feat(checks): warn when --allow-existing-red runs against a stale local base #70

Merged

triform-admin merged 3 commits from agent/checks-stale-base-warn into main 2026-06-24 00:24:09 +00:00

Conversation 0 Commits 3 Files changed 2 +463

triform-admin commented 2026-06-24 00:23:29 +00:00

Owner

Copy link

Defect-2a stale-base detection in --allow-existing-red classification.

cargoless checks run --base <ref> --allow-existing-red now warns when the
local tracking ref is behind real origin before re-running checks against
the base worktree. Without the warning, a stale origin/dev silently resolves
to a pre-PYAML merge-base and classify_required_reds_at_base reports phantom
counts for whole-tree fingerprint checks — the wedge PYAML hit on donut
(~5 false-block dev-merge attempts).

What's in this PR

• base_freshness_warning() in BaseWorktree::create — uses read-only
  git ls-remote origin <short> to compare local tracking-ref tip vs real
  origin. Warns when they differ; classification proceeds regardless.
• Threshold gating: with rev-list --count succeeding (objects in odb),
  warn only if drift ≥ CARGOLESS_BASE_STALE_THRESHOLD (default 5).
  When count fails (the common stale-clone case — objects not in odb), warn
  regardless, because the user is reasoning against a base they cannot even
  fully describe locally.
• 5 unit tests under stale_base_tests covering: happy path silent, the
  exact production wedge (objects unavailable → "unknown number of commits"),
  count-known path, threshold honoured, unresolvable base silent.
• Disable with CARGOLESS_BASE_STALE_CHECK=0.
• Detect-only: no auto-fetch, no side effects on local refs.

CI

All 4 jobs green on the final commit 1051126: build / test / fmt / clippy.
ra-bench correctly skipped (S1 harness).

Commit history (will squash)

• 8b48840 feat(checks): warn when classify-at-base runs against a stale local tracking ref
• cb63c72 fix(checks): clippy clean stale-base helper
• 1051126 fix(checks): rustfmt edition-2024 single-line if-else

Defect-2a stale-base detection in `--allow-existing-red` classification. `cargoless checks run --base <ref> --allow-existing-red` now warns when the local tracking ref is behind real origin before re-running checks against the base worktree. Without the warning, a stale `origin/dev` silently resolves to a pre-PYAML merge-base and `classify_required_reds_at_base` reports phantom counts for whole-tree fingerprint checks — the wedge PYAML hit on donut (~5 false-block dev-merge attempts). ## What's in this PR - `base_freshness_warning()` in `BaseWorktree::create` — uses read-only `git ls-remote origin <short>` to compare local tracking-ref tip vs real origin. Warns when they differ; classification proceeds regardless. - Threshold gating: with `rev-list --count` succeeding (objects in odb), warn only if drift ≥ `CARGOLESS_BASE_STALE_THRESHOLD` (default 5). When count fails (the common stale-clone case — objects not in odb), warn regardless, because the user is reasoning against a base they cannot even fully describe locally. - 5 unit tests under `stale_base_tests` covering: happy path silent, the exact production wedge (objects unavailable → "unknown number of commits"), count-known path, threshold honoured, unresolvable base silent. - Disable with `CARGOLESS_BASE_STALE_CHECK=0`. - Detect-only: no auto-fetch, no side effects on local refs. ## CI All 4 jobs green on the final commit `1051126`: build / test / fmt / clippy. ra-bench correctly skipped (S1 harness). ## Commit history (will squash) - `8b48840` feat(checks): warn when classify-at-base runs against a stale local tracking ref - `cb63c72` fix(checks): clippy clean stale-base helper - `1051126` fix(checks): rustfmt edition-2024 single-line if-else

triform-admin added 3 commits 2026-06-24 00:23:29 +00:00

feat(checks): warn when classify-at-base runs against a stale local tracking ref ... 8b48840ddc

`cargoless checks run --base <ref> --allow-existing-red` re-runs project
checks against a base worktree to classify required reds as inherited
vs. new. When the local `origin/<ref>` lags real origin, the base
worktree resolves to a pre-merge-base tree and whole-tree fingerprint
checks report phantom diagnostic counts — classifications come back as
"new" because their fingerprints don't match the stale-base run.

This is the wedge the PYAML team hit on donut (~5 false-block dev-merge
attempts, reported as Defect 2 in 2026-06-23 ticket).

`BaseWorktree::create` now calls a read-only `git ls-remote origin
refs/heads/<ref>` before `git worktree add` and compares the remote tip
to the local tracking ref. When tips differ:

  - if `rev-list --count` succeeds (remote objects are in odb), warn
    when drift ≥ CARGOLESS_BASE_STALE_THRESHOLD (default 5);
  - if `rev-list --count` fails (the common stale-clone case), warn
    regardless — the operator is reasoning against a base they cannot
    even fully describe locally.

Disable with CARGOLESS_BASE_STALE_CHECK=0. Detect-only: the warning
suggests `git fetch origin <ref>`; classification continues regardless,
so this never blocks a build. `ls-remote` never advances local refs.

Unit tests use a file:// origin fixture to exercise: silent-when-fresh,
warn-when-objects-missing (the wedge case), warn-with-count-when-objects-
available, threshold-honoured-when-count-known, silent-for-unresolvable-base.

Belongs alongside the existing A1–A4 asks in
tf-multiverse/.docs-discovery/plan/cargoless-source-asks.md as A5.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

fix(checks): clippy clean stale-base helper ... cb63c72562

Three default-warn clippy lints flagged on 8b48840:
- collapsible_if: collapse `if let Some(n) = behind { if n < threshold {...} }`
  to `if behind.is_some_and(|n| n < threshold)` in base_freshness_warning.
- redundant_closure: replace `.map(|s| s.to_string())` with
  `.map(str::to_string)` in resolve_tracking_ref and ls_remote_origin_branch
  (one each). Same behaviour, no allocation difference.
- needless_borrow: drop the extra `&` on `local.parent().unwrap()` in the
  test fixture — the inner `Option<&Path>::unwrap()` is already `&Path`.

The format! call in base_freshness_warning loses redundant `short = short`
style named args (the args are now bound to local lets which the captured
identifier syntax picks up directly). Pure mechanical formatting change.

No behaviour change; this is the clippy gate on the Defect-2a stale-base
detection PR.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

fix(checks): rustfmt edition-2024 single-line if-else ... 1051126576

`cargo fmt --check` (workspace edition 2024) collapses
`if x.is_empty() { None } else { Some(name) }` onto one line. My earlier
`rustfmt <file>` (no flag) defaulted to edition 2015 which expands them
to 5 lines — fmt CI on cb63c72 caught the drift. Re-formatted with
`rustfmt --edition 2024` so the source matches what CI checks.

No behaviour change. Clippy/build/test all stayed green on cb63c72;
only fmt was red.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

triform-admin merged commit d7fe357e5d into main 2026-06-24 00:24:09 +00:00

triform-admin referenced this pull request from a commit 2026-06-24 00:24:10 +00:00

feat(checks): warn when --allow-existing-red runs against a stale local base (#70)

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

triform/cargoless!70

No description provided.