feat(coordinator): one BigBean's approval is enough, and the results come below it #119

Merged
HumanBean merged 1 commit from HumanBean/Beanstalk:feat/one-approval into main 2026-11-20 18:16:41 +00:00
Owner

Summary

Removes the rule that a diff touching nx2/, CMakeLists.txt, *.cmake, tools.lock, .farm/, .forgejo/ or Longshot's dev scripts needs two BigBeans, and that a manual run of such a PR head waits for a second one (seen on Longshot #1880: "This diff touches sensitive paths, so a second member of BigBeans must approve it too").

The rule protected nothing the sandbox doesn't:

  • every command from Longshot (the build scripts, the translator, the game) runs in the tester's sandbox (Workspace: only git and the tool download run outside, and they run nothing from the repository);
  • the tools come from main's tools.lock only, never the PR's;
  • a tester without a working sandbox gets no job (scheduler.NO_SANDBOX);
  • an ordinary .cpp change already runs as much of the PR's code on testers' PCs.

Changes:

  • trust.tier_for(approvers): one approver is reviewed, two double-reviewed; is_sensitive / sensitive_paths are gone.
  • The PR comment, beanstalk/approval, the state label and the dashboard no longer say "needs a second approval"; the dashboard's approve-only outcome goes.
  • Manual runs: no awaiting_approval state, no POST /api/v1/runs/{id}/approve, no approval field. No run in production ever waited for one, and Tooling's farm_run.py never called it.
  • regression.py's "not on main: no run" went too: it was this rule's needs_approval (it fired only off main and on sensitive files).
  • commits.sensitive and runs.needs_approval stay in the schema, unused, so an older image can still run on the database.
  • Docs: using.md, protocol.md, architecture.md, deploy.md, test-plan.md.

Also, the results come below the approval (Longshot #1860 had them edited into a comment from 08:18, far up the thread): a new approval deletes a sticky comment that shows no run yet (reporter.sticky_to_bottom, from commands.start), so update_pr posts the results in a new comment under it. One that can't be deleted is edited to say the results are below. A comment that shows a run is left to _sticky, which already folds it and posts a new one per run.

Testing

Tested: the coordinator suite in a python:3.13-alpine container with Postgres 17 (on the farm's server, not a tester): 587 passed, 3 failed, 1 skipped (with test_the_results_come_below_the_approval). The 3 failures (test_regression.py: test_tools_lock_starts_the_chain_again, test_superseded_and_announced, test_tick_waits_its_hours) fail the same way on main @ d7101761. Main also fails test_target_cell_names_what_each_report_ran_on (#117 removed reporter._hardware), which this branch fixes.

Still to test: after deploy, @Beanstalk on Longshot #1880 starts it with one approval.

🤖 Generated with Claude Code

## Summary Removes the rule that a diff touching `nx2/`, `CMakeLists.txt`, `*.cmake`, `tools.lock`, `.farm/`, `.forgejo/` or Longshot's dev scripts needs two BigBeans, and that a manual run of such a PR head waits for a second one (seen on Longshot #1880: "This diff touches sensitive paths, so a second member of `BigBeans` must approve it too"). The rule protected nothing the sandbox doesn't: - every command from Longshot (the build scripts, the translator, the game) runs in the tester's sandbox (`Workspace`: only git and the tool download run outside, and they run nothing from the repository); - the tools come from main's `tools.lock` only, never the PR's; - a tester without a working sandbox gets no job (`scheduler.NO_SANDBOX`); - an ordinary `.cpp` change already runs as much of the PR's code on testers' PCs. Changes: - `trust.tier_for(approvers)`: one approver is `reviewed`, two `double-reviewed`; `is_sensitive` / `sensitive_paths` are gone. - The PR comment, `beanstalk/approval`, the state label and the dashboard no longer say "needs a second approval"; the dashboard's approve-only outcome goes. - Manual runs: no `awaiting_approval` state, no `POST /api/v1/runs/{id}/approve`, no `approval` field. No run in production ever waited for one, and Tooling's `farm_run.py` never called it. - `regression.py`'s "not on main: no run" went too: it was this rule's `needs_approval` (it fired only off main *and* on sensitive files). - `commits.sensitive` and `runs.needs_approval` stay in the schema, unused, so an older image can still run on the database. - Docs: using.md, protocol.md, architecture.md, deploy.md, test-plan.md. Also, **the results come below the approval** (Longshot #1860 had them edited into a comment from 08:18, far up the thread): a new approval deletes a sticky comment that shows no run yet (`reporter.sticky_to_bottom`, from `commands.start`), so `update_pr` posts the results in a new comment under it. One that can't be deleted is edited to say the results are below. A comment that shows a run is left to `_sticky`, which already folds it and posts a new one per run. ## Testing **Tested:** the coordinator suite in a python:3.13-alpine container with Postgres 17 (on the farm's server, not a tester): 587 passed, 3 failed, 1 skipped (with `test_the_results_come_below_the_approval`). The 3 failures (`test_regression.py`: `test_tools_lock_starts_the_chain_again`, `test_superseded_and_announced`, `test_tick_waits_its_hours`) fail the same way on main @ d7101761. Main also fails `test_target_cell_names_what_each_report_ran_on` (#117 removed `reporter._hardware`), which this branch fixes. **Still to test:** after deploy, `@Beanstalk` on Longshot #1880 starts it with one approval. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
feat(coordinator): one BigBean's approval is enough, whatever the diff touches
Some checks failed
CI / Conventional Commit title (pull_request) Successful in 17s
CI / Workflow lint (actionlint) (pull_request) Successful in 17s
CI / Coordinator tests (pull_request) Has been cancelled
CI / Agent build and checks (pull_request) Has been cancelled
CI / Coordinator image (pull_request) Has been cancelled
6ca8aa9033
The rule that a diff touching nx2/, CMakeLists.txt, *.cmake, tools.lock,
.farm/, .forgejo/ or Longshot's dev scripts needs two BigBeans (and that a
manual run of such a PR head waits for a second one) is gone. It protected
nothing the sandbox doesn't: every command from Longshot (the build scripts,
the translator, the game) runs in a tester's sandbox, the tools come only from
main's tools.lock, and a tester without a sandbox gets no job. Any .cpp change
runs just as much of the PR's code on testers' PCs.

