Repository navigation
chore: cut v4.8.0 - #348
Conversation
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
📊 SMDA Performance Evaluation Benchmark ResultsGenerated on: 2026-09-14 14:51:22 Summary
Per-side Timing Context (best of 3/3 runs per file)
Paired per-file timing (positive speedup = PR faster):
Determinism (self-check across repeated runs)
Pairwise run matrix (individual run medians, diagnostic)Diagnostic only: each row compares one raw PR run against one raw base run. The headline verdict above uses paired per-file best-of-runs timings.
|
The release the
v4.8.0rc1rehearsal was for. No code changes:__version__moves to4.8.0and the candidate's section becomes## [v4.8.0] - 2026-09-14.What it carries
Four merged PRs, one topic: releasing is now something the repository does rather than something a person remembers.
Added — pushing a
vX.Y.Ztag builds, publishes to PyPI through trusted publishing, creates the GitHub release from this version's changelog section, and closes the milestone; five gates refuse to publish otherwise (#343).Changed —
SmdaConfig.VERSIONread fromsmda.__version__so a release bumps one line, and the PEP 639 licence expression with the project URLs (#343).Deprecated — the Python 3.11 policy: the floor follows MCRIT's and the two move in the same wave (#343).
Fixed — a pre-release version is ordered on report import instead of being treated as no version at all (#345).
Removed,SecurityandCompatibilityare dropped as empty. No compatibility constant moved, verified against thev4.7.0tag rather than inferred from the entries, so this release owes noCompatibilitysection.The candidate's heading is renamed rather than a second section added.
v4.8.0rc1's own notes were generated from the tree at its tag, so nothing is lost by the heading moving on, and the changelog keeps one entry per version a user can install. Its[v4.8.0rc1]link definition goes with it rather than being left dangling.tests/testPackageMetadata.py(#346) carries no entry: test-only, nothing undersrc/smda/orpyproject.tomlmoved.What the rehearsal established
v4.8.0rc1ran the whole path on 2026-09-14. Both halves of the gate behaved:pypienvironment and was rejected. Verify and build both passed, publish failed on the rejection, and the release job skipped — no GitHub release, no milestone touched, nothing on PyPI. A rejected deployment leaves no partial state.latest: 1.1.1, which is the pre-release marking working end to end rather than only in the guard's output.python -P: 33 functions oncutwail_xored,smda_version: 4.8.0rc1, and re-importing that report costs 0 recalculations with hashes intact. That is fix(common): order a pre-release version when importing a report #345 confirmed on an artifact a user could install, not only in the suite.The milestone gate is the one thing the rehearsal did not exercise — no milestone was named
v4.8.0rc1, so it reported a notice and continued. Av4.8.0milestone now exists reading 0 open / 4 closed, so this release runs that gate for real.Verification
On this branch, saturn:
python -m pytest tests/→ 2,172 passed, 1 skipped, 2,623 subtestsruff check .andruff format --check .→ cleanrelease_guard.py --tag v4.8.0→ passes,version=4.8.0 prerelease=false[Unreleased]spansv4.8.0...HEAD,[v4.8.0]spansv4.7.0...v4.8.0After this merges and
masteris green, the tag goes on and the deployment gets approved rather than rejected. That publish is the first one this repository has made without a person holding an API token.🤖 Generated with Claude Code