DDN-COLUMNAR

Typed, nullable, chunked column storage and value computation for D

Modules 24

  • ddn.columnar.array
    The chunk layer of `ddn.df`: the untyped, Arrow-style `ArrayData` plus the cheap typed views built on kernel entry (D1, D6, D8; ADR-2026-10-18-03, ADR-2026-10-18-06).
  • ddn.columnar.batch
    The structural storage carrier beneath `DataFrame`.
  • ddn.columnar.bitmap
    The bitmap layer of `ddn.df`: one representation for validity bitmaps and bit-packed boolean values (D7, ADR-2026-10-18-08).
  • ddn.columnar.builder
    The builder layer of `ddn.df`: mutable, single-use buffers that seal into immutable `ArrayData` chunks (D3; ADR-2026-10-18-09).
  • ddn.columnar.categorical
    The categorical-column layer of `ddn.df` (ADR-2026-10-23-13/-14): int32 codes in `values`, the labels chunk in `children[0]`. Nulls are validity-0 slots (no code); codes never compare directly - eq...
  • ddn.columnar.column
    The `Series` layer of `ddn.df`: a named, typed column of chunks (D11).
  • ddn.columnar.compute
    The compute layer of `ddn.df`: arithmetic; comparison, logic and aggregations follow in the later tasks of phase D (D12; ADR-2026-10-18-05).
  • ddn.columnar.convert
    The cast layer of `ddn.df`: converting chunks between dtypes (D12; ADR-2026-10-18-05).
  • ddn.columnar.decimal
    The decimal-column layer of `ddn.df` (ADR-2026-10-23-13/-14): exact fixed-point values stored as raw coefficients (int64 for precision <= 18, Int128 payload for 38), semantics from `ddn.num.fixed_d...
  • ddn.columnar.dispatch
    The dispatch layer of `ddn.df`: static expansion of a kernel over the fourteen kinds through one textual call site (D6, D6a, D14, D15; ADR-2026-10-18-10, ADR-2026-10-18-11).
  • ddn.columnar.exception
    The error type of `ddn.df`.
  • ddn.columnar.hash
    The hashing layer of `ddn.df`: per-row hashing, typed column equality and `Series.unique`/`nUnique` (D16; ADR-2026-10-18-15). The `GroupTable` itself lives in `ddn.df.groupby`, shaped by its two co...
  • ddn.columnar.list
    The list-column layer of `ddn.df` (ADR-2026-10-23-13/-14): offset-and-child storage where a null list (validity-0, empty span) is distinct from an empty list (valid, zero span), and child nulls liv...
  • ddn.columnar.memory
    The memory layer of `ddn.df`: aligned GC-owned buffers, the single trusted sealing step and the safe typed view (D2, D9, D13; ADR-2026-10-18-07).
  • ddn.columnar
    Typed, nullable, chunked column storage and value computation for D.
  • ddn.columnar.parallel
    The parallel-execution substrate of `ddn.df` (ADR-2026-10-19-02).
  • ddn.columnar.sort
    The sort layer of `ddn.df`: stable argument sorts over `(key, idx)` pairs with the index tie-break always applied (D17; ADR-2026-10-18-13).
  • ddn.columnar.strings
    The literal string namespace of `ddn.df` (ADR-2026-10-23-12): length, case mapping, stripping, slicing and literal matching over UTF-8 with *code-point indices* and full Unicode semantics via Phobi...
  • ddn.columnar.structs
    The struct-column layer of `ddn.df` (ADR-2026-10-23-13/-14): named field chunks in `children[i]` with parent validity marking whole-struct nulls, distinct from a valid struct of null fields. Field ...
  • ddn.columnar.temporal_access
    The temporal component accessors of `ddn.df` (ADR-2026-10-23-12): `year`/`month`/`dayOfMonth` on DATETIME (the UTC civil date of the instant - days-from-epoch by floor division, then the civil-from...
  • ddn.columnar.testkit
    The shared cross-chunking oracle harness of the DDN Analytics family.
  • ddn.columnar.types
    The logical types of `ddn.df`: `Kind`, `TimeUnit`, `DataType`, the physical and logical type maps, the supported element types and the arithmetic supertype (D4, D5; ADR-2026-10-18-03).
  • ddn.columnar.vector
    The packed-bit vector kernels of `ddn.df`: dense `core.simd` loops that turn comparisons into bitmap words and arithmetic into raw value buffers, so kernels never pay per-element `Nullable` or p...
  • ddn.columnar.window
    The window-function layer of `ddn.df` (ADR-2026-10-24-17): `over` (partitioned aggregation scattering back to input rows) and rolling windows (fixed-width receding frames). Both reuse the existing ...