Zarr protocol v3 2019-12 to 2020-12

[clbarnes]

Questions (in spec)

• node name character set (currently ASCII, potentially UTF-8)
• node name case sensitivity (currently sensitive, potentially insensitive)

Misc

• Paths start with /
• Zero-D arrays
• Groups' metadata document is optional
• Node names are case-sensitive
• No complex numbers
• Raw bits dtype
• Chunk IDs are prefixed with c, and default dimension separator is / (i.e."c1/2/3" not "1.2.3")
  • Zero-D array chunk ID is "c"

Metadata

Entry point metadata (formerly "bootstrap metadata")

• zarr.json (entry point metadata document in hierarchy root, formerly "bootstrap metadata") is the only definite JSON
• metadata_encoding is now a URL, not a MIME type (e.g. "https://purl.org/zarr/spec/protocol/core/3.0" instead of "application/json")
• metadata_key_suffix is new, e.g. ".json"

Array metadata

"{node_name}.array{metadata_key_suffix}"

• dtype -> data_type
• chunk_grid object now contains separator = "/"
• compressor is optional
• (some stuff about extensions)

Group metadata

"{node_name}.group{metadata_key_suffix}"

• no extensions

Stores

• "prefix" operations are group names with a trailing /

Writable:

• delete is now erase
• erase_prefix now exists