Beyond Python

The reference files currently stored in JSON format and can be readily interpreted in any language, as documented in detail in the References specification. The contents of each key is either encoded binary data or a URL/offset/size set. It can be loaded as long as the language being used can access the particular URL type.

To interpret the blocks as parts of a zarr dataset, the language should have an implementation of zarr, as well as whichever binary codecs the target requires (maybe nothing for plain binary, or common compressors like gzip, but might be more specific). You would need to write some code to expose the reference set as a storage object that zarr can use.

One example of a reference dataset used via a JS implementation, applied to multi-scale TIFF microscopy, can be found at https://observablehq.com/@manzt/ome-tiff-as-filesystemreference.