Type Inference

These types are exported from @ueberbrot/effect-api-query. Each describes a generated leaf’s contract:

Type	Result
InferInput<Leaf>	The accepted operation input
InferQueryData<Leaf, View>	Cached data for a supported query view, before select
InferQueryError<Leaf, View>	Execution and package failures carried by that query view
InferMutationData<Leaf>	The successful buffered mutation result
InferMutationError<Leaf>	The buffered mutation’s execution error

The helpers follow the utility tree’s contract and ready-client types. Literal declaration names preserve leaf paths and supported views.

Operation input

The utility tree from a client module supplies the leaf types:

const input: InferInput<typeof rpc.users.get> = { id: 1 }
type UserData = InferQueryData<typeof rpc.users.get>
type HistoryData = InferQueryData<typeof rpc.diagnostics.stream, 'streamed'>
type UserError = InferQueryError<typeof rpc.users.get>

RPC input is payload constructor input. Fields supplied by constructor defaults can be omitted. HTTP input is the decoded request with its declared params, query, headers, and payload containers. Multipart HTTP requests use FormData in the payload container. Inputless leaves use void. A call to an options builder can use skipToken to disable execution without changing the leaf’s operation input contract.

Query views and cached data

Query data and error helpers default to query. Streaming leaves require an explicit live or streamed view. A leaf constrains View to the views with supported key and option builders:

Leaf	Query views
Unary RPC	query, infinite
Buffered nonmultipart HTTP	query, infinite, metadata
RPC stream or supported HTTP SSE	streamed, live
Buffered multipart HTTP	Mutation-only

An unsupported explicit view fails typechecking. If a leaf has no ordinary query, omitting the view produces never. That leaf requires a supported explicit view.

Buffered and live query data normalize top-level successful undefined to null. Accumulated histories preserve each emitted value. Infinite data includes pages and page parameters. Metadata contains decoded data, status, and headers. See Data Normalization for ownership rules.

InferQueryData describes cache contents. A native select changes the returned value:

const selectedOptions = rpc.users.get.queryOptions({
  input: { id: 1 },
  select: (value) => value.name,
})
const selectedName = await queryClient.query(selectedOptions)
const cachedUser = queryClient.getQueryData(selectedOptions.queryKey)

In this example, selectedName is a string, and cachedUser is the decoded user or undefined. The generated key retains the unselected cache type. Broad prefix or filter reads can contain different views. Their results remain unknown until the application establishes their representation.

Standalone infinite keys use unknown for page parameters, as does InferQueryData<Leaf, 'infinite'>. An infiniteOptions(...) result retains the page parameter inferred from that call.

Mutation results

The write contract has separate data and error types:

type CreateUserInput = InferInput<typeof rpc.users.create>
type CreatedUser = InferMutationData<typeof rpc.users.create>
type CreateUserError = InferMutationError<typeof rpc.users.create>

Mutation data preserves the successful decoded value, including undefined. The options builder infers variables, data, errors, and the local onMutate result for mutation callbacks. Defaults and Policies describes how reusable typed callbacks preserve local inference.

Failures at runtime

Inferred errors describe declared execution and package failures. Execution errors retain the full Effect Cause. A custom runner rejection or an error from an application callback can pass through unchanged. The inferred union does not validate those unknown failures.

The package’s execution-error guards narrow execution errors. Domain failures need further narrowing before their fields can be read. See Handle Failures and the error reference.