRITweb → docker-mcrit.