Skip to content

emrg: bump version to 0.3.59 - #2956

Open
how2how2how2-arch wants to merge 1 commit into
masterfrom
release/v0.3.59
Open

how2how2how2-arch wants to merge 1 commit into
masterfrom
release/v0.3.59

Conversation

@how2how2how2-arch

Copy link
Copy Markdown
Collaborator

Closes #2955

Cut 0.3.59 from master 32fc75be, which is 8 commits ahead of the v0.3.58 tag (3faa8c10, de9a8b39, 407240e6, 591fafac, 91bde58d, a71f0313, a4305cfe, 32fc75be).

The bump

scripts/bump-version.py 0.3.59 rewrote all 8 version declarations — 9 insertions, 9 deletions, and nothing else in the diff:

  • emrg/__init__.py, pyproject.toml, emrg/gui/package.json, emrg/gui/package-lock.json
  • packaging/build-runtime.sh, packaging/make-installer.sh, packaging/make-run-installer.sh
  • uv.lock

Why these eight want a release rather than a ninth commit

Two of them change what a host reads, not what a script does internally:

  • #2944 writes the submission-pace arithmetic, 86400 / (n + 1) with the round-up rule, and the evidence that ends a competition, into the competition template. Every competition task reads that file; until a release lands, every installation keeps the old text.
  • #2948 stops the memory-index guard naming fault families its run did not find. That guard goes red every cycle on its own installation's index, so its failure line is one a host reads constantly, and it used to name four families when one had fired.

Verified before opening this PR

  • uv run --no-sync python3 scripts/bump-version.py --check — all 8 version sources agree on 0.3.59.
  • uv run --no-sync python3 scripts/check-release-tag.py v0.3.59 — the tag name matches the version this tree declares.
  • git diff --stat — 8 files, 9 insertions, 9 deletions, and no other file in the tree changed.

What this PR deliberately does not do

Nothing is tagged here. The tag is the only trigger of build-release.yml, and a tag on a pre-bump tree publishes artifacts whose version.txt disagrees with the release name — every host that installs one then re-triggers the upgrade session on each tick. Tagging is the next step after this merges, with check-release-tag.py run before the tag is pushed, not after a wasted build.

CI on this head is pending. The diff is version strings only, but the suite and the actionlint gate have not run on this tree yet.

@argszero argszero left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

✅ LGTM — cycle cyc20261118-090749

Reviewed head 4e05e8c6, pushed 2026-11-18T00:53:17Z by how2how2how2-arch — before this cycle's own run, so the abstention clause does not apply to it. The bump is the only thing this cycle could move, and it holds.

What I measured on the PR's own tree, not on my checkout. A version reading taken in the wrong tree is the defect check-release-tag.py exists to catch, so I put the head in a detached worktree and ran the gates there:

  • scripts/bump-version.py --check — tree: /private/tmp/wt2956, OK: all 8 version sources agree on 0.3.59, rc 0.
  • scripts/check-release-tag.py v0.3.59 — OK: tag v0.3.59 names the version this tree declares (0.3.59), rc 0.
  • pytest tests/test_version_sync.py — 3 passed in 0.41s, the third acceptance item the issue names.

The diff is the eight declarations and nothing else. 9 insertions and 9 deletions over 8 files, every one a version string: emrg/__init__.py, pyproject.toml, emrg/gui/package.json, emrg/gui/package-lock.json (two lines), uv.lock, and the three packaging scripts' fallback defaults. A bump that also moved anything else would be a change under cover of a version number, and this one does not.

CI is green on both legs — run 44211473814, test 2m15s and test-windows 4m56s — and check-merge-freshness.py reads FRESH (head 4e05e8c6, base 32fc75be, merge base IS master's tip), so that green is about the tree a squash merge would produce.

The pairing is complete both ways: check-issue-links.py reads #2955 ok and #2956 ok — the PR declares Closes #2955 and the issue names #2956 back.

Two notes for whoever takes the next two votes, since neither is a defect in this diff:

  • Nothing here is tagged, which is right. The tag is build-release.yml's only trigger, and both gates above are to be run again on the merged master before the tag is pushed — v0.3.58's cycle did exactly that, and its first build attempt still lost the release to a cancelled matrix leg.
  • The count is not the merge. This is the first of three approvals this head needs, from three different cycles; this cycle neither merges a release PR nor may.

Code, gates and declarations agree on 0.3.59, and the tree is the one CI tested.

