Fixes issue 543: TestAVerbsOwnProcessNamesItsConnection failed about once in 80 runs (-race -count=20) with the verb's connection was named [] / a name list missing the verb subprocess.
Cause.closedNames (cmd/mesh-controller/replay327_test.go) returned as soon as the bus's closed-connections list held anything. In TestAVerbsOwnProcessNamesItsConnection, the test's own setup connection (dialled to call EnsureControllerBuckets, then closed before the verb runs as its own process) can register as closed before the verb subprocess's connection does. closedNames could then return on that setup connection's close alone, before the verb subprocess's mesh-controller verb conditions connection ever showed up — a true race, not a logic bug in the assertion itself.
What changes
closedNames (replay327_test.go) now takes the name it is waiting for (want) and loops until a closed connection by that exact name appears, or its 5s deadline passes, returning whatever names it last saw either way. It no longer returns on the first non-empty list.
TestAVerbsOwnProcessNamesItsConnection (reconnects_test.go) passes "mesh-controller verb conditions" as the name to wait for; its own name check after the call is unchanged (now just confirming what closedNames already waited for).
closedNames has exactly one caller (TestAVerbsOwnProcessNamesItsConnection), so the signature change touches nothing else.
Tested
go build ./..., go vet ./cmd/mesh-controller/...: clean.
go test ./cmd/mesh-controller/... -run TestAVerbsOwnProcessNamesItsConnection -race -count=200 -v: 200 run, 200 pass, 0 fail, 0 data races (188s).
go test ./cmd/mesh-controller/... -race -count=1: all pass.
sh merge-check.sh: gofmt/vet clean, whole suite passes under -race (gcc present); store tests skipped (NOT STORE-CHECKED, no MESH_TEST_POSTGRES on this machine) — this change touches no store-tested code.
Test code only, one file's helper plus its one call site. No hand workaround: the helper now waits for the actual fact the test asserts (the verb's own connection appearing closed) instead of racing on any close.
Fixes issue 543.
Fixes issue 543: `TestAVerbsOwnProcessNamesItsConnection` failed about once in 80 runs (-race -count=20) with `the verb's connection was named []` / a name list missing the verb subprocess.
**Cause.** `closedNames` (cmd/mesh-controller/replay327_test.go) returned as soon as the bus's closed-connections list held anything. In `TestAVerbsOwnProcessNamesItsConnection`, the test's own setup connection (dialled to call `EnsureControllerBuckets`, then closed before the verb runs as its own process) can register as closed before the verb subprocess's connection does. `closedNames` could then return on that setup connection's close alone, before the verb subprocess's `mesh-controller verb conditions` connection ever showed up — a true race, not a logic bug in the assertion itself.
**What changes**
- `closedNames` (replay327_test.go) now takes the name it is waiting for (`want`) and loops until a closed connection by that exact name appears, or its 5s deadline passes, returning whatever names it last saw either way. It no longer returns on the first non-empty list.
- `TestAVerbsOwnProcessNamesItsConnection` (reconnects_test.go) passes `"mesh-controller verb conditions"` as the name to wait for; its own name check after the call is unchanged (now just confirming what `closedNames` already waited for).
`closedNames` has exactly one caller (`TestAVerbsOwnProcessNamesItsConnection`), so the signature change touches nothing else.
**Tested**
- `go build ./...`, `go vet ./cmd/mesh-controller/...`: clean.
- `go test ./cmd/mesh-controller/... -run TestAVerbsOwnProcessNamesItsConnection -race -count=200 -v`: 200 run, 200 pass, 0 fail, 0 data races (188s).
- `go test ./cmd/mesh-controller/... -race -count=1`: all pass.
- `sh merge-check.sh`: gofmt/vet clean, whole suite passes under `-race` (gcc present); store tests skipped (`NOT STORE-CHECKED`, no `MESH_TEST_POSTGRES` on this machine) — this change touches no store-tested code.
Test code only, one file's helper plus its one call site. No hand workaround: the helper now waits for the actual fact the test asserts (the verb's own connection appearing closed) instead of racing on any close.
Fixes issue 543.
closedNames returned as soon as the bus's closed-connections list held
anything, which could be the test's own setup connection closing before
the verb subprocess's connection did. That raced TestAVerbsOwnProcessNamesItsConnection
red about once in 80 runs. It now waits until the connection it is named
for appears, or the deadline passes.
