test: the test tree holds zero hand-woven concurrency gates — 79 → 0, allow file deleted by rbuergi · Pull Request #2806 · Systemorph/MeshWeaver

rbuergi · 2026-08-30T16:36:04Z

The ratchet landed in #2782 froze a measured inventory: 79 hand-woven concurrency gates across 23 files in test/, every one a ManualResetEventSlim, 32 of them in IoPoolTest.cs. This is the sweep. All 79 are converted, test moves into the guard's ProductionRoots, and test/HandWovenGateSites.allow is deleted — there is no allow file left, for any root.

Why a gate in a TEST is a defect, not a style choice

In product code the failure is obvious: a parked hub action block or grain turn is a deadlock. In a test the same primitive fails more quietly, and it has bitten twice already:

A stranded worker. IoPoolResidualNamesItsPoolTest put its release.Set() after an await of the method-timeout token. On timeout that line never ran and the leaf it had blocked kept a pool thread for two minutes, into the next test.
A ratchet swap. The first repair traded the event for AsyncSubject + .Wait() — an observable→blocking bridge, which parks the thread just the same and trips BlockingBridgeInTestRatchetGuard instead. Satisfying one guard by violating another is not a fix, and this PR deliberately does not do it: the blocking-bridge count is unchanged.

Two shapes, chosen by which way the signal travels

Producer → test — a handler reporting it started, a pooled leaf, a .Finally, a disposal callback:

var entered = new AsyncSubject<Unit>();
… entered.OnNext(Unit.Default); entered.OnCompleted(); …
await entered.Should().Within(Timeout5).Emit("the leaf must be running before the drain");

The negative direction is the same subject with NotEmit(within) — the one place a fixed wait is correct, because there is no positive signal to await.

Test → a worker it deliberately parks — a wedged action block, a leaf that ignores its cancellation token, a merge turn held open so an ack provably cannot arrive. The park stays: it is the subject. The kernel handle goes:

var releaseGate = 0;
… SpinWait.SpinUntil(() => Volatile.Read(ref releaseGate) == 1, TimeSpan.FromSeconds(60)); …
try     { … Volatile.Write(ref releaseGate, 1); … }
finally { Volatile.Write(ref releaseGate, 1); }   // 🚨 the load-bearing line

SpinUntil(predicate, timeout) is already the house idiom in these same files (IoPoolTest's own remarks named it), needs no using, and returns the same bool the event's Wait(timeout) did — so the sites that measured whether the release arrived (legAReleasedCleanly, GateTimedOut, pumpCompletedWhileBlocked, answered) keep their assertion unchanged. Like the event, it never pumps a SynchronizationContext, which is load-bearing where the test's whole point is that the blocked thread must not run a queued continuation.

🚨 Every release is in a finally, several of them newly added. Where a helper owns the park (WriteWaitsForCommitVerdictTest's OwnerGate), Dispose() releases and the call site is using var gate = await ParkOwnerMergeTurn(path).

Not touched: Thread.Join(timeout) and IsSet-style state reads. They answer "did that thread finish" / "has this happened", not "wait for a signal", and Join is the stricter primitive (#2792).

No assertion weakened, no timeout widened. Several void tests became async Task to await the reactive assertions — the normal shape.

Measured

	before	after
sites (guard's own scanner, comments + literals masked)	79	0
files	23	0
`test/MeshWeaver.Hosting.Test/IoPoolTest.cs`	32	0
`test/HandWovenGateSites.allow`	79 entries	deleted

Per project: Hosting.Test 39 · Messaging.Hub.Test 14 · Hosting.Monolith.Test 14 · Hosting.Orleans.Test 6 · Graph.Test 3 · Layout.Test 1 · Persistence.Test 1 · Autocomplete.Test 1.

Verified


`dotnet build -c Release -warnaserror`, one project per invocation	all 9 touched test projects → `0 Warning(s)` `0 Error(s)`
every converted class, run by filter	105 tests, `Passed!`, 0 failed — Hosting 30, Autocomplete 23, Messaging.Hub 21, Orleans 11, Graph 10, Monolith 8, Layout 1, Persistence 1
`HandWovenGateRatchetGuard` · `BlockingBridgeInTestRatchetGuard` · `ImpersonationScopeSiteRatchetGuard` · `ObservableToTaskBridgeGuard` · `RouterAsTestRequestOriginRatchetGuard`	20/20 green

Mutation proof — the ratchet still bites, in both directions

1. A new gate must fail. Planted new ManualResetEventSlim(false) in QueryInRenderDeadlockTest.cs:

[FAIL] HandWovenGateRatchetGuard.ProductionCodeHasNoHandWovenGate_OutsideTheSanctionedOne
🚨 A hand-woven concurrency gate parks a thread … Do NOT add an allow entry — there is no
allow file, for any root.
  test/MeshWeaver.Layout.Test/QueryInRenderDeadlockTest.cs (1)
Failed!  - Failed: 1, Passed: 3, Total: 4

Reverted → green.