- trust.tier_for takes only the approvers; is_sensitive/sensitive_paths go.
- The PR comment, beanstalk/approval, the state label and the dashboard no
  longer say "needs a second approval"; the dashboard's approve-only outcome
  goes.
- Manual runs: no awaiting_approval state, no POST /api/v1/runs/{id}/approve
  (no run ever waited for one), no approval field; jobs are reviewed with the
  requester as approver.
- The commits.sensitive and runs.needs_approval columns stay (unused) so an
  older image can still run on the database.
- regression.py's "not on main: no run" was this rule's needs_approval in
  disguise (it fired only off main AND on sensitive files): it goes with it.
- test_reports: #117 removed reporter._hardware; its test now checks
  _hardware_line, as main's suite failed there.

Assisted-by: Claude Code
Co-Authored-By: Claude Opus 5.5 <[email protected]>
HumanBean changed title from WIP: feat(coordinator): one BigBean's approval is enough, whatever the diff touches to WIP: feat(coordinator): one BigBean's approval is enough, and the results come below it 2026-11-20 18:11:31 +00:00
HumanBean force-pushed feat/one-approval from 6ca8aa9033
Some checks failed
CI / Conventional Commit title (pull_request) Successful in 17s
CI / Workflow lint (actionlint) (pull_request) Successful in 17s
CI / Coordinator tests (pull_request) Has been cancelled
CI / Agent build and checks (pull_request) Has been cancelled
CI / Coordinator image (pull_request) Has been cancelled
to 76d2eefa9b
Some checks failed
CI / Conventional Commit title (pull_request) Successful in 17s
CI / Workflow lint (actionlint) (pull_request) Successful in 18s
CI / Coordinator tests (pull_request) Has been cancelled
CI / Agent build and checks (pull_request) Has been cancelled
CI / Coordinator image (pull_request) Has been cancelled
2026-11-20 18:11:32 +00:00
Compare
HumanBean changed title from WIP: feat(coordinator): one BigBean's approval is enough, and the results come below it to feat(coordinator): one BigBean's approval is enough, and the results come below it 2026-11-20 18:11:55 +00:00
HumanBean force-pushed feat/one-approval from 76d2eefa9b
Some checks failed
CI / Conventional Commit title (pull_request) Successful in 17s
CI / Workflow lint (actionlint) (pull_request) Successful in 18s
CI / Coordinator tests (pull_request) Has been cancelled
CI / Agent build and checks (pull_request) Has been cancelled
CI / Coordinator image (pull_request) Has been cancelled
to f9689b38cc
Some checks failed
CI / Conventional Commit title (pull_request) Successful in 17s
CI / Workflow lint (actionlint) (pull_request) Successful in 18s
CI / Coordinator tests (pull_request) Failing after 1m24s
CI / Agent build and checks (pull_request) Successful in 43s
CI / Coordinator image (pull_request) Successful in 10s
2026-11-20 18:16:34 +00:00
Compare
HumanBean deleted branch feat/one-approval 2026-11-20 18:16:41 +00:00
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
Notifications
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
MagicBeanDev/Beanstalk!119
No description provided.