[V3] Allow for incomplete codec metadata using numcodecs.get_codec

[jhamman]

This PR refactors how codecs are instantiated in the current V3 implementation. This allows for default parameters to be omitted.

TODO:

• Changes documented in docs/release.rst
• GitHub Actions have all passed
• Test coverage is 100% (Codecov passes)

closes #1443

[codecov]

Codecov Report

Merging #1447 (15bbf7f) into main (98f74d5) will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##              main     #1447   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           37        37           
  Lines        15019     15026    +7     
=========================================
+ Hits         15019     15026    +7     

Impacted Files	Coverage Δ
zarr/meta.py	100.00% <100.00%> (ø)
zarr/tests/test_meta.py	100.00% <100.00%> (ø)

[rabernat]

I am in favor of making the handling of codec configuration the same in V2 vs V3.

I do think this needs a test to verify that the situation described in cgohlke/tifffile#211 is handled correctly after this change.

[jhamman]

Test failures are unrelated and are fixed here #1450

[rabernat]

Great work! Thanks for adding the test.

Once you note this change in docs/release.rst and marge main to get the final test passing, I think we are good to go.

Any comments on this change from other devs? I'd be curious to hear @grlee77's take on relaxing the strict codec metadata expectations, as he is the one who wrote the original code here.

[jhamman]

@zarr-developers/python-core-devs - looking for some final feedback here. If I don't hear any, I'll plan to merge ~7/12.

[martindurant]

LGTM. I suppose if the conf has invalid values, you get an exception when the codec is instantiated - do we have any expectation/test of how that should go?

For the related issue in TIFFfile, I will comment that many codec options, specifically for compressors, are only actually important at compress time, not when reading. It would be useful to know, though, if the defaults in python implementations are compatible with whatever TIFF assumes - I don't know how to do that rigorously.

[cgohlke]

It would be useful to know, though, if the defaults in python implementations are compatible with whatever TIFF assumes - I don't know how to do that rigorously.

The only compression method specified in the TIFF standard that has a codec in the numcodecs package is "deflate". The spec says that "each image segment consists of a single complete zlib data stream.". This is compatible what numcodecs.Zlib handles.

All the other dozens of compression methods found in TIFF are either nowhere specified specifically for TIFF (e.g. LZMA, Zstd) , not supported by numcodecs (PackBits, LZW, CCITT, JPEG), or both (e.g., JPEG2000, WebP, PNG, JetRaw).

The imagecodecs package provides many of those codecs made for TIFF. Few of them might require setting extra arguments, such as JPEG (jpegtables), WebP (hasalpha), or JetRaw (shape). This is handled through tifffile. If not, open an issue there.

[rabernat]

@jhamman - we just need release notes for this. That's important, since this change could hypothetically impact existing code.

[martindurant]

@cgohlke , thank you for the thorough reply. It seems we have nothing to worry about here.