2. A resurrected allow file must fail. Re-created test/HandWovenGateSites.allow:

[FAIL] HandWovenGateRatchetGuard.TheTestTreeAllowFileStaysDeleted
test/HandWovenGateSites.allow is back. It was deleted when the last of its 79 seeded sites
was converted, and `test` moved into ProductionRoots in the same change …
Failed!  - Failed: 1, Passed: 3, Total: 4

Deleted → green (4/4). That second check is new here: a cleared inventory that is quietly re-seeded is exactly how one grows back, and SourceScan.ReadAllowFile is explicit that a regenerated allow file blesses whatever happens to be in the tree.

Work products conserved

Removing Hand-Woven Concurrency Gates — the two shapes, why the finally is load-bearing, Thread.Join vs an event, what the sweep measured, and how the guard is proven.
A What's New entry (Category: Feature).
AGENTS.md, the /async skill and Doc/Architecture/AsynchronousCalls now state the rule as covering src/ or test/, and name the two replacement shapes so the next author does not reach for an event.

The only surviving exemptions are the two verified SanctionedGates: the one sealed inside IoPool, and the standalone ThumbnailGenerator CLI whose premise (no MeshWeaver ProjectReference) is re-checked on every run.

🤖 Generated with Claude Code

… allow file deleted The ratchet added yesterday froze the debt at a MEASURED 79 sites across 23 files, every one a `ManualResetEventSlim`, 32 of them in `IoPoolTest.cs`. All 79 are converted here, so `test` moves into `HandWovenGateRatchetGuard`'s `ProductionRoots` and `test/HandWovenGateSites.allow` is deleted — there is no allow file left, for any root. ## Two shapes, chosen by which way the signal travels - **Producer → test** (a handler reporting it started, a pooled leaf, a `.Finally`, a disposal callback) → an `AsyncSubject<Unit>` the producer completes, awaited through the house assertion helpers: `await x.Should().Within(...).Emit(because)`, or `.NotEmit(within)` for the negative. Never `.Wait()` on it — that swaps this ratchet for `BlockingBridgeInTestRatchetGuard`, the exact mistake made on #2786 and caught. - **Test → a worker it deliberately parks** (a wedged action block, a leaf that ignores its cancellation token, a merge turn held so an ack provably cannot arrive) → the park STAYS, because the park is the subject; the kernel handle goes. A volatile `int` the worker polls under a bounded `SpinWait.SpinUntil(predicate, budget)` — already the house idiom in these same files — returning the same `bool` the event's `Wait(timeout)` did, so sites that MEASURED whether the release arrived keep their assertion unchanged. 🚨 Every release is written in a `finally`, several of which are new. Without it an assertion that throws first leaves the worker parked for its full bound, into the next test — the two-minute pool-thread leak the allow file's own header records. No assertion weakened, no timeout widened. `Thread.Join(timeout)` and `IsSet`-style state reads were left as they are: they answer "did it finish" / "has it happened", not "wait". ## Measured before / after | | before | after | |---|---|---| | sites (guard's own scanner) | 79 | **0** | | files | 23 | 0 | | `IoPoolTest.cs` | 32 | 0 | | allow file | 79 lines | deleted | Touched: Hosting.Test (IoPoolTest 32, LiveQueryHandoffDropTest 6, LiveQueryForeignTrampolineTest 1), Messaging.Hub.Test (MessageHubTest 6, + 4 files × 2), Hosting.Monolith.Test (7 files × 2), Hosting.Orleans.Test (RoutingGrainDeliveryRetryTest 3, IoPoolSiloTeardownTest 2, RoutingGrainTurnIsolationTest 1), Graph.Test (3), Layout.Test (1), Persistence.Test (1), Autocomplete.Test (1). ## Verified - `dotnet build -c Release -warnaserror`, one project per invocation, on all 9 touched test projects → `0 Warning(s) 0 Error(s)` each. - Every converted class run by filter: 105 tests, `Passed!` with 0 failures — Hosting 30, Autocomplete 23, Messaging.Hub 21, Orleans 11, Graph 10, Monolith 8, Layout 1, Persistence 1, plus the 20 guard tests. - `HandWovenGateRatchetGuard`, `BlockingBridgeInTestRatchetGuard`, `ImpersonationScopeSiteRatchetGuard`, `ObservableToTaskBridgeGuard`, `RouterAsTestRequestOriginRatchetGuard` — all green (20/20). ## Mutation proof — the ratchet still bites, twice 1. Planted `new ManualResetEventSlim(false)` in `QueryInRenderDeadlockTest.cs` → `ProductionCodeHasNoHandWovenGate_OutsideTheSanctionedOne` **FAILED**, naming `test/MeshWeaver.Layout.Test/QueryInRenderDeadlockTest.cs (1)`. Reverted → green. 2. Re-created `test/HandWovenGateSites.allow` → the new `TheTestTreeAllowFileStaysDeleted` **FAILED** ("…is back"). Deleted → green (4/4). Work products conserved: `Doc/Architecture/RemovingHandWovenGates` carries the two shapes, the `finally` rule, the sweep's numbers and how the ratchet is proven; AGENTS.md, the `/async` skill and `Doc/Architecture/AsynchronousCalls` now state the rule as covering `test/` too. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Copilot

