fix: bind Refs parser to one token #252
No reviewers
glm-bot-andresmgsl
kimi-bot-andresmgsl
claude-bot-andresmgsl
andres
Labels
No labels
attention
blocked
blocker:ci-red
blocker:conflict
blocker:drill-pending
blocker:unrequested
bug
claimed
documentation
enhancement
epic
merge-next
needs-ruling
needs-triage
offsite
post-merge
ready
release
scope:docs
scope:guards
scope:labels
scope:release-flow
stale
state:addressing
state:bots-reviewing
state:building
state:needs-human
No milestone
No project
No assignees
4 participants
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference: heavy-duty/ceremony#252
Loading…
Reference in a new issue
No description provided.
Delete branch "build/234-bind-refs-token"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Closes #234
Rebinds each
Refskeyword to exactly the single reference token that immediately follows it, preventing same-line prose from creating false issue-flow edges.Worklog
refs_referencesto emit one bound token per keyword occurrenceCI fix round — run 1796 / job 2
origin/maindevelopment-version bump without rewriting pushed historychangelog-armedon the corrected treeAcceptance criteria
52and nothing else8,12, and175unmodifiedRefs #8, #9returns8alone175Refs #8. Refs #9.returns both local referencesrefs: nonereturn nothingtest/issueflow-reconcile.test.shpasses wholeTest evidence
b2048f6: the four new behavior checks fail for trailing prose, comma narrowing, cross-repo trailing prose, and repeated same-line declarations; the preserved forms pass.b105939: the original implementation made the full issue-flow reconciler suite pass.5232027:Refs #8Refs #9exposed a lost left boundary (8,9); fixed atd712f06to return8alone.d712f0636f4a27be8b052024ea90a74208d2b52c: issue-flow574 passed, 0 failed; changelog suites48 + 50 + 37 + 20passed; chunked ShellCheck clean; mawk 1.3.4; diff and worktree clean.Round log
Round at
4fb01e8bRound passed with no written reply.
Initial build round answered whole at head
d712f0636f.Refsoccurrence to its single immediately following token and retained the existing LOCAL/CROSS classification plus numeric sort.Refs #8Refs #9yields only8.changelog.d/234.md.574/574, changelog suites48/48 + 50/50 + 37/37 + 20/20, chunked ShellCheck, mawk 1.3.4, clean diff, and clean worktree.Every acceptance criterion and worklog item is checked in the PR body; no review point is outstanding.
📣 round answered at head
d712f0636fWIP: fix: bind Refs parser to one tokento fix: bind Refs parser to one token🔧 addressing round on head
d712f0636fClassification: ATTRIBUTABLE.
The actual
CI / self-guards (pull_request)log for run 1796, job 2 shows checkout and repository fetch both succeeded, then the repo-ownedchangelog-armedguard ran against this head and failed deterministically:these fragments were not consumed: changelog.d/238.md, changelog.d/234.md — re-run 'changelog-assemble 0.6.2'. The branch is based on the 0.6.2 release merge and does not yet carry currentmain'sca7ce6edevelopment-version bump, so this is branch state rather than an infrastructure/setup failure. I will not rerun the unchanged head.Round analysis:
origin/mainso the change is evaluated as 0.6.3-dev, while retaining the required #234 fragment; run the failing self-guard locally and the scoped regression suite; push the corrective commit(s).Round answered whole at head
4fb01e8b9f.CI / self-guards (pull_request)failure was ATTRIBUTABLE, not infrastructure: run 1796 job 2 checked out successfully and the repo-ownedchangelog-armedguard rejected the branch's released0.6.2version with surviving fragmentschangelog.d/238.mdandchangelog.d/234.md. The unchanged failing head was not rerun.origin/mainatca7ce6e9194d0b427c093cfc74f03eb2c84fe25fwithout rewriting pushed history. That carries the required0.6.3-devdevelopment-version bump while preserving the #234 fragment and parser changes.changelog-armed,changelog-monotonic,changelog-assembled,drill-recorded, andrunner-isolatedall pass;git diff --checkis clean; the worktree is clean and matches the pushed remote head.All CI-fix Worklog items are checked. I have not requested or re-requested any reviewer; the engine owns the panel request after this head settles.
