Function append_stamps_identity_and_time
pub async fn append_stamps_identity_and_time<F: Fixture>(
open: impl AsyncFn() -> F,
) -> RuleOutcomeExpand description
VT-4 — the four facts on a SequencedEvent are persisted, not
produced by the read that returned them.
VT-4’s Rejects: line names the adapter that makes identity or time up
at read time, and observes that it “passes every rule that reads back
what it just wrote inside one process”. That is true of a rule that reads
once, and it stops being true the moment one event is reached two
different ways: an identity derived from the row’s ordinal in the result
set, or a time taken from the connection’s clock in the row mapper, both
answer differently when the result set is a different shape — while
agreeing with themselves perfectly under any single query.
So this reads one event through Query::all() and again through a query
that selects it alone, and compares the whole SequencedEvent.
RowOrdinalIdentityStore and ReadTimeClockStore in
tests/mutation_coverage/mutants.rs are the two compiled versions, and
the first of them is right under Query::all() on a densely-allocated
store, which is what makes it survivable and what makes the second read
the whole rule.
The middle event, not the first, so that neither an ordinal-derived
identity nor a first-row special case can coincide with the right answer.
The types ascend on purpose: SortByEventTypeStore orders by (type, position) and would otherwise reorder the batch, which is
read_defaults_to_ascending_order’s subject and not this one’s.
The local shape of the identity — that id.position() is the position
this store assigned — belongs to append_stamps_a_local_event_id, and
the time’s own claim to append_stamps_a_recorded_time. Asserting either
here as well would be one rule wearing three names.