The data-access seam a backend provides per entity.
One instance is bound to one entity (table/model/collection). The data
source owns query evaluation end to end: it applies where, orderBy,
offset/limit and select itself, and the repository above it must not
re-apply any of them.
Transaction-scoped instances may defer their writes. A backend whose
store has no interactive transaction (Cloudflare D1, where a pre-declared
batch() is the only unit of atomicity) has to buffer writes and apply them
at commit. On such a backend the write methods below describe what WILL be
written rather than what already is, and reads observe committed state only.
A backend that defers must document it; D1Adapter is the worked example.
findAll(query: NormalizedQuery): Promise<Record<string, unknown>[]>
Find every entity matching the normalized query.
Find a single entity by its primary key value.
The key is an EntityKey: a scalar string/number or a
composite record. A composite record is refused by the backend when it
maps to a single column (e.g. Prisma's where: { id }), so widening this
parameter is source-compatible for callers but breaking for an
out-of-repo implementor whose method still declares the scalar-only form.
create(data: Partial<Record<string, unknown>>): Promise<Record<string, unknown>>
Insert a new entity.
Update an existing entity by primary key.
findPage(query: NormalizedQuery): Promise<PageResult>
Find a page of entities by cursor pagination.
Optional. The member exists on every in-repo adapter but is declared
optional so an out-of-repo implementor that cannot page by cursor keeps
compiling. When it is absent, the repository above refuses cursor paging
by name (UnsupportedQueryFeatureError): absence means "cannot page by
cursor", never "there are no more rows" — the distinction an absent
member cannot otherwise preserve.
count(where: Record<string, unknown>,filter?: FilterExpression): Promise<number>
Count entities matching a filter.