connection

connection() wraps a collection fetch to provide the utilities for working with GraphQL cursor connections.

The underlying collection step may yield:

• an array,
• an object with an items array ({ items: any[] }), or
• support the .items() method.

export function connection<...>(
  step: StepRepresentingList<...>,
  params?: ConnectionParams,
): ConnectionStep<...>

interface ConnectionParams {
  fieldArgs?: FieldArgs;
  edgeDataPlan?: ($item: Step<any>) => Step;
}

Wrapping a step in connection() may mutate the step!

If your step implements paginationSupport and applyPagination, then connection(step) will call step.applyPagination(...). If you reuse the same step elsewhere, the pagination limits may leak across.

Create fresh steps for connections. Steps without paginationSupport are unaffected.

step.paginationSupport

If your underlying step can handle pagination directly, it should expose its capabilities via a paginationSupport object.

If paginationSupport is present, support for limit is assumed. Support for other features is indicated via the flags:

• offset?: boolean The step supports numeric offsets — i.e. skipping a fixed number of records before returning results. If combined with cursor, the offset is applied after the cursor.

• cursor?: boolean The step supports cursor-based pagination (after). If you support cursor but not reverse, then reverse queries (last, before) will result in an error, so do not expose those arguments in your GraphQL schema. If you cannot handle cursor + offset together, set cursor: true and offset: false (since cursor is more impactful than offset).

• reverse?: boolean Indicates support for reverse pagination (requires cursor). In reverse mode, the step must apply limit, offset, and after starting from the end of the collection and working backwards. A typical strategy is: invert the sort order in your data source, apply the arguments as usual, then restore the original order. Do not return the final list reversed.

• full?: boolean Advanced. The step implements the full ConnectionHandlingStep contract. In this case, the step itself receives the raw GraphQL pagination parameters (first, last, before, after, offset, etc) via setters such as .setFirst(...) and must yield an object of the form:

  {
    items: ReadonlyArray<TItem> | AsyncIterable<TItem>,
    hasNextPage: boolean,
    hasPreviousPage: boolean
  }

step.applyPagination($params)

When step.paginationSupport is present (but does not contain full: true), connection() will call step.applyPagination($params) with a step that yields a PaginationParams object containing:

• limit: number | null — maximum rows to fetch, where null means no limit
• reverse: boolean — whether to paginate backwards (always false if unsupported)
• offset: number | null — rows to skip (applied after after if cursor is set); null (and 0) means no offset
• after: string | null — exclusive lower bound cursor (or upper bound in reverse mode); null if no cursor was provided, or if you didn’t indicate support for cursor
• stream: ExecutionDetailsStream | null — streaming hints (e.g. stream?.initialCount)

There are other properties in this object starting with __ — these must be ignored, and may change over time. They're used internally by connection().

Your step must honour the subset of fields corresponding to the features you declared in paginationSupport.

step.cursorForItem($item)

If the step indicates cursor: true in paginationSupport, the step must implement a cursorForItem($item: Step<TItem>): Step<string> method. This is called for each item in the collection to produce its stable, opaque cursor string.

Cursors must be consistent across forward and reverse pagination; the same item must always map to the same cursor value.

If you don’t indicate support for cursor, connection() will automatically fall back to numeric cursors (based on the item’s index).

Performance

Without paginationSupport, connection() handles everything in memory. That means fetching the entire dataset, which is usually inefficient.

Adding paginationSupport pushes supported concerns (limit, offset, cursor, reverse) down into your data source for efficiency.

Other notes

• If reverse is not supported, don’t expose before/last args in your schema, or we may be forced to fetch the whole collection.
• If cursor is supported, the step must also implement cursorForItem($item: Step<TItem>): Step<string> to provide stable cursors.
• Optional hooks nodeForItem, edgeForItem, and listItem give finer control over how nodes and edges are represented.
• connectionClone() can produce an unpaginated copy of the step, e.g. for totalCount or aggregates.