Use of int8 for call_genotype results in integer overflow with complex variants

[timothymillar]

I'm working with some microhaplotype variant calls in which the number of the alleles can be >300 (likely due to poor quality calls at some loci). It would be ideal if the call_genotype dtype was configurable and/or automatically set based on the max_alt_alleles parameter. Similar discussion in #584.

[jeromekelleher]

+1 - this is definitely an issue. Setting based on max_alt_alleles is good, and I think that seals the deal on mapping any alleles we can't represent to missing data.

[tomwhite]

It would be ideal if the call_genotype dtype was configurable and/or automatically set based on the max_alt_alleles parameter.

+1

BTW I just opened #643, which is tangentially related. @timothymillar I wonder if you see a few very long alleles in your data?

[timothymillar]

I wonder if you see a few very long alleles in your data

I'm currently working with alleles with lengths 120bp. These are fixed sized "chunks" across the genome (targeted sequencing) so fixed length strings are suitable. But I think we'll use more variable allele lengths in future as these chunks need some tuning.
We also use freebayes quite regularly which can produce highly variable allele lengths. However, most of the data I'm working with is targeted sequencing so the datasets are quite small in the variants dimension.

[tomwhite]

Good to know - thanks.