Skip to content

RELEASE.md

Latest commit

 

History

History
91 lines (80 loc) · 4.17 KB

RELEASE.md

File metadata and controls

91 lines (80 loc) · 4.17 KB
Raw

Release Checklist

MeshCraft ran a long stabilization phase (policy archived at docs/history/STABILIZATION.md) that is now substantively complete — gates A-E satisfied, remaining plan.md items are owner-gated/deferred, not active defect work. This checklist is for when a release is actually being cut; until then, treat it as the bar to clear, not a completed process.

Build clean

  • cmake-build-debug/ (Debug): full reconfigure (CLion's bundled cmake — see CONTRIBUTING.md) + full rebuild, 0 errors/warnings of note. sh "$CLION_CMAKE" -S . -B cmake-build-debug -DBUILD_TESTING=ON ninja -C cmake-build-debug
  • Release build (fresh directory, system cmake is fine): full configure + build. sh cmake -S . -B b-release -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=ON \ -DFETCHCONTENT_UPDATES_DISCONNECTED=ON cmake --build b-release -j4
  • Offline mode builds cleanly (-DFETCHCONTENT_UPDATES_DISCONNECTED=ON with dependencies already fetched from a prior build).
  • Each CNA-free component builds standalone, without the root project: mc3/, mcb/, mc3togltf/, mc3tomcb/ (see TESTING.md's standalone-build loop).

All tests pass

  • ctest --output-on-failure in cmake-build-debug/: all tests pass — run ctest -N for the current count (95 as of 2026-07-11; this number grows over time, do not hard-code it here).
  • Same in the Release build directory.
  • Each standalone component's own test(s) pass — check with ctest -N -L <label> (format/export/unit/lint/ai/ commands/registry/render) rather than trusting a hard-coded per-component count here.
  • No test was skipped due to a missing optional dependency (SQLite3, OpenSSL, LibXml2) that should actually be present on the release build machine — check the CMake configure log for "not found" messages for any of these.

Docs current

  • NEXT.md reflects actual current state (status/blocker/next task) — not stale from a prior session.
  • plan.md is internally consistent — run python3 test/validate_plan_consistency.py . (add a build directory and --run-tests to also catch live CTest drift, not just registered counts). plan.md (SYS-W13-01) is a bounded active plan using [DONE]/[IN_PROGRESS]/[BLOCKED]/[DEFERRED]/[PROPOSED] text markers on each row, not an emoji-marker summary table — that older format was retired when plan.md was cut down from 2,743 lines; historical detail lives in docs/history/.
  • README.md's "Current Features"/"Current Limitations" reflect what actually works — spot-check a few claims against the code, don't just trust the prose (this has caught real staleness before — see NEXT.md's history for an example).
  • MC3_FORMAT.md covers every element mc3.xsd declares (no N1-N7 gaps).
  • mc3.xsd declares every attribute/element Mc3XmlWriter.cpp actually writes (see CONTRIBUTING.md's "schema/writer symmetry" note — this has been a real, recurring source of bugs).
  • TESTING.md's test count and per-test descriptions match ctest -N's actual output.

Sample files valid

  • Every test/*.mc3.xml fixture validates against mc3/mc3.xsd (covered by the xsd_validation ctest, but worth a manual spot check if any fixture changed recently).
  • mc3togltf exports at least one representative scene (test/features.mc3.xml or similar) to .glb and the result opens cleanly in a glTF viewer (not just "exit code 0" — a malformed-but-technically-valid GLB can still fail to render).
  • mc3tomcb round-trips a representative scene (.mc3.xml → .mcb → .mc3.xml) losslessly.

Before tagging

  • git status clean, all intended changes committed.
  • develop (or the release branch) pushed and in sync with the remote.
  • No secrets committed (API keys, tokens) — this project has had a PAT-in-.git/config issue before; double-check nothing similar snuck into a tracked file.