dataframe-parquet

Parquet reader and writer for the dataframe ecosystem.

Stackage Nightly 2026-07-26:	1.4.0.0
Latest on Hackage:	1.4.0.0

See all snapshots dataframe-parquet appears in

MIT licensed by Michael Chavinda

Maintained by [email protected]

This version can be pinned in stack with:dataframe-parquet-1.4.0.0@sha256:ed68df005313c152e9883136458af4314fc9edecf631755ba14b8c1dce9f9eb1,2900

Module documentation for 1.4.0.0

DataFrame
- DataFrame.IO
  - DataFrame.IO.Parquet
  - DataFrame.IO.Utils
    - DataFrame.IO.Utils.RandomAccess
- DataFrame.Typed
  - DataFrame.Typed.IO
    - DataFrame.Typed.IO.Parquet

Depends on 16 packages(full list with versions):

base, bytestring, containers, dataframe-core, dataframe-operations, dataframe-parsing, directory, filepath, Glob, pinch, snappy-hs, text, time, vector, zlib, zstd

Used by 3 packages in nightly-2026-07-26(full list with versions):

dataframe, dataframe-lazy, dataframe-parquet-th

DataFrame.IO.Parquet — pure-Haskell Parquet 2.0 reader and writer (with snappy, zstd, gzip codecs, dictionary decoding, nested list/repeated columns, and predicate pushdown). Heavy package — pulls in pinch and compression codecs. Reading directly from HuggingFace (hf://) datasets lives in the separate dataframe-huggingface package. Most users want dataframe-csv instead unless they specifically need Parquet.