Summary
bcftools call -m (no -v) emits a record at every position with reads. On a 33× WGS BAM that's ~1 billion records / ~23 GB output, vs ~10M records / ~1 GB with -v. The flag is documented, but the default-without--v behavior reliably surprises users following the canonical WGS pipeline; the 100× output blow-up then breaks every downstream consumer (plink2 --vcf OOMs building metadata; bcftools view -R becomes intractable — see #2557).
A one-time stderr warning when -v is omitted on whole-genome input would catch the trap with zero behavior change.
Reproducer
bcftools mpileup -f hg38.fa wgs.bam | bcftools call -m -Oz -o out.m.vcf.gz # ~1B / ~23 GB
bcftools mpileup -f hg38.fa wgs.bam | bcftools call -mv -Oz -o out.mv.vcf.gz # ~10M / ~1 GB
Both are accepted; outputs diverge by 100×. The variant-calling howto uses -mv, but nothing flags the absence of -v as unusual.
Suggested fix
In vcfcall.c after flag parsing:
if (call_mode == MULTIALLELIC && !variants_only && !gvcf_mode) {
fprintf(stderr,
"WARNING: `bcftools call -m` invoked without -v; will emit a record at "
"every callable position (often 100× larger than expected for WGS). "
"Add -v for variant sites only.\n");
}Skipping the warning under -g covers the intentional-invariant-output case. An optional second gate ("only warn if input regions cover > 1 contig and > 1 Mbp") would further suppress noise for capture-panel runs.
A warning preserves backward compatibility: existing scripts unchanged, new users get one chance to spot the trap.
bcftools 1.16, 1.23.1; help text in 1.23.1 confirms current behavior.
Summary
bcftools call -m(no-v) emits a record at every position with reads. On a 33× WGS BAM that's ~1 billion records / ~23 GB output, vs ~10M records / ~1 GB with-v. The flag is documented, but the default-without--vbehavior reliably surprises users following the canonical WGS pipeline; the 100× output blow-up then breaks every downstream consumer (plink2 --vcfOOMs building metadata;bcftools view -Rbecomes intractable — see #2557).A one-time stderr warning when
-vis omitted on whole-genome input would catch the trap with zero behavior change.Reproducer
Both are accepted; outputs diverge by 100×. The variant-calling howto uses
-mv, but nothing flags the absence of-vas unusual.Suggested fix
In
vcfcall.cafter flag parsing:Skipping the warning under
-gcovers the intentional-invariant-output case. An optional second gate ("only warn if input regions cover > 1 contig and > 1 Mbp") would further suppress noise for capture-panel runs.A warning preserves backward compatibility: existing scripts unchanged, new users get one chance to spot the trap.
bcftools 1.16, 1.23.1; help text in 1.23.1 confirms current behavior.