Disabling compression in version 1.2.0 (Zarr 3)

[orliac]

Hi,
In version 1.2.0 (Zarr 3), how can I disable compression when converting a MSv2 dataset with

xradio/src/xradio/measurement_set/convert_msv2_to_processing_set.py

Line 56 in f40eedb

def convert_msv2_to_processing_set(

?
With former version it used to worker passing None but with the latest version it fails with:

  File "/.../.../lib/python3.11/site-packages/zarr/registry.py", line 214, in _parse_bytes_bytes_codec
    raise TypeError(f"Expected a BytesBytesCodec. Got {type(data)} instead.")
TypeError: Expected a BytesBytesCodec. Got <class 'tuple'> instead.

[FedeMPouzols]

It looks like the issue is that:

• With Zarr v2 we used to pass the value directly to the Zarr encoding dictionary, keyword "compressor" (only one value). If one passes None it works.
• With Zarr v3 the keyword "compressor" has been replaced by "compressors" which takes a tuple/iterable. Zarr does not seem to check if the tuple juts has a None compressor and tries to use it directly as a BytesBytesCodec. And when building that tuple in xradio/ms/utils/zarr/encoding the case where the compressor is None was not considered. We'd need to support the use of None/nullable with something like:

diff --git a/src/xradio/measurement_set/_utils/_zarr/encoding.py b/src/xradio/measurement_set/_utils/_zarr/encoding.py
index c22a173c..d90a0922 100644
--- a/src/xradio/measurement_set/_utils/_zarr/encoding.py
+++ b/src/xradio/measurement_set/_utils/_zarr/encoding.py
@@ -6,4 +6,5 @@ def add_encoding(xds, compressor, chunks=None):
 
     for da_name in list(xds.data_vars):
         da_chunks = [chunks[dim_name] for dim_name in xds[da_name].sizes]
-        xds[da_name].encoding = {"compressors": (compressor,), "chunks": da_chunks}
+        compressors = (compressor,) if compressor else ()
+        xds[da_name].encoding = {"compressors": compressors, "chunks": da_chunks}

This seems to work, and I could open a PR to get this for the next release.

However the error that I get if I pass None to the compressor of the convert function is:

... lib/python3.13/site-packages/zarr/registry.py", line 212, in _parse_bytes_bytes_codec
TypeError: Expected a BytesBytesCodec. Got <class 'NoneType'> instead.

which is what I would expect (using Zarr 3.2.1).

Regardless of this, why would one disable compression? Is it because the overhead is too high in some cases?
An alternative is using a fast low-overhead codec config like for example: zarr.codecs.BloscCodec(cname='lz4', clevel=1), or even with no compression, along the lines of: zarr.codecs.BloscCodec(cname="lz4", clevel=0, shuffle=zarr.codecs.BloscShuffle.noshuffle)
Incidentally I couldn't find any truly "no-op" compressor in the new Zarr v3 codecs hierarchy.

[orliac]

Thanks for the feedback @FedeMPouzols! I'll try your suggestions.
As part of a project I have to compare reading from uncompressed data and reading + uncompressing compressed data with different compressors, hence the need to be able to disable the compression completely when generation the Zarr files. And same as you, I couldn't find a no-op compressor just to satisfy the expected type.