Function acknowledged_writes_survive_a_reopen
pub async fn acknowledged_writes_survive_a_reopen<F: Fixture>(
open: impl AsyncFn() -> F,
) -> RuleOutcomeExpand description
An append acknowledged with Ok is still there after a reopen.
This rejects an adapter that acknowledges before its COMMIT returns: a
pooled store doing its work in spawn_blocking and answering on the
join, a Durable Object relying on output-gate semantics it does not
actually have, any connection setup carrying PRAGMA synchronous = OFF.
This rule is formally CF-14, which is deferred to phase 8, and landing
it now is a named exception. Its adapter far end — a real store that
can lose a write under a real fault — is phase 8’s to build. It is here
because without it the fixture contract ships with exactly one
capability-gated rule and no second one to observe, leaving REOPEN’s
provisional marker and the skip machinery both untested.
Its fixture far end is not deferred and is in the tree:
DurableFixture in tests/fixture_instruments.rs replays a durable log
into a fresh store on reopen, so this body executes on every gate run
rather than reporting a skip under all three shipped harnesses — and
LosingFixture in tests/mutation_coverage/mutants.rs, which never
records at all, is the registered mutant that shows it can fail.
MemoryFixture and LocalFixture both decline REOPEN for honest
reasons and still report the skip, which is what CF-18 is for.
The handle is dropped before the reopen on purpose: a handle that carried the data across would make the rule pass for the wrong reason.