RFC: layers should accept a zarrita Store directly for data loading

[keller-mark]

As noted in the title, I am making this issue as a "request for comment"-style thing to encourage discussion. It is just an idea, and the benefits may not outweigh the negatives, so i want to hear others' thoughts.

User story

Viv currently implements loader classes conforming to the PixelSource abstraction. Such an abstraction was necesssary for historical reasons to unify things between the different image formats Bioformats-Zarr, OME-TIFF, OME-Zarr.

However, requiring the intermediate PixelSource may be unnecessary if we can instead unify things at the Zarr-store-level.

Preferred solution

• Refactor the layers so that they accept a Zarrita store, where the root of the store corresponds to an OME-NGFF image.
• Implement stores that "virtually" interpret non-Zarr images as OME-NGFF images, so that Viv still supports the same range of image formats.
  • See Support OME-TIFF through custom OmeTiffStore vizarr#254 and https://github.com/keller-mark/tiff-as-virtual-zarr.js

Possible alternatives

The status quo, keeping the PixelSource abstraction.

[ilan-gold]

So if the layers see a zarrita store, they guess that it's an OME store? What of the metadata, like physical sizes which the layers sometimes need?

[keller-mark]

Correct. Well not guess, assume. They could use zod-ome-ngff to validate the metadata in the root group.attrs.
The physical size info from the OME-XML would be translated to the OME-NGFF metadata model.

[ilan-gold]

They could use zod-ome-ngff to validate the metadata in the root group.attrs

Yup, just meant there would be a check.

The physical size info from the OME-XML would be translated to the OME-NGFF metadata model.

Ah true, I guess our layers don't use anything else actually. The only one might be the interleaved RGB layer, but that is tiff data and not OME

[dahannes]

This would simplify things significantly for us. We're currently working around the sealed loading pipeline to support client-side SigV4 request signing for S3 — which required separate solutions for zarr (#944) and TIFF (#957).

Unifying at the store level would let us use a single CredentialedHTTPStore (zarrita-compatible, already implemented) for all formats. The format-specific loading code, metadata shape differences (OME-XML vs NGFF), and custom transport adapters we maintain today would all go away.

Happy to help test or provide feedback on an early implementation if that's useful.