Giorgi / DuckDB.NET / 35915328756

88%

push

github

Split the collection writer into list and array writers

ListVectorDataWriter handled both LIST and ARRAY columns, branching on
IsList in the constructor, the item positions, the size check, NULL
handling and growth. The two types store their items differently, and
both of the recent fixes came from mixing them: a list-only running
offset applied to arrays, and list growth that missed how DuckDB moves
array items.

- CollectionVectorDataWriter holds what both share: the item writer,
  the per-type switch that writes a collection's items from a start
  position, and Dispose.
- ListVectorDataWriter keeps the running offset, list entries and
  ResizeVector.
- ArrayVectorDataWriter has the size check, row-based positions and the
  WriteNull override.

The InitializeWriter override from 7dc9f6e now lives only in
ArrayVectorDataWriter, not in the list writer as that commit's message
says. When an outer list grows, DuckDB's Vector::FindResizeInfos walks
into ARRAY items but stops at LIST buffers, so only an array has to pass
the refresh on. A list's items move only when that list grows, and its
own ResizeVector refreshes them.

Behaviour is unchanged. The size check now runs once, before any item
is written, with the same message. New tests cover a growing list of
arrays of arrays (INTEGER[2][2][]) and of arrays of lists
(INTEGER[][2][]).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KxGp85FzaVDRPo7eXVBFrf

1580 of 1877 branches covered (84.18%)

Branch coverage included in aggregate %.

79 of 92 new or added lines in 4 files covered. (85.87%)

3184 of 3514 relevant lines covered (90.61%)

384257.78 hits per line