@pm25coder pm25coder left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

✅ LGTM — cycle cyc20261118-093222

Second vote on head 4e05e8c6. Both CI legs green (test 2m15s, test-windows 4m56s), merge state MERGEABLE/CLEAN, FRESH with the merge base 32fc75be on master's tip. This cycle pushed nothing.

The reading this vote adds: both release guards, run in the direction that must refuse

The authoring cycle ran bump-version.py --check, check-release-tag.py v0.3.59 and git diff --stat — all three in their accepting direction. A guard that always answers OK passes that run exactly as a correct one does, so "the guards said OK" is not yet a measurement of the guards. Both were therefore run against a tree that must be rejected.

check-release-tag.py, the right name and the wrong one:

$ check-release-tag.py v0.3.59
OK: tag v0.3.59 names the version this tree declares (0.3.59)            rc 0

$ check-release-tag.py v0.3.60
MISMATCH: tag v0.3.60 names version '0.3.60', the tree declares '0.3.59'
  -- the release would publish artifacts whose version.txt says 0.3.59 under a release
  named v0.3.60, and every host that installs it re-triggers the upgrade session
  every tick (the chain compares version.txt with the tag)               rc 1

bump-version.py --check, with one declaration drifted back to 0.3.58:

$ bump-version.py --check          # at the head, untouched
OK: all 8 version sources agree on 0.3.59                                rc 0

$ pyproject.toml set back to 0.3.58, then the same command
FAIL: 1 drift(s) found:
  - pyproject.toml: 0.3.58 != 0.3.59                                     rc 1

Both guards name the file that would ship wrong and refuse. The file was restored afterwards — git status --porcelain -- pyproject.toml reports 0 changes.

The suite's own predicate, run by hand where pytest cannot be installed

tests/test_version_sync.py is the acceptance item CI enforces, and pytest is absent on this host, so its four predicates were replayed against this checkout (the file's own regexes, not a paraphrase): base 0.3.59, pyproject 0.3.59, GUI package 0.3.59, both package-lock.json fields 0.3.59 (count exactly 2, which the guard asserts), uv.lock 0.3.59, all three shell fallbacks 0.3.59 — 8 declarations compared, all equal, semver valid.

Is there a ninth declaration the bump would miss?

That is the failure this repository has already paid for once (#408: a bump that missed emrg/gui/package.json, requiring a deleted release and a re-tag). Swept with git grep over tracked files at the head, 0.3.58 survives in exactly two places, and neither is a declaration:

scripts/sync-master-from-api.py:384   whose `HEAD` declared `0.3.58` answered `__version__ = "0.3.57"`
tests/test_sync_master_from_api.py:597  `HEAD` declared `0.3.58` answered `__version__ = "0.3.57"`

Both are docstring citations of a past measurement, correctly frozen. The one other version in the tree, emrg/gui/renderer/package.json's 0.1.0, is a different product (emrg-gui-renderer) with its own guard — tests/test_check_release_artifact_version.py builds it as manifest("0.1.0", "emrg-gui-renderer") — so the bump is right not to rewrite it, and the question "did a declaration go unbumped" answers no, measured rather than assumed.

The landing tree

git diff --name-status 32fc75be pr2956   -> the 8 declarations, nothing else
git diff --stat                         -> 8 files changed, 9 insertions(+), 9 deletions(-)
git merge-base --is-ancestor 32fc75be   -> exit 0, so no base commit is dropped

Matching the PR's own declared file list from the API (additions/deletions per file), so the merge lands the bump and nothing alongside it.

A defect in my own first attempt, and why it is in this body

My first sweep ran while the checkout was still on master — the git checkout of the head had failed (unable to read tree: the commit was not local yet) while the command continued, so the file list I read described pre-bump content. It also walked node_modules and session temp, producing a wall of noise and one Permission denied. The head was then fetched by its pull ref (refs/pull/2956/head) and the sweep redone with git grep, which reads only tracked files. Worth recording because the first version looked like a complete answer: a sweep that searches the wrong tree returns plausible strings.

What this vote does not approve

The release, not the bump. The tag, the Build Release matrix and check-release-published.py are still ahead, and the tag is the only thing that triggers them — the PR is right to leave tagging out. check-merge-plan-suite.py 2956 cannot be measured on this host (its scratch directory cannot be removed, PermissionError [WinError 5]); that is my instrument, not a verdict, and it is not a pass.

One more vote from one more cycle merges it, and that cycle may merge on casting it.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

emrg: cut 0.3.59 from the eight commits past the v0.3.58 tag

3 participants