Results 506 comments of Tom White

Thanks for giving an example @jeromekelleher. I think your interpretation is correct. The NS values are simply missing, not the end of a vector that needs padding/filling.

> Sigh - skipping those tests pushes the required coverage below 100% so the build still fails. > > Any suggestions here? Temporarily push coverage down, and create an issue...

As a proof of concept I did this for Modal using the following code: ```python import fsspec import modal from sgkit.io.vcf.utils import ( build_url, url_filename, ) from sgkit.io.vcf.vcf_partition import partition_into_regions...

Note that `concat_zarrs` could be avoided completely if we had #1131

That would be great! The code needs a bit more attention before I'd be happy for people to copy it.

Yes, it is guaranteed. The format is documented here: https://github.com/pystatgen/vcf-zarr-spec/blob/main/vcf_zarr_spec.md, where it says: > Note that the `REF` and `ALT` fields are combined into a single Zarr array.

Very true! Are you interested in contributing a patch to fix these issues?

> In the current implementation, it is expected that regularization has happened before the method is called. Did I overlook an implementation of the regularization? No, [I skipped the regression...

#1125 and linked issues may be related too.

It fails when `asv` tries to create a conda env. I can't reproduce locally, so a bit of a mystery.