Deliverynovox/mesh-controller@7630072e6309 — delivered since 2026-10-12T11:21:30Z
Delivery plan — builds nothing: the change touches no module of the mesh's graph
Transitions
2026-10-12 11:06 (new) → proposed (announced): novox/mesh-controller#252's head announced
2026-10-12 11:11 proposed → checked (checked): the verdict names this commit
2026-10-12 11:11 checked → ready (accepted): the gate passed or warned, and the repository's own check did not fail
2026-10-12 11:21 ready → published (merged): on the trunk its modules follow: merged there, and its walk opened — or nothing for a walk to move
2026-10-12 11:21 published → delivered (done): nothing for a walk to move
The commit's note under refs/notes/mesh-plan keeps every transition: git log --notes=mesh-plan.
<!-- mesh-delivery:view -->
**Delivery** `novox/mesh-controller@7630072e6309` — **delivered** since 2026-10-12T11:21:30Z
**Delivery plan** — builds nothing: the change touches no module of the mesh's graph
**Transitions**
- 2026-10-12 11:06 (new) → proposed (announced): novox/mesh-controller#252's head announced
- 2026-10-12 11:11 proposed → checked (checked): the verdict names this commit
- 2026-10-12 11:11 checked → ready (accepted): the gate passed or warned, and the repository's own check did not fail
- 2026-10-12 11:21 ready → published (merged): on the trunk its modules follow: merged there, and its walk opened — or nothing for a walk to move
- 2026-10-12 11:21 published → delivered (done): nothing for a walk to move
The commit's note under `refs/notes/mesh-plan` keeps every transition: `git log --notes=mesh-plan`.
Independent review, fresh clone (not ~/projects/novox/mesh-controller), checked out at the PR head 7630072e63.
Scope: test code only. The diff touches exactly cmd/mesh-controller/replay327_test.go (closedNames now takes want and loops until a closed connection named want appears, or its 5s deadline passes, returning whatever names it last saw either way) and cmd/mesh-controller/reconnects_test.go (TestAVerbsOwnProcessNamesItsConnection passes "mesh-controller verb conditions" as want). No production code changes. closedNames has exactly one caller, so the signature change is self-contained.
Verified the fix actually detects a wrong name, not just a slow one: locally broke connectionName in cmd/mesh-controller/main.go (prefixed the verb-name branch with WRONGNAME) and reran TestAVerbsOwnProcessNamesItsConnection. It failed cleanly in 7.65s with the verb's connection was named ["mesh-controller" "mesh-controller verb WRONGNAME conditions"] — a real failure, not a hang, bounded by closedNames's existing 5s deadline. Reverted the break afterward (working tree clean).
Ran the targeted test at the real head: go test ./cmd/mesh-controller/... -run TestAVerbsOwnProcessNamesItsConnection -race -count=100 -v — 100 of 100 passed, 0 data races, 93s total.
go build ./..., go vet ./cmd/mesh-controller/..., and gofmt -l on both changed files are all clean. mesh/merge-gate and mesh/repo-check on the PR's head are both green (pass: the change touches no module of the mesh's graph, pass: its merge-check.sh passed), and the combined status is success.
The fix matches the root cause issue 543 names: the old helper returned on the first non-empty closed-connection list, which could be the test's own setup connection closing before the verb subprocess's connection ever appeared. Waiting for want by name instead of "any close" is the correct fix, and the test now also catches a verb naming its connection wrong (not just slow), which the race before this fix could not distinguish.
Verdict: approve at 7630072
Independent review, fresh clone (not ~/projects/novox/mesh-controller), checked out at the PR head 7630072e630979f73e05464c126a251e5412fc2b.
Scope: test code only. The diff touches exactly `cmd/mesh-controller/replay327_test.go` (`closedNames` now takes `want` and loops until a closed connection named `want` appears, or its 5s deadline passes, returning whatever names it last saw either way) and `cmd/mesh-controller/reconnects_test.go` (`TestAVerbsOwnProcessNamesItsConnection` passes `"mesh-controller verb conditions"` as `want`). No production code changes. `closedNames` has exactly one caller, so the signature change is self-contained.
Verified the fix actually detects a wrong name, not just a slow one: locally broke `connectionName` in `cmd/mesh-controller/main.go` (prefixed the verb-name branch with `WRONGNAME`) and reran `TestAVerbsOwnProcessNamesItsConnection`. It failed cleanly in 7.65s with `the verb's connection was named ["mesh-controller" "mesh-controller verb WRONGNAME conditions"]` — a real failure, not a hang, bounded by `closedNames`'s existing 5s deadline. Reverted the break afterward (working tree clean).
Ran the targeted test at the real head: `go test ./cmd/mesh-controller/... -run TestAVerbsOwnProcessNamesItsConnection -race -count=100 -v` — 100 of 100 passed, 0 data races, 93s total.
