Data Normalization

Query-stable RPC payloads

A payload-bearing query constructs and normalizes its payload before it builds the key. The ready RPC client constructs that value again during execution, so the payload Schema must be query-stable: reconstruction must preserve every field that affects the RPC result. Constructor defaults then describe both execution and cache identity.

Inputs that normalize to the same payload then share a cache entry. Custom key encoders also receive the normalized payload.

Decoded HTTP inputs

HTTP builders accept the decoded params, query, headers, and payload values declared by the endpoint. They do not apply RPC constructor defaults. Key preparation schema-encodes those values, while the ready HttpApiClient encodes them for execution.

Every declared request container remains required, even when its fields are optional. For example, an endpoint with optional query filters still takes input: { query: {} } when no filter is selected.

No-content and undefined query results

TanStack Query rejects undefined as successful query data. When an RPC or HTTP query succeeds with undefined, the generated query resolves to null instead. The exported QueryData<A> type models that rule.

An HTTP endpoint using HttpApiSchema.NoContent decodes its successful 204 response to undefined, so its ordinary and infinite queries cache null.

Mutations keep the original success type. A mutation that succeeds with undefined still resolves to undefined.

The packed RPC consumer and packed HTTP consumer exercise these normalization rules.