glm-5.3-flash
6f9cfb5778
Replace the harness's blocking std::thread::sleep with tokio::time::sleep across all nine call sites, note the no-blocking-calls-inside-rows posture in the helper's doc comment, and mark the task completed. Both engines' contract suites re-run green with assertions and tolerances unchanged (ADR-024 \u00a75 item 4).
82 lines
3.5 KiB
Markdown
82 lines
3.5 KiB
Markdown
---
|
|
id: suite-harness-current-thread
|
|
name: Contract-suite harness — async sleep in `past_stamp_sleep`, current-thread flavor readiness
|
|
status: completed
|
|
depends_on: []
|
|
scope: narrow
|
|
risk: low
|
|
impact: component
|
|
level: implementation
|
|
tags: [wave-6, contract-suite]
|
|
---
|
|
|
|
## Description
|
|
|
|
The one suite-side change wave 6 carries
|
|
([ADR-024](../architecture/decisions/024-mem-engine.md) §5 item 4; the
|
|
review-003 posture note at `properties.rs:1683`): the suite's
|
|
`past_stamp_sleep` helper uses a blocking `std::thread::sleep` inside
|
|
async rows — safe under the multi-thread test flavors both existing
|
|
engines' targets use, but it would stall a current-thread runtime. The
|
|
mem engine's suite column **must** run under a current-thread flavor
|
|
(`#[tokio::test(flavor = "current_thread")]`) to keep the mem
|
|
engine's internals honestly current-thread-clean — which is what
|
|
wasm32 needs.
|
|
|
|
- Replace `past_stamp_sleep`'s blocking sleep with `tokio::time::sleep`
|
|
(async). The helper's semantics are unchanged (a bounded
|
|
pre-`run_at` settle before reads); the existing rows' determinism
|
|
posture is untouched — verify the two legs using it re-run green on
|
|
both engines with no timing drift (tolerances unchanged).
|
|
- Document (one line in the helper's doc comment or the factory
|
|
contract notes) the harness posture: **no blocking calls inside
|
|
suite rows** — the mem column enforces the flavor that catches
|
|
violations.
|
|
|
|
This task is engine-independent and can run any time before
|
|
`mem-engine-integration` wires the third column.
|
|
|
|
## Acceptance Criteria
|
|
|
|
- [x] `past_stamp_sleep` is async (`tokio::time::sleep`)
|
|
- [x] Both engines' existing suites re-run green with identical
|
|
assertions and tolerances (determinism posture intact)
|
|
- [x] The no-blocking-calls posture noted where the harness's
|
|
determinism posture is documented
|
|
- [x] Clippy `-D warnings`, fmt clean; workspace gates green
|
|
|
|
## References
|
|
|
|
- docs/architecture/decisions/024-mem-engine.md §5
|
|
- docs/reviews/003-wave-5-general-review.md §Finding 5 (the
|
|
`properties.rs:1683` posture note)
|
|
- docs/architecture/decisions/022-contract-suite-layout.md (the
|
|
factory contract the flavor rides)
|
|
|
|
## Notes
|
|
|
|
- The posture note went into `past_stamp_sleep`'s doc comment rather
|
|
than the factory contract notes: the posture is a harness rule about
|
|
row internals (what a row may call), not part of the
|
|
`StoreFactory` shape, and the doc comment sits exactly where a row
|
|
author reaches for a sleep. The comment cites ADR-024 §5 for the
|
|
current-thread flavor that enforces it.
|
|
- Five rows exercise the helper across its two use shapes (the
|
|
lapse-settle legs and the scheduler runner-less-pause leg):
|
|
`lock_ttl_expiry_and_reacquisition`, `job_handle_validity_predicate`,
|
|
`queue_depth_reclaim_and_dead_letter`, `sweep_no_stranded_rows`,
|
|
`scheduler_bounded_catchup`. All nine call sites gained `.await`;
|
|
assertions and tolerances untouched.
|
|
|
|
## Summary
|
|
|
|
`past_stamp_sleep` now awaits `tokio::time::sleep` instead of blocking
|
|
on `std::thread::sleep` (alkstore-contract-suite/src/properties.rs),
|
|
making every suite row current-thread-safe for the mem engine's
|
|
`#[tokio::test(flavor = "current_thread")]` column (ADR-024 §5 item 4);
|
|
the doc comment records the no-blocking-calls-inside-rows posture.
|
|
Both engines' `contract_suite` re-ran fully green (25/25 each, SQLite
|
|
15.0 s / Postgres 21.3 s — no timing drift, tolerances unchanged); the
|
|
full workspace build, tests, `clippy -D warnings`, and `fmt --check`
|
|
pass. PostgreSQL leg ran against the POC container (`pglo-poc`, port
|
|
15432, creds per docs/reviews/003 §env note). |