pond-ts API Reference (core)
    Preparing search index...

    Type Alias PublicColumnForKind<K>

    PublicColumnForKind: K extends "number"
        ? Float64Column
        | ChunkedFloat64Column
        : K extends "boolean"
            ? BooleanColumn
            | ChunkedBooleanColumn
            : K extends "string"
                ? StringColumn
                | ChunkedStringColumn
                : K extends "array" ? ArrayColumn | ChunkedArrayColumn : never

    Public column type for a given declared schema kind. Used by the schema-narrowed TimeSeries.column(name) signature (RFC §7.2) so series.column('value') is typed as Float64Column | ChunkedFloat64Column, series.column('host') as StringColumn | ChunkedStringColumn, etc.

    Both packed and chunked variants carry the full public method surface (per the augmentations below). Chunked methods delegate to materialize().method() for v1 — correct, but ~2× the cost of the packed-native path. A future PR may add chunked-native implementations for the hot reductions; the v1 contract holds either way.

    Why include chunked in the narrow return rather than packing- only? An earlier draft narrowed to packed-only and pushed chunked callers to the wide column(name: string) overload, but L2 review on PR #155 flagged this as a runtime type-safety hole: concatSorted and other substrate paths can produce chunked columns at runtime where the type system would say packed. Better to widen the type and pay the materialize-per-method cost on the rare chunked path than to lie at the type level about what column(name) can return.

    Type Parameters

    • K extends "number" | "boolean" | "string" | "array"