Function reissued_conditional_batch_lands_once
pub async fn reissued_conditional_batch_lands_once<F: Fixture>(
open: impl AsyncFn() -> F,
) -> RuleOutcomeExpand description
A conditional append whose condition matches its own events is at-most-once under verbatim reissue.
ES-24 [FROZEN], and the guarantee costs nothing to provide because it
falls out of the condition the caller already wrote. After a dropped
future (ES-22, ES-23) the caller does not know whether its append landed.
Reissuing the identical batch resolves it: a refusal means the first
attempt landed, Ok means it had not and now has. Either way the store
holds exactly one copy, and the caller needs no identity, no idempotency
key and no new API — which is what makes this clause severable from
EventId.
What it rejects is a store that writes before it decides.
WriteThenCheckStore — autocommit plus a separate probe — extends its log
and then evaluates the condition, so the batch’s own events are in the
set the probe reads and the first attempt is refused while its rows
stay down. A caller following the documented resolution procedure reads
that refusal as “my write already landed”, stops, and has written nothing
at all on the one path where the answer matters.
§Two deliberate choices in the arrangement
The store is seeded first, so that no assertion here runs against an
empty store: condition_against_an_empty_store_admits_the_append owns
that degenerate case and NullAggregateProbeStore is its mutant, and a
rule that started empty would fail it for a reason that is not this
rule’s.
The second attempt is asserted with is_err, not with a match on
ConditionViolated. Which error a rejection is reported as is ES-25’s
MUST and condition_rejection_is_reported_as_condition_violated owns it;
demanding the discriminant here would make ViolationAsStoreErrorStore
fail a rule about reissue for a reason that has nothing to do with
reissue.