Repository navigation
fix(minimal): on a VM-backed host, min ls prints a hostname listener failure on its HOSTNAME PROXY - #3372
Conversation
On a VM host, min ls printed a held hostname-proxy port on the BOX NAMES row and no HOSTNAME PROXY row. NET-020 puts the cause on the HOSTNAME PROXY row. The surface now folds the cause out, so the cause row always prints and BOX NAMES does not repeat it (#3046).
Drop the #3046 exception. The cause row is HOSTNAME PROXY on every lane, it prints once, and BOX NAMES does not repeat the held port.
ls_across_vms_names_one_proxy_verdict_per_vm pinned the old fold, where the cause moved into BOX NAMES. NET-020 keeps it on HOSTNAME PROXY.
ls_keeps_the_answerer_row_beside_a_down_proxy pinned the old fold. The cause stays on HOSTNAME PROXY beside any verdict, as NET-020 says.
|
Important Review skippedBot user detected. To trigger a single review, invoke the
|
|
/revise make Evidence: the PR body says Do:
|
|
Round 1 of 10 on a /revise note: the coder changed nothing. Revise round: I could not reproduce the agent-verify exit 100. I ran the agent-verify steps for this PR's changed crate (minimal): fmt check, clippy -p minimal -D warnings, and nextest -p minimal. They all passed at the current head 262155e9, three times. The full nextest run also passed at the old head 811bd1b, where the sandbox reported exit 100. So no test fails here at either head. The sandbox failure is probably a flaky test or a sandbox-specific condition. Its log does not name the test, so I cannot name it. I changed no file, as the directive asks for a failure that this change does not cause. One side effect of my investigation: a worktree build that shared the target dir left stale boxes artifacts. I touched the .rs files (mtime only, no content change) to force a rebuild. git status is clean. |
|
Approved: every review thread resolved at @norrietaylor resolved the last thread, which releases the approval this review had withheld. Nothing to do. |
There was a problem hiding this comment.
arch-review: PASS · the full ruling
The change conforms to the design. Section 7.1 line 255 requires that a failed bind is a surfaced error. The design does not say which min ls row shows it. NET-020 puts the cause on the HOSTNAME PROXY row, and both listing paths now print it there once. One wording question is outside the design: for an unconfirmed publish, BOX NAMES says "not serving" but HOSTNAME PROXY says "not confirmed yet". Make the two rows agree in this PR or in a follow-up.
Ruled at 811bd1b against gominimal/arch@15b603e6afb8 by claude-opus-5-5.
…cked-host-min-ls-prints-a-hostname-lis
A proxy whose publish is unconfirmed printed "not serving" on BOX NAMES, while HOSTNAME PROXY beside it said "not confirmed yet". The row now keeps the headline for the cause and leaves the detail to HOSTNAME PROXY.
|
Round 2 of 10 on the runtime's review findings, pushed as
Verify Outside the task's
Converted to a draft: something above needs a person before review. Review threads the coder judged:
On a VM host with the proxy port held, min ls prints the holder and the remedy once, on the HOSTNAME PROXY row. The BOX NAMES row says only '.min.internal: the hostname proxy is not serving'. When the publish is unconfirmed, the BOX NAMES row says 'not confirmed yet', the same as the HOSTNAME PROXY row. Notes from the agent: Revise round 2. A new function, resolver::name_surface_line_beside, writes the BOX NAMES row beside a proxy-down cause. A folded Proxy verdict keeps the headline for its cause ('not serving' or 'not confirmed yet') without the detail. Both min ls paths (one VM and across VMs) call it. The unit test a_proxy_down_sibling_is_named_once_beside_any_verdict now pins the BOX NAMES text for each cause, PublishUnconfirmed included. I did not run 'bash scripts/session-e2e.sh proxy_recovers_when_its_port_frees'. This box has no built min or minimald and no session host (no /dev/kvm). This round does not change the script. Its BOX NAMES check only refuses the held port on that row, and the new text still meets it. |
There was a problem hiding this comment.
arch-review: PASS · the full ruling
The new commit keeps the cause headline on the BOX NAMES row. For an unconfirmed publish, the row now says "not confirmed yet" and agrees with the HOSTNAME PROXY row. The cause detail and the remedy still print once, on the HOSTNAME PROXY row, as NET-020 requires. The failure stays surfaced, as §7.1 line 255 requires. The merge from main has no conflict edits. The earlier PASS stands for this head.
