BCFtools sort zero position overflow

[Rapsssito]

There is an overflow issue when sorting a VCF file with zero positions. Here is an example:

Unsorted VCF

MT	0	1915430843:1	N	]MT:16640]N	99	PASS	.
MT	0	3110098432:1	N	]MT:16638]N	99	PASS	.
MT	16638	3110098432:2	G	G[MT:0[	99	PASS	.
MT	16640	1915430843:2	G	G[MT:0[	99	PASS	.

Commands

bcftools sort test.vcf -o test.sorted.vcf

Sorted VCF

MT	4294967296	3110098432:1	N	]MT:16638]N	99	PASS	.
MT	4294967296	1915430843:1	N	]MT:16640]N	99	PASS	.
MT	16638	3110098432:2	G	G[MT:0[	99	PASS	.
MT	16640	1915430843:2	G	G[MT:0[	99	PASS	.

Note how the 0 from 3110098432:1 has been wrongly changed to 4294967296.

[Rapsssito]

@pd3, could you check if you are able to reproduce this issue?

[jkbonfield]

It shouldn't go wrong like this ideally, but note the input is invalid. VCF is 1-based, not 0-based. The zeros have been converted to -1 (as internally the code is using 0-based coordinates) which has caused a wrap-around.

[Rapsssito]

@jkbonfield, actually, the VCF specification allows positions with 0. This is an example from VCFv4.2:

[pd3]

This problem is not specific to bcftools sort but to a round trip through BCF via htslib. The format stores 0-based position in hts_pos_t type and 0 coordinate overflows. This needs to be fixed in htslib.

[Rapsssito]

@pd3, I have created the corresponding issue in htslib: samtools/htslib#1475

[daviesrob]

Fixed by samtools/htslib#1476