Error decoding v2 array

[jeromekelleher]

Zarr version

3.0.0a5

Numcodecs version

0.13.1

Python Version

3.10.12

Operating System

Linux

Installation

pip install

Description

Reading a v2 array written with Zarr Python 2 gives an error when creating the Blosc codec.

I've attached the .zarray and one chunk file below to reproduce.

import zarr

z = zarr.open("zarr_bug")
print(z)
print(z[:])

Gives

<Array file://zarr_bug shape=(204714,) dtype=int32>
Traceback (most recent call last):
  File "/home/jk/work/github/vcf-zarr-publication/zarr_bug.py", line 6, in <module>
    print(z[:])
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/core/array.py", line 919, in __getitem__
    return self.get_orthogonal_selection(pure_selection, fields=fields)
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/_compat.py", line 43, in inner_f
    return f(*args, **kwargs)
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/core/array.py", line 1361, in get_orthogonal_selection
    return sync(
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/core/sync.py", line 91, in sync
    raise return_result
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/core/sync.py", line 50, in _runner
    return await coro
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/core/array.py", line 476, in _get_selection
    await self.codec_pipeline.read(
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/codecs/pipeline.py", line 427, in read
    await concurrent_map(
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/core/common.py", line 53, in concurrent_map
    return await asyncio.gather(*[func(*item) for item in items])
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/codecs/pipeline.py", line 260, in read_batch
    chunk_array_batch = await self.decode_batch(
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/codecs/pipeline.py", line 177, in decode_batch
    chunk_array_batch = await ab_codec.decode(
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/abc/codec.py", line 125, in decode
    return await _batching_helper(self._decode_single, chunks_and_specs)
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/abc/codec.py", line 409, in _batching_helper
    return await concurrent_map(
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/core/common.py", line 53, in concurrent_map
    return await asyncio.gather(*[func(*item) for item in items])
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/abc/codec.py", line 422, in wrap
    return await func(chunk, chunk_spec)
  File "/home/jk/.local/lib/python3.10/site-packages/zarr/codecs/_v2.py", line 30, in _decode_single
    compressor = numcodecs.get_codec(self.compressor)
  File "/home/jk/.local/lib/python3.10/site-packages/numcodecs/registry.py", line 42, in get_codec
    config = dict(config)
TypeError: 'Blosc' object is not iterable

Steps to reproduce

Get the attached tar file (renamed to .txt to work around annoying file type restrictions...)

zarr_bug.txt

• mv zarr_bug.txt zarr_bug.tar
• tar -xf zarr_bug.tar
• Run code above

I've included one chunk here and the .zarray as hopefully the simplest way to reproduce.

Additional output

No response

[jeromekelleher]

Including the .zarray here for ease of reference:

{
    "chunks": [
        10000
    ],
    "compressor": {
        "blocksize": 0,
        "clevel": 7,
        "cname": "zstd",
        "id": "blosc",
        "shuffle": 0
    },
    "dimension_separator": "/",
    "dtype": "<i4",
    "fill_value": -1,
    "filters": null,
    "order": "C",
    "shape": [
        204714
    ],
    "zarr_format": 2
}

[d-v-b]

thanks for reporting this bug. looking at the last few lines of the traceback:

  File "/home/jk/.local/lib/python3.10/site-packages/zarr/codecs/_v2.py", line 30, in _decode_single
    compressor = numcodecs.get_codec(self.compressor)
  File "/home/jk/.local/lib/python3.10/site-packages/numcodecs/registry.py", line 42, in get_codec
    config = dict(config)

I'm wondering if the problem comes form calling get_codec(self.compressor) instead of get_codec(make_this_a_dict(self.compressor)).

[jhamman]

Thanks for the great report @jeromekelleher - I'm not able to reproduce this with 3.0.0-rc.1 so I'm closing.