Ruled at c9b5318 against gominimal/arch@15b603e6afb8 by claude-opus-5-5.
|
Round 3 of 10 on a failed check suite: the coder changed nothing. Revise round 3: no change on purpose. The failed macOS |
There was a problem hiding this comment.
Reviewed against D3046 at c9b53181. surface_beside_proxy_down folds the cause out of the surface verdict, and proxy_down_line_beside always prints it on the HOSTNAME PROXY row, so the cause is named once there, as NET-020 says. BOX NAMES keeps only the headline (name_surface_line_beside: 'not serving' or 'not confirmed yet'). Both listing paths and the e2e expectation follow. The tests/cli.rs change updates two tests to the new split and is accepted out of layer. Arch: PASS on head.
There was a problem hiding this comment.
minimal-review — approved
Verdict APPROVE — every review thread is resolved; @norrietaylor closed the last one. The review of c9b53181ee6e withheld this approval on the threads open then. Posted as a comment: the App cannot approve its own pull request; the check and the label carry it.
Defect
Fixes #3358
crates/minimal/src/resolver.rscrates/minimal/src/cmd/list.rscrates/minimal/src/tests.rsscripts/session-e2e.shTask marker
Proof
On a VM host with the hostname proxy port held, min ls prints the holder and the remedy once, on a HOSTNAME PROXY row, for one VM and across VMs. The BOX NAMES row says only '.min.internal: the hostname proxy is not serving' and does not repeat the cause. The unit tests show this rendering. No macOS VM host was available, so nobody saw it on a real VM.
Notes from the agent: Out of scope: crates/minimal/tests/cli.rs. Two tests there (ls_across_vms_names_one_proxy_verdict_per_vm, ls_keeps_the_answerer_row_beside_a_down_proxy) expected the old fold, with no HOSTNAME PROXY row beside a Proxy verdict. I changed only that expectation, to one cause row. Without that change, the minimal suite fails. After the change, the full minimal suite passes (1103 passed) and clippy is clean.
Design: surface_beside_proxy_down now folds a Proxy or ProxyNotServing verdict beside a cause into plain LiveSurface::Proxy. The BOX NAMES row then uses the existing words '.min.internal: the hostname proxy is not serving'. proxy_down_line_beside now always returns the cause line. It keeps its signature and has a debug_assert that the caller folded the surface first.
Printed text: for a PublishUnconfirmed cause, the BOX NAMES row now says 'not serving'. The HOSTNAME PROXY row says 'not confirmed yet'. The code cannot prove 'not serving' in that case. A reviewer can decide whether the folded row needs its own wording for that cause.
e2e: the script counts the cause by the holder phrase 'holds 127.0.0.1:.' and no longer counts 'the hostname proxy is not serving'. That phrase now prints on both rows. The script also fails if BOX NAMES names the held port. The minvmd branch of the script was not run, because this box has no VM lane.
Tests
cargo nextest run -p minimal a_proxy_down_sibling_is_named_once_beside_any_verdictcargo nextest run -p minimal proxy_held_line_names_holder_without_restartcargo nextest run -p minimal ls_warning_clears_on_recovery_with_no_guest_portbash scripts/session-e2e.sh proxy_recovers_when_its_port_freesVerify
min task run agent-verifyexited 100, so this PR is a draft.Out of scope
These files are outside the task's
layers:, so this PR is a draft:crates/minimal/tests/cli.rs