Create empty genotype arrays when there are no samples

[tomwhite]

Fixes #463

[jeromekelleher]

Ah, hmm, I think I misunderstood. I think we should continue to have no call_genotype arrays when there is no GT in the input VCF by default. However, we do allow for creating empty genotypes if you specify a ploidy value explicitly? Or is this only handling cases where GTs are specified in the header, but there's no samples?

It's not clear to me what happens when the ploidy value that you specify disagrees with what's in the GTs. It is simplest to just raise an error for now?

In any case I think we need to spell out the semantics somewhere in the documentation and it's probably best to that now while we're making the change.

[tomwhite]

I think we should continue to have no call_genotype arrays when there is no GT in the input VCF by default.

That's inconsistent with all the other genotype fields though. If there is a GQ field defined in the header but no samples then vcf2zarr will create empty call_GQ fields.

It's not clear to me what happens when the ploidy value that you specify disagrees with what's in the GTs. It is simplest to just raise an error for now?

Yes, that's probably best.

[jeromekelleher]

There's two orthogonal things here:

1. Is GT included in the header?
2. Are there samples present with GT data?

I think we have to deal with both? If there's no GT in the header and no samples present, we should not output call_genotype by default I think. But, we can force call_genotype to be included, by specifying the ploidy value.

[tomwhite]

I agree they are orthogonal.

Case	GT in header	samples with GT	create call_genotype
1	No	No	No
2	No	Yes	Error
3	Yes	No	Yes
4	Yes	Yes	Yes

Currently for case 3 we don't create call_genotype, but I'm arguing that we should, for consistency with other format fields.

I think specifying ploidy is also orthogonal. In case 1 it could force creating call_genotype (although it doesn't have to), and it's needed for case 3 in order to specify diploid for example (which I why I introduced it). For case 4 we could error if it disagrees with the data, as you suggested earlier.

What do you think?

[jeromekelleher]

SGTM. What do we do in case 2 at the moment?

[jeromekelleher]

(If we're already handling case 2 sensibly I think we should leave it - experience has shown that you just have to accept malformed VCFs)

[tomwhite]

Hopefully this is a bit better. For case 1 setting ploidy doesn't have any effect. I haven't changed case 2 (or checked what it does). I've implemented case 3. And for case 4 it will error if the set ploidy is less than the maximum ploidy in the data.