Assertion failure / segfault in `nw()` during liftover of pangenome-derived VCF

[teepean]

I am trying to lift over the VCF from here:

https://zenodo.org/records/19245725

The error is:

bcftools: plugins/liftover.c:1615: nw: Assertion `cell == 0' failed.
Aborted (core dumped)

or:

Segmentation fault (core dumped)

Environment

• bcftools 1.23 / htslib 1.23
• Linux 7.0.0-10-generic x86_64
• liftover plugin version 2026-02-02 (latest as of 2026-04-06, commit 909d230)

Minimal reproducer

The crash is triggered by a single variant at chr22:11977555 with:

• REF length: 50,523 bp
• 23 ALT alleles totaling ~1,044,892 characters
• Produced by minigraph-cactus (v2.9.2) graph deconstruction + vcfbub

A minimal VCF (crash_chr22.vcf.gz, 548K) containing only this variant is attached.

# Reproduce the crash with the attached single-variant VCF:
bcftools +liftover --no-version -Ou crash_chr22.vcf.gz -- \
  -s hg38.fa \
  -f hg19.fa \
  -c hg38ToHg19.over.chain.gz \
  --reject /dev/null --reject-type v > /dev/null

The chain file used: https://hgdownload.cse.ucsc.edu/goldenpath/hg38/liftOver/hg38ToHg19.over.chain.gz

crash_chr22.vcf.gz

[freeseek]

This should not be happening but this is definitely not a variant that will be liftover correctly by BCFtools/liftover

[teepean]

This should not be happening but this is definitely not a variant that will be liftover correctly by BCFtools/liftover

Any way to avoid it?

[freeseek]

Do not attempt to use the tool on large CNVs. It is really only meant for short SNPs and indels