bcftools corrupts duplicate GT format fields

[anthakki]

Running bcftools (here filter but seems to affect other commands as well) on a VCF with duplicate GT (genotype) FORMAT fields seems to change all but the last GT value. Looks like ./., 0/0, 0/1, 1/1 get converted to 0,0, 2,2, 2,4, and 4,4, respectively. I'm not 100% sure if duplicate GT values are legal, but I would expect an error instead of invalid data. Non-GT fields don't seem to have the problem. I'm using bcftools 1.19, but this can also be reproduced in bcftools 1.12.

Minimized test case follows. I would expect the payload to match that of the input.

$ cat foo.vcf 
##fileformat=VCFv4.1
##FORMAT=<ID=GT,Number=1,Type=String>
##FORMAT=<ID=X,Number=1,Type=Integer>
##contig=<ID=chr1>
#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO	FORMAT	A
chr1	1	.	A	C	.	.	.	GT:X:GT:X	0/1:9:0/1:9
$ bcftools filter foo.vcf | sed '/^#/d'
chr1	1	.	A	C	.	PASS	.	GT:X:GT:X	2,4:9:0/1:9

[pd3]

Duplicate tags are not allowed. I am not sure if it is explicitly stated in the VCF specification, but that was the intention.

The parsing is done in htslib, ideally it should give a warning and drop the duplicate fields. Obviously, easiest solution is to avoid producing invalid VCFs :)