📣 round answered at head
4fb01e8b9f🔎 reviewing head
4fb01e8b9f4fb01e8b9f🔎 reviewing head
4fb01e8b9fApprove — verified at head
4fb01e8b9fin a detached throwaway worktree.Acceptance criteria (issue #234) — all run, all pass:
test/issueflow-reconcile.test.shwhole: 574 passed, 0 failed on this box, whose awk is mawk 1.3.4 (the runner shape the mawk criterion names); the crew!64 fixture is in-file and returns52alone; the pre-existingrefs_bodyfixture still returns8\n12\n175and its check is untouched by the diff.Refs #8, #9→8; cross-repo forms (Refs heavy-duty/rig#4, with, #12) → empty; paren case returns175;Refs #8. Refs #9.→ both;Refs #8Refs #9→8;fix refs parsing from #200andCloses #40; refs: none→ nothing;open_pr_issuesunion checks unchanged and passing..github/scripts/shellcheck-all.sh): clean over all 64 tracked scripts. No\xescapes introduced.Failure cases actually fail: I re-ran the four must-change shapes through the pre-PR parser (origin/main's
refs_references, reproduced verbatim) — crew line yields the false65, comma yields8 9, cross-repo prose yields12, repeated declaration yields8only. The new tests genuinely discriminate; the builder's red-baseline claim atb2048f6is consistent with what I measured.Code: the rewrite binds each keyword occurrence to exactly the one token the spec's regex allows, loops over occurrences, deletes the
sub(/[.(;].*/)truncation and the guard it served, keeps theissue_references→ LOCAL →sort -nutail unchanged. The retained final byte (RSTART + RLENGTH - 1) is always a digit, so it can only suppress a false fresh boundary and cannot loop forever (every match consumes ≥ keyword+token). Output now comes fromtolower($0); I checked classification is case-blind where it matters (#N→ LOCAL, any qualified token → CROSS and dropped), so lowercasing changes nothing observable. Changelog fragment present.CI: head status is
successfor all seven checks, including theCI / self-guards (pull_request)that failed run 1796 job 2 on the prior head; changelog suites armed/monotonic/assembled/changelog pass locally (50/20/37/103).Not verified, relied on instead: the historical run-1796 log itself (I verified the corrected outcome, not the old failure — the builder's account plus the current green guard stand in); actionlint and the full
test/run.shmatrix withCEREMONY_REQUIRE_NPM/YQ=1(no npm/yq drive here — relied on the green CI jobs for the same head).On-panel (author codex-bot recuses; this verdict is a gating one, not advisory). Non-blocking nit, builder's discretion: none worth logging.
Reviewed against issue #234's acceptance criteria and the repo's conventions.
refs_referencesnow loops over everyRefsoccurrence and emits only the immediately following#Norowner/repo#Ntoken, matching the siblingrefs-not-closingparser.sub(/[.(;].*/, "", line)truncation is removed; the token boundary replaces it.issue_references, so cross-repo refs stayCROSSand the output stayssort -nulocal issues.test/issueflow-reconcile.test.shcover the crew!64 fixture, comma-list narrowing, cross-repo prose, parenthesized prose, multiple occurrences on one line, concatenated keyword misspellings, and non-Refs uses of the word.test/issueflow-reconcile.test.shpasses whole (574/574).shellcheck -xis clean on the changed script and test..github/scripts/{marker-check,vendored-check,self-ref-check}.share green on this head.test/run.shonly fails the package-json cases becausenode/npmis not installed on this box; every other test file passes, and the PR's CI/test status is SUCCESS at this head.actionlint.This head satisfies #234.
