8000 Merge remote-tracking branch 'origin/main' into agent/3421-an-interrupted-in-box-task-run-prints-no-socat-o · gominimal/minimal@970f7bb · GitHub
[go: up one dir, main page]

Skip to content

Merge remote-tracking branch 'origin/main' into agent/3421-an-interru…

Minimal Arch / arch-review failed Nov 14, 2026 in 0s

CONDITIONS

Baseline. This re-reviews 970f7bb5 against the CONDITIONS on b8cc8ab2 (this PR), on arch 15b603e6. The checkout has no arch-map.md, so I found the sections by search, as the earlier rulings did. The ledger entry for b8cc8ab2 is cut off after its first lines, so I could not read its conditions word for word. since.diff holds two merges of main and one PR commit, f11f8340. git diff a42a9d6d 970f7bb5 touches only env.rs and env_min_helper.sh, so the merges add no PR change. f11f8340 keeps socat's stderr for non-run methods and adds a terminal-stdin leg to the test.

Governs. Design principle 4 (line 334): a payload command writes only the payload to stdout, and diagnostics go to stderr. Design principle 5 (line 335): min's own failures use reserved codes, and a stop by signal borrows the shell's 128 + signal. The exit-code table: 124 is a box ended by its own timeout (line 370), and 125 is a min-side runtime failure (line 371).

Ruling. The stderr part conforms. Only run sends socat's stderr to /dev/null, and the broken-stream line stays for a stream that ends with no exit: line and no signal. The exit code does not conform. The helper decides that a signal end "is not a failure of min" and then returns 125, which line 371 reserves for a min-side failure. C1 returns sock_status (128 + signal) for that case. C2 changes the test that pins 125. Neither condition changes the diagnostic text. The task's proof says "exits with the same code as today". That plan line contradicts lines 335 and 371, and the design governs (rubric rule 9). The plan line has no requirement id, so the condition goes to the code and not to the spec. BCLI-014 is not addressed: the daemon relays a run's own timeout as an exit: line, and this change does not touch that path. The helper cannot tell a timeout from another signal, so its comments must not claim one.

Not checked. I did not build or run the tests. I did not confirm how bash in the helper itself reacts to the SIGINT of a Ctrl-C. A non-interactive bash in read can die on the signal before it reaches the code this change adds. In that case the shell gives 130, and C1 matches it. The help_word_in_argument_slot_reaches_rpc failure in the PR body's verify run is a test fact that a build settles. minimal-review owns it.

Conditions

  • C1: crates/minimald/src/env_min_helper.sh __min_rpc: in the run branch after the stream ends with no exit: line, when sock_status is above 128, return $sock_status (128 + signal, for example 130 for SIGINT and 143 for SIGTERM) with no line, in place of return 125. Keep return 125 and the broken-stream line for every other stream that ends with no exit: line. In the comment on lines 283-284, remove "or the run's timeout" and state that the helper exits with the shell's 128 + signal code.
  • C2: crates/minimald/src/env.rs min_helper_run_ended_by_signal_prints_no_connection_line: the check closure asserts out.status.code() == Some(143) (the fake socat ends on SIGTERM) in place of Some(125), for both the forwarded-stdin and the terminal-stdin runs. Change the comment above the test so that it says 128 + signal, not "exits 125 as before", and does not name the run's timeout. Keep the last assertion: a stream with no signal and no exit: line exits 125 and prints the connection line.

Ruled at 970f7bb against gominimal/arch@15b603e6afb8 by claude-opus-5-5; arch_rulings 1293, job adf0f2ea-aa74-43bc-aff4-e952cdbbde2f.

0