Possible bug in htslib/bcftools 1.1: [E::bcf_hdr_add_sample_len] Duplicated sample name

[sahwa]

If we have a test.vcf and use htslib/bcftools 1.1 to:

bcftools view test.vcf

We get

[E::bcf_hdr_add_sample_len] Duplicated sample name .... Failed to read from test.vcf: could not parse header

We get the same error for running any bcftools command.

But using an old version bcftools (1.9-207-g2299ab6 Using htslib 1.9-271-g6738132), then the file can be viewed OK.

[sahwa]

Edit - just figured that this is because one of the sample names contains an unusual special character "B�R1" - once I edited the name to remove the vcf could be parsed OK.

[daviesrob]

This is due to the way HTSlib looks for tabs in the header line, which currently (on x86-64) mistakes UTF-8 characters for the end of the sample name. As a result is splits the name "B�R1" into several parts, and complains about a duplicate when it gets to "R1". Casting *q to uint8_t makes the comparison unsigned and keeps the entire name intact.

The VCF4.3 specification says it uses UTF-8 so presumably names like this ought to be allowed?

[jmarshall]

The VCF4.3 specification says it uses UTF-8 so presumably names like this ought to be allowed?

The VCF spec is (as usual) vague on this. samtools/hts-specs#18 motivated UTF-8 by saying “in order to address the need to represent non-ASCII characters in INFO field values, VCF files are assumed to be encoded in UTF-8 […]” which I read as intending UTF-8 for use primarily in free text description fields and the like, as in SAM, but not necessarily in fields like these sample IDs that tools need to compare/etc. PR samtools/hts-specs#414 proposes rules for VCF sample IDs but remains vague. In a different context, ga4gh/seqcol-spec#2 (comment) wisely noted that allowing arbitrary Unicode would make testing sample IDs for equivalence (e.g. bcftools view -s) difficult.

Fortunately this HTSlib parsing code could be fixed to trigger only on \t and \n while remaining agnostic as to whether UTF-8 is actually allowed here.