issue_comments

Agree! Now we just need to decide between chunksizes and chunk_sizes...

It seems better to introduce a new property on both DataArrays and Datasets that always returns a dict

That's a good suggestion - then we can have backwards compatibility whilst also allowing intuitive code that treats dataarrays and datasets similarly, e.g:

python def is_core_dim_chunked(obj, core_dim): return len(obj.chunksizes[core_dim]) > 1

chunksizes seems too similar to chunks

I think chunksizes is quite good: it is in keeping with sizes, and auto-complete would also show both chunks and chunksizes when a user types .ch[tab] which I think is helpful.

For DataArrays there is an underlying chunks property so it makes sense to forward it (like shape and dtype). Though perhaps we should only forward those properties that are common to all duck arrays.

It seems better to introduce a new property on both DataArrays and Datasets that always returns a dict (Like sizes vs shape). I came up with two names but don't like either of them: chunksizes seems too similar to chunks; dims_chunks doesn't really seem great either.

There is a similar problem for dtype as @crusaderky points out here

The honest answer is that I didn't think too carefully about this when originally implementing Xarray's Dask wrapper back in 2015.

I guessed that might be the case!

I'm not sure whether making this consistent is worth the effort of a significant breaking change though

Still leaves this question though ^ . I made a draft PR in https://github.com/pydata/xarray/pull/5846.

The honest answer is that I didn't think too carefully about this when originally implementing Xarray's Dask wrapper back in 2015.

DataArray.chunks forwards to chunks on Dask arrays (a tuple), but that didn't make sense for Dataset.chunks due to the lack of a dimension ordering.

Variable.chunks also returns a tuple, which again I feel is weird given that variables have named dimensions.

There is another difference between ds.chunks and da.chunks - the former checks for inconsistent chunking between different variables when called (and will raise ValueError Object has inconsistent chunks along dimension {dim}. This can be fixed by calling unify_chunks()."). In contrast da.chunks doesn't check, and so it's possible to have a DataArray whose data variable is chunked inconsistently with its coordinate variables and not be warned about it.