Pull request overview

This PR completes the “test tree to zero” sweep by removing all remaining hand-woven concurrency gates from test/, deleting the last allow file, and updating the ratchet + documentation so the whole repo (including tests) is now held at zero.

Changes:

Replaced ManualResetEventSlim-style gates across the test suite with AsyncSubject<Unit> (producer → test) and volatile flags + bounded SpinWait.SpinUntil (test → deliberately parked worker).
Moved test/ into the guard’s zero-tolerance roots and deleted test/HandWovenGateSites.allow, adding a regression check that fails if it reappears.
Added/updated docs (architecture + What’s New) and repo guidance (AGENTS.md, /async skill, AsynchronousCalls) to describe the two sanctioned replacement shapes.

Reviewed changes

Copilot reviewed 31 out of 31 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
test/MeshWeaver.Persistence.Test/PartitionObjectsSubscriberIndependenceTest.cs	Replaces event gate with volatile flag + `SpinUntil` to keep the “blocked context” scenario without kernel handles.
test/MeshWeaver.Messaging.Hub.Test/TeardownHubCreationFreezeTest.cs	Converts handler/test gating to `AsyncSubject` + volatile release; test becomes `async Task`.
test/MeshWeaver.Messaging.Hub.Test/MessageHubTest.cs	Converts multiple gate sites; adds `finally`-release to avoid stranding a parked turn thread on assertion failure.
test/MeshWeaver.Messaging.Hub.Test/InflightHubCreationDrainTest.cs	Converts construction/test gating; wraps release in `finally` to prevent stranded construction thread.
test/MeshWeaver.Messaging.Hub.Test/HostedHubDisposalJoinTest.cs	Same conversion pattern as inflight construction drain; ensures release is guaranteed.
test/MeshWeaver.Messaging.Hub.Test/DisposalRaceNackTest.cs	Converts deterministic handler stall/release gates to `AsyncSubject` + volatile flag.
test/MeshWeaver.Layout.Test/QueryInRenderDeadlockTest.cs	Replaces `ManualResetEventSlim` with volatile flag + `SpinUntil` to preserve the “blocked subscribing thread” subject.
test/MeshWeaver.Hosting.Test/LiveQueryHandoffDropTest.cs	Removes event gates; introduces `GatedAdapter` helpers using `AsyncSubject` + volatile release; converts key tests to `async Task`.
test/MeshWeaver.Hosting.Test/LiveQueryForeignTrampolineTest.cs	Replaces event with volatile flag + `SpinUntil` for trampoline-blocking probe.
test/MeshWeaver.Hosting.Test/IoPoolTest.cs	Broad gate conversion to `AsyncSubject` + volatile releases, adding `try/finally` releases to prevent stranded pool threads.
test/MeshWeaver.Hosting.Orleans.Test/RoutingGrainTurnIsolationTest.cs	Converts injected stall gate from `ManualResetEventSlim` to volatile flag + `SpinUntil`.
test/MeshWeaver.Hosting.Orleans.Test/RoutingGrainDeliveryRetryTest.cs	Converts callback completion signals to `AsyncSubject<Unit>` and awaits via assertion helpers.
test/MeshWeaver.Hosting.Orleans.Test/IoPoolSiloTeardownTest.cs	Replaces event gates with `AsyncSubject` + volatile state to assert teardown joins pooled leaf.
test/MeshWeaver.Hosting.Monolith.Test/WriteWaitsForCommitVerdictTest.cs	Reworks merge-turn parking helper to be `async` and uses `AsyncSubject` + volatile release via an `OwnerGate`.
test/MeshWeaver.Hosting.Monolith.Test/ShutdownRoutingRejectClassificationTest.cs	Converts mesh action-block stall gating to `AsyncSubject` + volatile release in `finally`.
