impg syng2gfa: failed to find symbol N offset 0 in linear rskip

[MarsicoFL]

impg syng2gfa fails with failed to find symbol 515 offset 0 in linear rskip

Affected: impg 0.4.1 (this main branch) and pangenome/impg:1778787077 Docker image.

impg syng2gfa aborts immediately after loading the GBWT with a fatal error
in syng's rskip (run-length compressed skip list) lookup. The error is
deterministic across inputs of any size — reproduces on as little as 2
yeast haplotypes × chrI (124 kB). The companion impg syng index build
runs cleanly; only the GFA export step is broken.

Reproduction

# Get two yeast S. cerevisiae chrI sequences (any small FASTA pair works,
# this is just what we used).
agc getctg yeast235.agc S288C#0#chrI W303#0#chrI > yeast2_chrI.fa
bgzip yeast2_chrI.fa
samtools faidx yeast2_chrI.fa.gz

# Build index — succeeds.
impg syng -f yeast2_chrI.fa.gz -o yeast2_chrI

# Export GFA — fails.
impg syng2gfa -a yeast2_chrI --sequence-files yeast2_chrI.fa.gz -o out.gfa

Expected

A valid GFA 1.0 (or 1.1 with --gfa-version 1.1) written to out.gfa.

Actual

[2026-05-14T20:51:38Z INFO  impg::commands::syng2gfa] [syng2gfa] loading syng index from prefix 'yeast2_chrI'
[2026-05-14T20:51:38Z INFO  impg::commands::syng2gfa] [syng2gfa] building sequence index for 1 file(s) (gap fill)
read GBWT with 10349 vertices and 20944 edges from yeast2_chrI.1gbwt
Linear max 16 nSym 2 free 13
    DIR  0 sym 511 offset 42 count 4
    DIR  1 sym 515 offset 42 count 2
    1: i 15 iSym 0 count 4 sum 4 sSum 4
    2: i 14 iSym 1 count 2 sum 6 sSum 2
[2026-05-14T20:51:38Z INFO  impg::commands::syng2gfa] [syng2gfa] 10349 syncmer segments (length 63 bp), GFA 1.0, gap fill: FASTA/AGC
FATAL ERROR: failed to find symbol 515 offset 0 in linear rskip

out.gfa is created but empty (0 bytes).

What I tried (no effect)

• --gfa-version 1.0 vs --gfa-version 1.1 — both fail at the same point.
• Subsampling: same error with 6 strains × chrI (1.35 MB) and 2 strains ×
  chrI (124 kB). Error is independent of dataset size.
• --parallel-dictionary on the index build — index still builds, GFA
  export still fails identically.
• Local build (cargo install from main, Rust 1.95 + conda-provided htslib)
  vs Docker image pangenome/impg:1778787077 — same bug in both, so this
  isn't a build artifact.

Adjacent observation (possibly related)

impg syng -f yeast6.fa.gz (6 yeast haplotypes, whole genomes, ~72 MB)
fails during the index build itself on chrIV with:

FATAL ERROR: syngBWTadd -ve output retVal -3 callCount 139595 k 262830

Both errors are inside the GBWT/rskip layer and may share a root cause.

Environment

• impg: 0.4.1 (this main, commit pinned by submodule) and Docker
  pangenome/impg:1778787077
• syng vendor submodule SHA: dc18190342c32349a060816f1fafcc51185d2922
• OS: Ubuntu 24.04.1, glibc 2.39
• gcc 13.3.0, cmake 3.28.3, Rust 1.95.0
• Data: yeast 1011 dataset (yeast235.agc from garrisonlab S3 bucket),
  chr-I extracted via agc getctg.