idea: referencefilesystem as a way to prepend to zarr

[martindurant]

This is not really a kerchunk process, but if you wanted to extend a zarr dataset by prepending chunks to it, you would currently write the new files and be forced to rename all the other chunks, as well as update the metadata - which would change how it looked to other users.

However, with our system, the new could files could be written with different names (or even "-1" negative names) and then construct a new view containing the new prepended chunks without changing the original dataset at all.

cc @jakirkham

[jakirkham]

Thanks Martin! 🙏

Also was thinking about issue ( zarr-developers/zarr-specs#82 ) when this came up. Maybe it is worth exploring this problem more generally in that context

[martindurant]

Yes, exactly! Lots of possibilities here, but it's not exactly simple :)