Repository navigation
PASS
Inputs. Re-review of #3593 at 5cd03066, on arch 8b824707, the same arch commit as the baseline PASS at 3cd97043. since-conditions.json is empty, and the latest ruling on this PR is a PASS. So no condition is open, and the compare is enough. I checked that since.diff equals git diff 3cd97043..5cd03066: one commit, crates/minimald/src/connection.rs only. No findings.json and no arch-map.md are staged, so I used the section the earlier rulings found by search.
Governs: specs/networking/deployment-and-egress-gateway.md §4.1 line 54. The classifier puts every host_ip box process in the cohort identity, and minimald must make the box's cgroup placement inescapable.
Ruling. The commit narrows log_in_box_dial_end(). Exit 2 is a "box stopped" warning only when the trimmed stderr equals INJECTION_BOX_STOPPED_TEXT. Exit 5 is an "unplaced" warning only when stderr starts with INJECTION_UNPLACED_TEXT. Any other failure with those codes is logged at info as the relay's own failure. This is the same test that min box exec applies in crates/minimal/src/cmd/boxes/exec.rs. watch_in_box_dial() passes stderr.trim_end(), so the equality against the trimmed constant holds. The log now records a placement failure only when the shim reported one, which makes the record more exact. The commit changes no placement, leaf join or refusal path. NET-079 keeps the status conforms from the baseline. The new test a_shim_exit_code_without_the_shim_s_line_is_the_relay_s_own_failure pins the negative case, and the existing tests now use the real shim line.
Not checked: I did not run the tests, because a build is out of bounds. The PR body says that just e2e, just test-cross and just test-root-integration did not run. This commit does not change that, and the baseline rulings accepted it for a log-only path.
Ruled at 5cd0306 against gominimal/arch@8b824707cde6 by claude-opus-5-5; arch_rulings 1528, job a5409a1f-12bf-4abc-aa53-d3adf78cc745.