test/MeshWeaver.Hosting.Monolith.Test/ReleaseRequestWatcherHighWaterTest.cs	Converts “park owner write queue” gate to `AsyncSubject` + volatile release with `finally` to avoid wedging the class.
test/MeshWeaver.Hosting.Monolith.Test/QueuedWriteAdvancesOnHandoffTest.cs	Converts parked-turn handshake to `AsyncSubject` + volatile release, adding `finally` release.
test/MeshWeaver.Hosting.Monolith.Test/OwnerDisposalNackTest.cs	Same conversion for owner merge executor park/release.
test/MeshWeaver.Hosting.Monolith.Test/LayoutRenderTeardownDrainTest.cs	Converts render-in-flight and release signals to `AsyncSubject` + volatile release, removes event disposal.
test/MeshWeaver.Hosting.Monolith.Test/LateNackReenqueueTest.cs	Converts parked-turn handshake to `AsyncSubject` + volatile release with `finally`.
test/MeshWeaver.Hosting.Monolith.Test/DeleteNodeBehaviorTest.cs	Updates commentary to reflect reactive-await assertion shape (no hand-woven gate).
test/MeshWeaver.Graph.Test/SyncedQueryRlsForeignTrampolineTest.cs	Replaces event gate with volatile flag + `SpinUntil` to preserve trampoline-blocking subject.
test/MeshWeaver.Graph.Test/NodeTypeAdoptionRegistryTest.cs	Replaces event gating with `AsyncSubject<Unit>` and reactive assertion waits.
test/MeshWeaver.Documentation.Test/HandWovenGateRatchetGuard.cs	Moves `test` into zero-tolerance roots, deletes ratcheted inventory logic, and adds “allow file stays deleted” check.
test/MeshWeaver.Autocomplete.Test/AutocompleteIntegrationTest.cs	Removes `ManualResetEventSlim` waiting in timing diagnostics in favor of `AsyncSubject` + reactive waiting.
test/HandWovenGateSites.allow	Deletes the seeded inventory allow file (no longer tolerated anywhere).
src/MeshWeaver.Documentation/Data/WhatsNew/2026-08-30-the-test-tree-holds-zero-hand-woven-gates.md	Adds What’s New entry explaining the sweep and new invariant.
src/MeshWeaver.Documentation/Data/Architecture/RemovingHandWovenGates.md	Adds architecture doc describing the two replacement shapes and why `finally`-release is required.
src/MeshWeaver.Documentation/Data/Architecture/AsynchronousCalls.md	Updates rule to explicitly cover `test/` and points to the new companion doc.
AGENTS.md	Updates repo-wide guidance to make the “no gates in test” rule explicit and actionable.
.claude/skills/async/SKILL.md	Updates `/async` skill to include `test/` and the two sanctioned replacement shapes.

Suppressed comments (1)

test/MeshWeaver.Hosting.Test/IoPoolTest.cs:498

After switching the async gate to an AsyncSubject<int>, complete it by publishing a value and calling OnCompleted() (idempotently, including in the finally) rather than TrySetResult on a TaskCompletionSource.

            Volatile.Write(ref release, 1);
            asyncGate.TrySetResult(0);

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

github-actions · 2026-08-30T16:42:57Z

Test Results (shard 3)

444 tests ±0 444 ✅ ±0 4m 44s ⏱️ -5s
6 suites ±0 0 💤 ±0
6 files ±0 0 ❌ ±0

Results for commit b3cf53c. ± Comparison against base commit afb5a5e.

github-actions · 2026-08-30T16:45:46Z

Test Results (shard 5)

2 184 tests ±0 2 180 ✅ ±0 6m 31s ⏱️ +49s
10 suites ±0 4 💤 ±0
10 files ±0 0 ❌ ±0

Results for commit b3cf53c. ± Comparison against base commit afb5a5e.

github-actions · 2026-08-30T16:46:21Z

Test Results (shard 0)

1 178 tests ±0 985 ✅ ±0 5m 26s ⏱️ -8s
5 suites ±0 193 💤 ±0
5 files ±0 0 ❌ ±0

Results for commit b3cf53c. ± Comparison against base commit afb5a5e.

This pull request removes 1 and adds 1 tests. Note that renamed tests count towards both.

