Changelog

This changelog covers every package in this repository. Version headings follow Zarr.jl releases.

Unreleased

• consolidate_metadata(g) of a Zarr v3 group now writes the consolidated metadata into the group's zarr.json. Previously it wrote a v2 .zmetadata file.

• consolidate_metadata(store, "", ZarrFormat(3)) now includes the children of the store root. Previously their paths started with /, so they were not found.

v0.11.2 - 2026-10-16

• Zarr v3 arrays that zarr-python wrote with a zlib compressor, as the numcodecs.zlib codec, can now be read, and ZarrZlib.ZlibV3Codec writes that codec. #360

• zarr_format is now supported for ZGroup, returning the group's Zarr format. #359

• ZarrCore is bumped to 0.11.2 to release the zarr_format(::ZGroup) support. #362

• ZarrZlib is bumped to 0.11.1 to release ZlibV3Codec and numcodecs.zlib support. #362

v0.11.1 - 2026-10-07

• Zarr v3 arrays can now store dimension names. Pass dimension_names=("x", "y") to zcreate and read them back with dimension_names(z), so arrays written by Zarr.jl open with named dimensions in xarray. Give names in Julia's dimension order, and use nothing for an unnamed dimension. #345, fixes #319

• resize! and append! now throw an error on read-only arrays instead of rewriting the stored metadata and deleting chunks. Resizing through a consolidated store now fails without changing the array, for both Zarr v2 and v3. #342

• Integer arrays using DeltaFilter now round-trip correctly when values wrap around the type's limits, such as Int8[-128, 127, -128]. #336, fixes #335

• Zarr v2 arrays that use the shuffle or fletcher32 filter with element types wider than one byte can now be read. Previously they could be written, but reading them threw a BoundsError. #354

• ZarrHTTP can now serve a DirectoryStore from its root path. It also rejects requests for files outside the served directory, including .. paths and symlinks that point elsewhere, with a 400 response that explains why. #344

• ZarrCore is bumped to 0.11.1 to release the ZarrCore changes above. #356

v0.11.0 - 2026-10-06

Zarr.jl is now split into several smaller packages. using Zarr still loads all of them, so most code needs no changes.

• The core array, group, metadata, and storage code now lives in the new lightweight ZarrCore package. #317

• The HTTP, GCS, S3, and ZIP storage backends moved into ZarrHTTP, ZarrGCS, ZarrS3, and ZarrZip. AWSS3 is still an optional dependency of ZarrS3. #330

• Blosc, zlib, and Zstandard support moved into ZarrBlosc, ZarrZlib, and ZarrZstd. Each package provides both the Zarr v2 compressor and the Zarr v3 codec. Blosc is the default compressor when ZarrBlosc is loaded; with ZarrCore alone, data is stored uncompressed by default. #330

• Compressors, codecs, and URL-based stores register themselves when their package loads. If several URL patterns match, the most specific one is used. To turn off automatic registration, set RegisterAtInit = false for ZarrCore in LocalPreferences.toml and call each package's register!() yourself. #330

• Zarr.jl now declares an explicit public API. Exported names are unchanged. Store, codec, filter, and compressor types, documented extension points, missing_chunk_return_code!, and gcs_credentials are public. Internal names such as Metadata and ZarrFormat are no longer available as Zarr.x; use Zarr.ZarrCore.x instead. DateTime64, PermanentZarrCache, BloscCodec, and GzipCodec are no longer accessible through Zarr. #317, #330

• Fixed reading Zarr v3 arrays opened with fill_as_missing=true, which threw cannot reinterpret UInt8 as Union{Missing,Float64}. #333

• For contributors: each package now lives in its own top-level directory, and the root project is a Julia workspace. #339

v0.10.2 - 2026-08-19

• Support HTTP.jl 2.x and drop 1.x #322

• Add logo and favicon to docs #307

• Support reading and writing variable-length strings #311

• Add more informative show method for S3Store #320

• Fix complex fill values in python interop tests #316

• Remove lru keyword from zopen, should use DiskArrays.cache instead

v0.10.1 - 2026-07-07

• Add caching to menu, update sharding description #306

• Added support for the Zarr V3 fixed_length_utf32 string data type #305

• Added consolidated_metadata reading and writing support for v3 #287

• Fix the ConcurrentRead write path silently dropping chunk writes; affected stores such as S3Store wrote array metadata but no data chunks #297

• Added reading compat for stores produced in python with numcodecs.blosc #286

• Bump HTTP compat to 2 #284

• V2 performance improvements #280

  • readblock! fast path for single-chunk full-reads that bypasses the readtask channel and the chunk-shaped scratch buffer, decoding straight into the output array

  • writeblock! fast path for single-chunk full-overwrites that bypasses the readtask/writetask channels and the chunk-shaped scratch buffer, encoding straight from the input array into the store

  • Zero-copy chunk write for NoCompressor + no filters: the single-chunk fastpath hands a reinterpret(UInt8, ain) view straight to the store, skipping the chunk-sized Vector{UInt8} allocation + memcpy

  • NoCompressor writes: replace append! over the reinterpret view with bulk resize! + copyto! in the generic zcompress! fallback

  • NoCompressor reads: bulk-copy zuncompress! method dispatched on ::NoCompressor bypasses copyto!(::Array, ::ReinterpretArray)'s element-by-element walk

  • getchunkarray_undef skips the dead zero-fill of the chunk-shaped scratch buffer on full-overwrite paths

• Added manual pagination in order to go beyond the default 1k #282

• Added wait to writetask in writeblock! #281

• Fix getattrs for v3 #277

• Add CachingStore, a store that caches reads from a remote store in a local cache store #231

• Fix CondaPkg branch in CI, use release version instead #273

• Fix creation of on-disk arrays that do not fit in memory #269

• Add another consistent caching approach through zarrcache #299

v0.10.0 - 2026-04-24

• Enable sharding_indexed codec for Zarr v3 #241

  • outer chunks (shards) are now split into inner chunks with a byte-range index, enabling efficient partial reads of large arrays

  • Python zarr interoperability tests for sharded arrays using real fixtures generated via PythonCall/zarr-python

  • Function-based codec registration system with typed CodecEntry for extensible V3 codec parsing

  • sharding_indexed codec context propagation: inner codec parser now receives element-size context so codecs like blosc get the correct typesize

  • sharding_indexed decode double-shift bug for index_location=:start: shard index offsets are now treated as absolute byte offsets from the shard start in all cases

  • sharding_indexed decode fill value: zdecode! now accepts an optional fill_value argument so missing inner chunks are filled with the array fill value instead of zero(T)

  • pipeline_decode! for V2Pipeline now accepts the fill_value keyword argument (ignored, but required to match the unified call site in ZArray.readblock!)

  • Ragged inner chunks in ShardingCodec handled correctly

• fixes UI str parsing #259

• added zarr_format to ZGroup #258

• Add support for S3Path and remove deprecated global_aws_config() #253

• (docs) get started #246

• update badges #245

• setup vitepress #243

• Registration system for chunk key encoding #242

• test: Consolidate CondaPkg.add calls #240

• test: Read Julia-generated v3 fixtures with Python zarr #239

• Use CondaPkg.jl#206 for nightly, fix manifest_uuid_path #238

• fix docs build #237

• consolidate methods #235

• Add AbstractChunkKeyEncoding #234

• V3 codec pipeline, group creation, and Python interop #232

• Add read support for big endian #230

• Indent method/type signatures on docstrings #229

• Promote header levels in tutorial.md #228

• Continue on v3 PR #226

• move AWSS3 into extension #224