[chappie-daemon]

a_close_during_the_invite_mint_leaves_the_event_but_not_the_read synchronised a CLOSE against the invite mint with a fixed 50 ms sleep. Under load the CLOSE handler could be delayed past that sleep, the mint resumed first, and the test observed the stored EVENT and EOSE it asserts must not arrive — a false failure with a correct fix.

Closes #963.

The fix

The test moves down one level and is ordered by program order instead of by a timer:

• it owns a Connection, a FrameSink and a context, per the family tests/connection.rs:414-424 established;
• it spawns subscribe::begin (the seam fix(relay): run a handler's ordered prefix on the connection loop #978 introduced, whose prefix registers the read before returning) and calls the synchronous unsubscribe::handle in its own task;
• it asserts !read_in_flight("invite") before gate.add_permits(1), so the abort is a program-order predecessor of the release — nothing between them can suspend;
• then it drains one EVENT (kind 28935) and asserts the sink is empty.

sleep/Duration in the test body: 2 → 0.

What proves it still decides

The oracle is a mutation, not a pass rate. With abort_read moved after reading.await, the test fails at a closed subscription streams nothing and the leaked frame is ["EVENT","invite",{…"kind":1,…"content":"stored"…}] — #963's exact defect. Also measured: changing begin_read(&subscription_id) to begin_read("") fails the registration assertion, so that assertion can still catch a begin that stopped registering.

The 200 ms injection that proves the old test red cannot discriminate the new one — it never enters handlers::dispatch — and build.md says so rather than reporting a green as evidence.

What it costs

The wire-level combination this test used to cover — invite EVENT plus CLOSE mid-mint, over a socket — is now covered by nothing. That is #983, filed with the shape a future fix needs (the relay publishes its per-connection registry; a wire test polls it between the CLOSE and the release; production source, five Relay::new sites). Every other piece the test gave up is carried elsewhere, and the carriers are green at this base.

Two smaller deltas found while doing this

• a_close_aborts_an_in_flight_read_without_a_frame stays green under the same delay it exists to catch (relay: a close-aborts-read test stays green under the delay it exists to catch #982) — not this branch's, and that issue needs re-measuring after fix(relay): run a handler's ordered prefix on the connection loop #978 moved this path.
• Two sentences around this test claim more than it proves (relay: two sentences around the CLOSE-during-mint test claim more than it proves #984).

Verification

Base rewrite/main @ 1cb6265 (the branch was rebased onto it after #978 merged mid-review and broke the call site). cargo test --workspace --all-features and the release build both exit 0 at this head; handlers target 52 passed; 0 failed; all five repository commands clean; git merge-tree --write-tree produces a tree byte-identical to HEAD^{tree}.

Notes

• Commit hooks did not run: .husky/_/husky.sh is absent and the hook drives the legacy pnpm checks, so commits used HUSKY=0 git commit --no-verify.
• Documentation repair in this job was capped at one round by the operator as an experiment; the residue is relay: two sentences around the CLOSE-during-mint test claim more than it proves #984 rather than a third round.

[changeset-bot]

🦋 Changeset detected

Latest commit: fa1f09b

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 0 packages

When changesets are added to this PR, you'll see the packages that this PR includes changesets for and the associated semver types

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[greptile-apps]

Confidence Score: 5/5

[Low impact] The test-only change appears safe to merge.

Summary

Replaces the CLOSE-during-mint test's socket timing with direct handler calls.

• The invite-mint test orders CLOSE before releasing the mint.

No actionable issues found. No production code changes.

chappie-daemon acknowledged the lost socket coverage (#983), deferred wording fixes (#984), and identified a pre-existing weakness in the neighboring test (#982).

_{Reviews (1) · Last reviewed commit: "docs(relay): say what the CLOSE test own..." · Reviewed by Greptile}