MeshWeaver.Documentation.Test.HandWovenGateRatchetGuard ‑ TheTestTreeInventoryOnlyShrinks

MeshWeaver.Documentation.Test.HandWovenGateRatchetGuard ‑ TheTestTreeAllowFileStaysDeleted

github-actions · 2026-08-30T16:46:27Z

Test Results (shard 4)

10 files ±0 10 suites ±0 6m 4s ⏱️ +32s
3 084 tests ±0 3 084 ✅ ±0 0 💤 ±0 0 ❌ ±0
3 753 runs +2 3 753 ✅ +2 0 💤 ±0 0 ❌ ±0

Results for commit b3cf53c. ± Comparison against base commit afb5a5e.

github-actions · 2026-08-30T16:46:32Z

Test Results (shard 2)

624 tests ±0 624 ✅ ±0 6m 11s ⏱️ +4s
6 suites ±0 0 💤 ±0
6 files ±0 0 ❌ ±0

Results for commit b3cf53c. ± Comparison against base commit afb5a5e.

github-actions · 2026-08-30T16:46:35Z

Test Results (shard 1)

988 tests ±0 988 ✅ ±0 6m 1s ⏱️ +12s
5 suites ±0 0 💤 ±0
5 files ±0 0 ❌ ±0

Results for commit b3cf53c. ± Comparison against base commit afb5a5e.

github-actions · 2026-08-30T16:47:02Z

Test Results

42 files ±0 42 suites ±0 35m 0s ⏱️ + 1m 26s
8 502 tests ±0 8 305 ✅ ±0 197 💤 ±0 0 ❌ ±0
9 171 runs +2 8 974 ✅ +2 197 💤 ±0 0 ❌ ±0

Results for commit b3cf53c. ± Comparison against base commit afb5a5e.

This pull request removes 1 and adds 1 tests. Note that renamed tests count towards both.

MeshWeaver.Documentation.Test.HandWovenGateRatchetGuard ‑ TheTestTreeInventoryOnlyShrinks

MeshWeaver.Documentation.Test.HandWovenGateRatchetGuard ‑ TheTestTreeAllowFileStaysDeleted

Copilot AI lite review requested due to automatic review settings August 30, 2026 16:36

Copilot started reviewing on behalf of rbuergi August 30, 2026 16:36

View session

Copilot AI reviewed Aug 30, 2026

View reviewed changes

Comment thread test/MeshWeaver.Hosting.Test/IoPoolTest.cs

rbuergi added this pull request to the merge queue Aug 30, 2026

rbuergi mentioned this pull request Aug 30, 2026

IoPoolTest holds two TaskCompletionSource-as-a-signal gates (one without RunContinuationsAsynchronously) #2809

Closed

Merged via the queue into main with commit b05e0ea Aug 30, 2026
26 checks passed

rbuergi deleted the test/hand-woven-gates-to-zero branch August 30, 2026 17:04

rbuergi mentioned this pull request Aug 30, 2026

fix(#2809): the two IoPoolTest gates use the sanctioned bridge, and release in a finally #2822

Merged

Navigation Menu

Repository navigation

test: the test tree holds zero hand-woven concurrency gates — 79 → 0, allow file deleted - #2806

test: the test tree holds zero hand-woven concurrency gates — 79 → 0, allow file deleted#2806
rbuergi merged 1 commit into
mainfrom
test/hand-woven-gates-to-zero

rbuergi commented Aug 30, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

github-actions Bot commented Aug 30, 2026

Uh oh!

github-actions Bot commented Aug 30, 2026

Uh oh!

github-actions Bot commented Aug 30, 2026

Uh oh!

github-actions Bot commented Aug 30, 2026

Uh oh!

github-actions Bot commented Aug 30, 2026

Uh oh!

github-actions Bot commented Aug 30, 2026

Uh oh!

github-actions Bot commented Aug 30, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

rbuergi commented Aug 30, 2026

Why a gate in a TEST is a defect, not a style choice

Two shapes, chosen by which way the signal travels

Measured

Verified

Mutation proof — the ratchet still bites, in both directions

Work products conserved

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

github-actions Bot commented Aug 30, 2026

Test Results (shard 3)

Uh oh!

github-actions Bot commented Aug 30, 2026

Test Results (shard 5)

Uh oh!

github-actions Bot commented Aug 30, 2026

Test Results (shard 0)

Uh oh!

github-actions Bot commented Aug 30, 2026

Test Results (shard 4)

Uh oh!

github-actions Bot commented Aug 30, 2026

Test Results (shard 2)

Uh oh!

github-actions Bot commented Aug 30, 2026

Test Results (shard 1)

Uh oh!

github-actions Bot commented Aug 30, 2026

Test Results

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants