Jerome Kelleher

OK, looks like this mostly just works. We're successfully compiling against numpy 2.0, and running tests against numpy 1.x. We can't really test against numpy 2.0 at runtime until there's...

Excellent catch --- yes, we should do our very best to be tidy here, even if the code ends up being a bit convoluted.

Time units should probably be a parameter somewhere, and not something we expect to be in the Zarr dataset? It's unlikely that ``sites_time`` or ``individuals_time`` is ever actually going be...

If there is more than one contig present in the VCF Zarr we require a ``contig`` parameter to be provided in the constructor. If a ``contig_length`` is present, we use...

Yeah, it seems that if there are multiple contigs in the VCF then it would be helpful to tell tsinfer which one it should use. You're right that this cuts...

Mixed_ploidy isn't a property of the dataset, and was introduced in sgkit as a way of forcing calls to appear as if they are *not* mixed ploidy (from the output...

That involves a full scan of the data, though, as you have to examine every element of the genotype matrix to find any -2s. That's definitely not an operation we...

Option 3 is straightforward enough all right.

Can't do it on create - it involves reading all values and can take 10s of minutes.

I don't think we're doing to support Xarray as an input format ultimately as it pulls in too much complexity. We should definitely provide better error messages though, as the...