Skip to content

fix(map): the Code map gate's fixtures and history, and the one-time pin bootstrap - #1063

Merged
cryptskii merged 2 commits into
mainfrom
fix/code-map-pins-bootstrap
Sep 29, 2026
Merged

cryptskii merged 2 commits into
mainfrom
fix/code-map-pins-bootstrap

Conversation

@cryptskii

Copy link
Copy Markdown
Collaborator

Main's Code map check has been red since #1057 merged. It was merged before its last two steps, and this PR is those two steps.

What was red on main (run 36556460780, at 8b28149)

  • The evidence pins refuse what they must. Two causes:
  • The intent manifest holds against the map. It reported "pins: not checked: specs/requirements/INTENT_PINS.tsv does not exist". The pins had not been bootstrapped yet.

The same run's Pin evidence job passed. With nothing pinned, it ran every requirement row's evidence tests on 8b28149, 289 distinct tests in 18 cargo runs, and every one passed. Its log and the run's code-map artifact are what the bootstrap below pins from.

Commits

  1. fix(ci): commit the board-log fixtures, and give the pins' tests history.
    • A .gitignore negation now tracks exactly ci/fixtures/board-logs/*.log.
    • The Code map job's checkout now uses fetch-depth: 0.
  2. feat(map): the one-time pin bootstrap.
    • ci/intent_pins.py pin-unpinned wrote 561 pins, one per requirement row with known holes included, all at 8b28149.
    • Before writing, it checked that:
      • this tree's compiled code is 8b28149's;
      • CI's map fingerprints as this tree;
      • the map indexed every build;
      • every evidence test resolves to one definition and passed in CI's log.

Verified against CI's map (android and node indexed)

  • The intent manifest reads 1,074 rows, 0 failing and 6 known holes.

  • The pins read 561/561 PINNED.

  • ci/test_intent_pins.py passes 14/14.

  • Live on the real pins, both edits restored afterwards:

    • a CONFORMANCE §8 status edit (MR-STOR-0146 Met → Partial) stales both its pins under status;
    • a manifest citation edit stales its pin under citation.

    The first is the markdown-only change the old path filter skipped.

On this PR's CI

  • Code map rebuilds the map on this commit. The pins file is not a map input, so its facts are unchanged and every pin should read PINNED.
  • Pin evidence sees all 561 pins as new against main and reruns their evidence on this commit: the full proof the bootstrap rests on, run once more.

From here, only pin KEY and repin KEY change a pin, one key at a time. Making Code map a required check waits for the owner's approval.

CI's pin tests failed on two assumptions the local run could not show.

The four board-log fixtures were never committed. `.gitignore`'s `*.log`
made `git add` of their directory skip them silently, so they existed only
on the machine that captured them. A negation, `!ci/fixtures/board-logs/*.log`,
now tracks exactly those four real `cargo test` runs.

The Code map job checked out one commit, while the pins' tests read history:
a commit's parent, and the repository's first commit, which holds no pins.
Its checkout now fetches full history, as the Pin evidence job's already
does.

ci/test_intent_pins.py: 14/14.
specs/requirements/INTENT_PINS.tsv pins every requirement row of the intent
manifest, known holes included. It was written by
`ci/intent_pins.py pin-unpinned` from CI's own run of main at 8b28149
(run 36556460780):
- the code-map artifact: the Linux map with the Android, storage-node and
  host test builds indexed;
- the Pin evidence job's log: with nothing pinned yet, it ran every
  requirement row's evidence tests on that commit, 289 distinct tests in 18
  cargo runs, and every one passed.

Before writing, the command checked that:
- this tree's compiled code is 8b28149's (`code_at`);
- CI's map fingerprints as this tree (`fresh`);
- the map indexed every build;
- every evidence test resolves to one definition and passed in that log.

Against that map (--built android,node):
- the intent manifest reads 1,074 rows, 0 failing, 6 known holes;
- the pins read 561/561 PINNED.

Checked live against the real pins: a CONFORMANCE §8 status edit
(MR-STOR-0146 Met -> Partial) stales both its pins under `status`, and a
manifest citation edit stales its pin under `citation`. That is exactly the
markdown-only change the old path filter skipped. Both edits were restored.

From here only `pin KEY` and `repin KEY` change a pin, one key at a time.
@cryptskii
cryptskii merged commit 11647a4 into main Sep 29, 2026
24 checks passed
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.

1 participant