Apache Arrow nanoarrow 0.9.0 Release

Published 14 Aug 2026
By The Apache Arrow PMC (pmc)

The Apache Arrow team is pleased to announce the 0.9.0 release of Apache Arrow nanoarrow. This release consists of 38 resolved GitHub issues from 5 contributors.

Release Highlights

In addition to a number of bugfixes and minor build system improvements, we added several new features in nanoarrow 0.9.0.

  • Dictionary decoding support in IPC reader
  • Reference-counted array/buffer support
  • LZ4 decompression support in R and Python bindings

See the Changelog for a detailed list of contributions to this release.

Features

Dictionary decode support

Whereas the nanoarrow IPC reader suppports most Arrow IPC features, dictionary support was a long requested gap in the reader functionality (mostly requested by users of the DuckDB nanoarrow extension, which uses nanoarrow's reader). Dictionary encoding is used to reduce the size of frequently repeated values and is the serialized equivalent of the "dictionary" data type that is exposed in most Arrow implementations.

In nanoarrow 0.9.0 built with the IPC feature enabled, streams that include the most common forms of dictionary encoding (i.e., dictionary replacement) should now work out of the box. This includes nested/complex dictionary types and dictionary replacement but does not include "delta" dictionaries (i.e., dictionaires that grow larger as more values are encountered in the encoded values).

In R this is accessible via read_nanoarrow(); in Python this is accessible via nanoarrow.ArrayStream.from_readable(); in C this is available via the higher level ArrowIpcArrayStreamReader API. Lower level users of the ArrowIpcDecoder will have to update existing usage to use the ArrowIpcDecoder...WithDictionaries() variants of some functions to support input with dictionary schemas or batches.

Reference-counted array/buffer support

In previous versions (since the introduction of the IPC reader), the ArrowIpcSharedBuffer has supported reading IPC streams and sharing an underlying set of data buffers for a group of arrays; however, this wasn't quite sufficient for the more complex case of decoding a dictionary and attaching cheaply-cloned shared "values" arrays for potentially many batches. Version 0.9.0 moves this functionality to the ArrowSharedBuffer and expands it to support moving all of an array's buffers into a shared state that can be more cheaply cloned.

LZ4 decompression support in R and Python

While LZ4 decompression support was has long been available via the pluggable decoder framework (and available since 0.8.0 as a built-in compile time option), reading IPC streams with LZ4 buffer compression was not possible in the R or Python bindings. In 0.9.0, the requisite configuration options were added such that the packages are built with LZ4 when it is available on the system.

import io
import nanoarrow as na
import pyarrow as pa

buf = io.BytesIO()
batch = pa.record_batch({"x": range(1000)})
with pa.ipc.new_stream(buf, batch.schema, options=pa.ipc.IpcWriteOptions(compression="lz4")) as w:
    w.write_batch(batch)

buf.seek(0)
na.ArrayStream.from_readable(buf).read_all()
# nanoarrow.Array<non-nullable struct<x: int64>>[1000]
# {'x': 0}
# {'x': 1}
# {'x': 2}
# {'x': 3}
# {'x': 4}
# {'x': 5}
# {'x': 6}
# {'x': 7}
# {'x': 8}
# {'x': 9}
# ...and 990 more items

library(nanoarrow)
library(reticulate)

# IPC Write with compression not available in arrow/R
pa <- reticulate::import("pyarrow")
io <- reticulate::import("io")

buf <- io$BytesIO()
batch <- arrow::record_batch(x = 1:1000)
writer <- pa$ipc$new_stream(buf, batch$schema, options = pa$ipc$IpcWriteOptions(compression = "lz4"))
writer$write_batch(batch)
writer$close()

nanoarrow::read_nanoarrow(as.raw(buf$getvalue())) |>
  tibble::as_tibble()
#> # A tibble: 1,000 × 1
#>        x
#>    <int>
#>  1     1
#>  2     2
#>  3     3
#>  4     4
#>  5     5
#>  6     6
#>  7     7
#>  8     8
#>  9     9
#> 10    10
#> # ℹ 990 more rows

Contributors

This release consists of contributions from 5 contributors in addition to the invaluable advice and support of the Apache Arrow community.

$ git shortlog -sn apache-arrow-nanoarrow-0.9.0.dev..apache-arrow-nanoarrow-0.9.0
    36  Dewey Dunnington
     2  Andrew Kane
     2  Bryce Mecum
     1  Michael Osipov
     1  Oliver Borchert