Feature Request: First-class representation of nested attributes

[etienneschalk]

Is your feature request related to a problem?

Formats such as Zarr store group attributes in JSON (e.g. .zattrs), so attributes can be deeply nested (objects and arrays). The current xarray APIs and built-in representations (e.g. repr(), HTML repr in Jupyter) typically expose attributes in a flat or single-level way. That makes it hard to inspect or document complex metadata in the IDE or in notebooks. Users either see a flattened view, truncated output, or have to manually traverse raw attribute dicts instead of seeing the structure xarray actually has at hand.

Xarray current HTML representation (only top-level attributes are shown, with full content):

Describe the solution you'd like

First-class support for nested attributes in xarray’s APIs and/or representations, for example:

• Repr / HTML repr: Optionally render nested attributes as an expandable/collapsible tree (or at least with clear indentation/structure) instead of a single-level or flattened dump, so users can see the full hierarchy in Jupyter and other environments that rely on xarray’s built-in repr.

The goal is that opening a dataset with rich nested metadata (e.g. from Zarr) feels natural in xarray: the repr and APIs should reflect that nesting rather than flattening or hiding it.

Describe alternatives you've considered

• Manual traversal: Users can already walk ds.attrs (or group attrs) and pretty-print themselves. This doesn’t improve the default experience in Jupyter or the consistency of tooling.
• Downstream-only handling: Some tools (e.g. Scientific Data Viewer) can render nested attributes in a tree view on the viewer side. That helps when using those tools but doesn’t improve the experience for users who only use xarray’s built-in repr or other consumers that follow xarray’s representation.
• Leaving attributes as “opaque” dicts: Keeps implementation simple but under-serves users and formats (like Zarr) where nested metadata is the norm.

Example: Scientific Data Viewer nested representation of nested attributes with collapsible sections:

Additional context

• This would benefit anyone opening Zarr (or similar) datasets with rich .zattrs-style metadata in Jupyter or scripts.
• Downstream projects (e.g. Scientific Data Viewer) could align their UIs with xarray’s representation for consistency when the same dataset is opened in xarray and in other tools.
• Reference: Scientific Data Viewer – xarray and nested attributes (doc that motivated this request).