Closed rsignell closed 1 week ago
Since there is still no in-spec way to have non-uniform chunks in Zarr V3, this is not possible today. But it should be trivial to support in Icechunk.
Edit: FWIW, an alternate way to do this would be to support lazy array concatenation. Like, concatenating virtual arrays of different shape into one big virtual array.
Thanks Ryan! Indeed we were planning to concat, but just thought it would be cool to harness the power of Icechunk! Someday soon it sounds like! 🤓
We have some yearly files that were rechunked with
{'time':-1}
that resulted in datasets with chunk sizes of 365,366,365 in time due to leap year. With Zarr v2 this could not be represented as a virtual dataset, but with icechunk and v3, should it be expected to work?