Repository navigation
CONDITIONS
Inputs. Re-review of #3603 at 6dcb1875, on arch 8b824707, the same arch as every earlier ruling on this PR. The baseline is the CONDITIONS ruling at ea331ed5. since-conditions.json carries its C1, C2 and C3, and since.diff starts at ea331ed5. No findings.json or arch-map.md is staged, so I opened design §7.1 (LocalVM, specs/networking/deployment-and-egress-gateway.md lines 243–255) directly.
Compare. since.diff holds the merge 98ce5434/6dcb1875 of main (BEP, box CLI, telemetry, installer, spec 02/25/28 edits; none touches declared ingress) and one PR commit, 271460bf. That commit only rewrites the test declared_ingress_bound_at_activation in crates/minimald/src/net/policy.rs. It adds the private seam bind_declared_on under bind_declared_at_publish so the test drives the activation entry, then an attach, and asserts two identical exposes with no rollback. The seam is a pure refactor: bind_declared_at_publish still passes control.host_unbinds_declared_forwards(). The test now proves the rule it cites better than before (rule 3). Spec 18 and the PR body are unchanged since ea331ed5.
Governs: §7.1 line 255, the bind discipline: bind before the name is published, a failed bind is a surfaced error, sockets held for the session's lifetime, and ingress revocation unbinds. §7.1 line 245: the helper holds a declared port's forward it bound at publish regardless of guest requests.
Conditions from ea331ed5:
- C1 — open. The
## Proofsection still says box stop does not free the declared ports at once, but gives noResidual:line with an issue for that undone task leg. The only residual is #3646 (native half). - C2 — open. The NET-121 sub-requirement at spec 18 lines 1044–1047 is unchanged.
finalizeincrates/minimald/src/box_actor/setup.rsstill has the(true, None) => {}arm at line 902, where a VM-backed box with no switch address binds nothing at activation. The comment does not name that case or give it an issue. - C3 — open. The
## Teststable still records exit 4 for the verify line, andagent-verifytimed out. No passing run of the rewritten test is recorded.
Ruling. The change is in the right owner (net/policy.rs, daemon layer) and the new seam does not move a check across a boundary. The three conditions are unchanged and carried whole, with C3 restated for this head. This follows the two CONDITIONS rulings at ea331ed5. No departure from precedent.
Not checked: whether the rewritten test passes. That needs a run, which is C3.
Conditions
- C1: In the PR body
## Proof, add aResidual:line for the undone task leg "box stop on a VM host frees the ports at once". Give it the URL of a filed issue. The issue states that the host side (the host-side creator toward minvmd, never the in-VM daemon) must withdraw the box's row, or unbind its declared forwards, at box stop. It must not wait forDETACH_GRACEincrates/minvmd/src/box_registry.rs. A residual without an issue is refused (rule 6). - C2: In
docs/specs/18-spec-box-networking/18-spec-box-networking.md, edit the new NET-121 sub-requirement (line 1044) so that it states only what the code does. Its comment (line 1047) must name the case where the VM-backed box has no switch address. That is the(true, None)arm of thefinalizematch incrates/minimald/src/box_actor/setup.rs, where the first attach binds the forwards, so the name is published before the bind. Give that case an issue URL, or add it to #3646 and say so in the comment. - C3: Run the verify line
cargo nextest run -p minimald declared_ingress_bound_at_activationuntil it passes, and record the passing result in the PR's## Teststable. At head6dcb1875the table still shows exit 4 for that line, andagent-verifytimed out after 1800s. The commit271460bfrewrote the test, but no run of the rewritten test is recorded. A proof that did not run is open (rule 4).
Ruled at 6dcb187 against gominimal/arch@8b824707cde6 by claude-opus-5-5; arch_rulings 1496, job 6ad682a8-7bfa-4f60-b05f-9c99e1dfc24e.