Jerome Kelleher

This is progressing as-of #2303 but still needs some work to complete. Some todos: - [ ] There are redundant calls to lineage_reset_segments, probably mainly in the GC code path....

Notes on the checklist above - the first task has been shelved for issue #2386. We should do some documentation on the second part all right before closing this issue.

I'm going to merge this as it should be significant improvement in terms of code readability, and I think it also fixes a few memory leaks and performance issues. I'll...

Overall the write path is very unclear to me also. The documentation needs to be clearer on what's xarray's responsibility, and what is coming from sgkit. However, this would need...

There's only two possible metrics right now ('euclidean' or 'correlation') so I vote we make the code less clever and just code in the function names directly for those two...

Ah, I see. I'm reluctant to remove the code as we put quite a lot of effort in and it's our main usage of GPUs... Perhaps @aktech would like to...

Thanks for the bug report @percyfal

Should that be an option in vcf2zarr @percyfal?

It's a common thing to have loads of contigs in the header with no variants. We could keep track of the number of variants for each contig during explode, and...

Requiring a single chunk in the samples dimension is a bit restrictive - perhaps we could add a note on how that could be generalised?