🔎 reviewing head
4fb01e8b9fca7ce6e) vs #234 acceptance criteria, repo constraints, code/testsApprove at head
4fb01e8b9fb8547831485112710686930d12c25c. Reviewed the whole PR (all six commits, baseca7ce6e) against #234's acceptance criteria, this repo's constraints, then the code.What I ran, in a detached worktree at this head
test/issueflow-reconcile.test.sh— 574 passed, 0 failed (mawk 1.3.4, bash 5.2.37).refs_referencesback toca7ce6e's body and left everything else at this head. The suite then reported 569 passed, 5 failed, and the five that fail are exactly the new behavior checks:ignores issue prose after the bound token,comma-list narrowing is deliberate,prose after a cross-repo Refs token stays non-local,every Refs occurrence on one line contributes its bound token,a concatenated Refs spelling is not a second keyword occurrence. The failure cases actually fail.52,65→52;Refs #8, #98,9→8;Refs heavy-duty/rig#4, #1212→∅;Refs #8. Refs #9.8→8,9. Preserved forms all unchanged: the:101fixture8,12,175;Refs #175 (split from #150)→175;Refs heavy-duty/rig#4→∅;Also refs: #8 and heavy-duty/rig#4.→8;fix refs parsing from #200→∅;Closes #40; refs: none→∅.Refsoccurrence on the same line — spec item 1, intended. Every reference it drops (585 lines) is either a cross-repo token in the keyword's bound position or prose after the bound token — the two narrowings #234 decides on purpose. A separate assertion over the same corpus: whenever the first keyword is immediately followed by a local#N, thatNis in the output. Zero losses of a bound reference..github/scripts/shellcheck-all.sh— clean, rc=0 (chunked helper, as the criterion requires).marker-check.sh,vendored-check.sh,self-ref-check.sh— all green.changelog-armeddriven the wayaction.ymldrives it (CHANGELOG=CHANGELOG.md VERSION_SOURCE=file FRAGMENTS_DIR=changelog.d):version '0.6.3-dev' agrees with fragment mode, rc=0. That is the CI-fix round proven on the tree, not just on the badge. Changelog suites: armed 50, assembled 37, assemble 48, monotonic 20, changelog 103 — all 0 failed.Acceptance criteria
Every box in #234 checks out: crew!64 fixture →
52alone;:101fixture byte-identical in the diff and still8\n12\n175; comma narrowing pinned by a test whose name says it is deliberate; both cross-repo forms and the paren form pinned;Refs #8. Refs #9.returns both; the two negatives return nothing;open_pr_issuesunion block at:104-110unchanged (context-only in the diff); suite passes whole; chunked shellcheck clean; awk is mawk-compatible with no\xescapes. Tasks likewise: thesub(/[.(;].*/, "", line)truncation and the^(#|…)#[0-9]+guard are both gone, and theissue_references→LOCAL→sort -nutail is untouched.changelog.d/234.mdis present under### Fixedand cites(#234).Closes #234is the right keyword — #234 marks no criterion post-merge.Code
The loop terminates: the match ends in
[0-9]+, so each pass advancesrestby at leastRLENGTH-1 ≥ 6bytes. The retained final byte is always a digit, which is why^cannot manufacture a keyword boundary out of a concatenated suffix —d712f06's fix, and the comment names why with its issue number, per CONTRIBUTING. The greedysub(/^.*refs[[:space:]:]+/, "", token)is safe rather than lucky: the reference-token charset[[:alnum:]_.-/]contains neither space nor colon, so a secondrefs[[:space:]:]+cannot occur inside a matched region — the fuzz found no misbinding.Nits — non-blocking, builder's discretion
rest = tolower($0)means the printed token is the lowercased one, where the old code matched on a lowered copy but sliced the original line. Today this is invisible:refs_referenceskeeps onlyLOCALrows and those are digits. It does mean a cross-repoOwner/Repo#4reachesissue_referencescase-folded, so if that side ever becomes load-bearing the case is already gone.actions/refs-not-closingaccepts[[]?#[0-9]+, the reconciler does not. Unchanged by this PR, and #234 putsrefs-not-closingout of scope explicitly, so this belongs in a discussion, not here.What I could not verify, and what I relied on instead
test/run.shis not whole-green in this box: 29 test files passed, 2 failed —test/labels-bootstrap.test.sh(7 passed, 7 failed) andtest/labels-dispatch.test.sh(7 passed, 14 failed), every failure ayq: command not found. I checked out baseca7ce6ein a second detached worktree and ran both files there: identical failure counts. Pre-existing environment gap on this box, not this PR.test/labels-scope.test.shlikewise emitsSKIP: yq not found. For the runner-side proof of those files I rely on CI run 1819 job 0.claimed→post-mergetransition and theOPEN_PR_ISSUESunion — are covered here only by the suite's unit and gather tests, which pass.glm-bot-andresmgsl referenced this pull request2026-08-24 19:18:27 +00:00