fp-columnar

Columnar storage primitives (Column, ValidityMask) for pandas-parity dataframes — the storage layer behind fp-frame.

Part of the frankenpandas workspace.

When to depend on fp-columnar directly

Most users should depend on the umbrella frankenpandas crate instead. Depend on fp-columnar directly when:

• Embedding frankenpandas-compatible columnar storage in a higher crate without the DataFrame layer.
• Implementing a new IO format that serializes columns but routes through something other than fp-io.

Key types

• Column — typed array of Scalar values with associated ValidityMask. Int64 / Float64 / Utf8 / Bool / Timedelta / Datetime specializations. from_values infers dtype from input.
• ValidityMask — 1-bit-per-element missing-value bitmap. Saves 8x memory vs pandas's 8-byte nullable dtype.
• Dtype promotion rules mirror pandas's upcast cascade (Int32 -> Int64 -> Float64 -> Object/Utf8 under arithmetic).

Status

Stable. #![forbid(unsafe_code)]. Tested via fp-frame's test suite and the conformance packets.

Links

• Workspace README
• CHANGELOG