norm does not left-align <DUP> that has INFO/END but no INFO/SVLEN #2601

🤖 Written by Claude

bcftools norm left-aligns a symbolic <DEL> whether its length comes from INFO/SVLEN or INFO/END, and a <DUP> when INFO/SVLEN is present. A <DUP> described only by INFO/END, though, is left where it is. Follow-up to #2145.

Using the norm.symbolic.3.fa test reference (ACTCACAAAGGGATTACAGATTACAGATTACAGGGGATCCAGAT):

#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO
1	26	.	G	GATTACAG	.	PASS	SVTYPE=DUP
1	26	.	G	<DUP>	.	PASS	SVTYPE=DUP;END=33
1	26	.	G	<DUP>	.	PASS	SVTYPE=DUP;SVLEN=7;END=33

bcftools norm -f norm.symbolic.3.fa (1.24, and develop at 347c273):

1	11	.	G	GGATTACA	.	PASS	SVTYPE=DUP
1	11	.	G	<DUP>	.	PASS	SVTYPE=DUP;SVLEN=7;END=18
1	26	.	G	<DUP>	.	PASS	SVTYPE=DUP;END=33

The explicit allele and the SVLEN <DUP> move to position 11, but the END-only <DUP> stays at 26. The same happens on real data (GRCh37 NC_000003.11:128200306, SVTYPE=DUP;END=128201507), whereas the equivalent <DEL> with only END is realigned.

In realign(), a <DUP> takes its length only from INFO/SVLEN; without it sv_len is 0 and the record returns ERR_SYMBOLIC unchanged. VCF 4.2/4.3 callers commonly write END without SVLEN.