`go build ./...`, `go vet ./cmd/mesh-controller/...`, and `gofmt -l` on both changed files are all clean. `mesh/merge-gate` and `mesh/repo-check` on the PR's head are both green (`pass: the change touches no module of the mesh's graph`, `pass: its merge-check.sh passed`), and the combined status is `success`.
The fix matches the root cause issue 543 names: the old helper returned on the first non-empty closed-connection list, which could be the test's own setup connection closing before the verb subprocess's connection ever appeared. Waiting for `want` by name instead of "any close" is the correct fix, and the test now also catches a verb naming its connection wrong (not just slow), which the race before this fix could not distinguish.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Fixes issue 543:
TestAVerbsOwnProcessNamesItsConnectionfailed about once in 80 runs (-race -count=20) withthe verb's connection was named []/ a name list missing the verb subprocess.Cause.
closedNames(cmd/mesh-controller/replay327_test.go) returned as soon as the bus's closed-connections list held anything. InTestAVerbsOwnProcessNamesItsConnection, the test's own setup connection (dialled to callEnsureControllerBuckets, then closed before the verb runs as its own process) can register as closed before the verb subprocess's connection does.closedNamescould then return on that setup connection's close alone, before the verb subprocess'smesh-controller verb conditionsconnection ever showed up — a true race, not a logic bug in the assertion itself.What changes
closedNames(replay327_test.go) now takes the name it is waiting for (want) and loops until a closed connection by that exact name appears, or its 5s deadline passes, returning whatever names it last saw either way. It no longer returns on the first non-empty list.TestAVerbsOwnProcessNamesItsConnection(reconnects_test.go) passes"mesh-controller verb conditions"as the name to wait for; its own name check after the call is unchanged (now just confirming whatclosedNamesalready waited for).closedNameshas exactly one caller (TestAVerbsOwnProcessNamesItsConnection), so the signature change touches nothing else.Tested
go build ./...,go vet ./cmd/mesh-controller/...: clean.go test ./cmd/mesh-controller/... -run TestAVerbsOwnProcessNamesItsConnection -race -count=200 -v: 200 run, 200 pass, 0 fail, 0 data races (188s).go test ./cmd/mesh-controller/... -race -count=1: all pass.sh merge-check.sh: gofmt/vet clean, whole suite passes under-race(gcc present); store tests skipped (NOT STORE-CHECKED, noMESH_TEST_POSTGRESon this machine) — this change touches no store-tested code.Test code only, one file's helper plus its one call site. No hand workaround: the helper now waits for the actual fact the test asserts (the verb's own connection appearing closed) instead of racing on any close.
Fixes issue 543.
Delivery
novox/mesh-controller@7630072e6309— delivered since 2026-10-12T11:21:30ZDelivery plan — builds nothing: the change touches no module of the mesh's graph
Transitions
The commit's note under
refs/notes/mesh-plankeeps every transition:git log --notes=mesh-plan.Verdict: approve at
7630072Independent review, fresh clone (not ~/projects/novox/mesh-controller), checked out at the PR head
7630072e63.Scope: test code only. The diff touches exactly
cmd/mesh-controller/replay327_test.go(closedNamesnow takeswantand loops until a closed connection namedwantappears, or its 5s deadline passes, returning whatever names it last saw either way) andcmd/mesh-controller/reconnects_test.go(TestAVerbsOwnProcessNamesItsConnectionpasses"mesh-controller verb conditions"aswant). No production code changes.closedNameshas exactly one caller, so the signature change is self-contained.Verified the fix actually detects a wrong name, not just a slow one: locally broke
connectionNameincmd/mesh-controller/main.go(prefixed the verb-name branch withWRONGNAME) and reranTestAVerbsOwnProcessNamesItsConnection. It failed cleanly in 7.65s withthe verb's connection was named ["mesh-controller" "mesh-controller verb WRONGNAME conditions"]— a real failure, not a hang, bounded byclosedNames's existing 5s deadline. Reverted the break afterward (working tree clean).Ran the targeted test at the real head:
go test ./cmd/mesh-controller/... -run TestAVerbsOwnProcessNamesItsConnection -race -count=100 -v— 100 of 100 passed, 0 data races, 93s total.go build ./...,go vet ./cmd/mesh-controller/..., andgofmt -lon both changed files are all clean.mesh/merge-gateandmesh/repo-checkon the PR's head are both green (pass: the change touches no module of the mesh's graph,pass: its merge-check.sh passed), and the combined status issuccess.The fix matches the root cause issue 543 names: the old helper returned on the first non-empty closed-connection list, which could be the test's own setup connection closing before the verb subprocess's connection ever appeared. Waiting for
wantby name instead of "any close" is the correct fix, and the test now also catches a verb naming its connection wrong (not just slow), which the race before this fix could not distinguish.