Prefixed genotype phasing

[pd3]

As reported in bcftools samtools/bcftools#2422 (comment), there is a problem with the new prefixing of partially phased genotypes, likely introduced by b14fffb4

I believe the intention was to explicitly prefix half-phased genotypes such as /0/1|2. The prefix is wrong in /0|1 (if anything, it should be |0|1, but no, it should not be prefixed in this case).

A copy-and-paste test case to reproduce the problem

printf '##fileformat=VCFv4.5\n##FORMAT=<ID=GT,Number=1,Type=String,Description="">\n##contig=<ID=chr6>\n#CHROM\tPOS\tID\tREF\tALT\tQUAL\tFILTER\tINFO\tFORMAT\taaa\nchr6\t160136611\t.\tC\tT\t.\t.\t.\tGT\t0|0\n' | bcftools view -o rmme1.bcf --write-index

printf '##fileformat=VCFv4.3\n##FORMAT=<ID=GT,Number=1,Type=String,Description="">\n##contig=<ID=chr6>\n#CHROM\tPOS\tID\tREF\tALT\tQUAL\tFILTER\tINFO\tFORMAT\tbbb\nchr6\t160136611\t.\tC\tT\t.\t.\t.\tGT\t0|1\n' | bcftools view -o rmme2.bcf --write-index

bcftools merge rmme1.bcf rmme2.bcf

[daviesrob]

This is happening because the two files specify different VCF versions, and the VCF parser attempting to be backwards compatible with itself by not setting the is_phased bit on the first GT value if the file format is 4.3 or earlier. Really we want to store one of three values - phased, unphased, or not set - but we can't do that in a single bit. The consequence is that you need to take the format version into account when interpreting the data.

If you change your example files so that they're both 4.3 or both 4.5 then the merge works as expected and outputs 0|1. It's the mixture that causes the problem.

[pd3]

I understand the versioning. However, I believe this is a different topic: the genotype /0|1 makes no sense, same as /0|1|2|3 wouldn't. Since N-1 out of N alleles are phased, the last one must be phased as well. If anything, it should be |0|1. However, it really should be just 0|1.