Easy-search crashes during mult-iteration nucleotide searches #1128

Description

Multi-iteration (>1) untranslated (nucleotide-to-nucleotide) easy-search workflow (MMseqs2 Version: 18.8cc5c) fails after reporting "Query sequence has a result with no diagonal information" (segmentation fault). Single-iteration searches and translated searches work correctly.

Steps to Reproduce

  <output.tsv> <tmp_dir> \
  --threads 20 \
  --num-iterations 2 \
  --search-type 3

Expected Behavior

Search completes successfully and outputs alignment results.

Actual Behavior

1. Search begins normally
2. Query sequences are processed
3. Several errors appear: Query sequence <N> has a result with no diagonal information. Please check your database.
4. Program terminates with segmentation fault

Error Messages

(...)
Calculation of alignments
Query sequence 265 has a result with no diagonal information. Please check your database.
Query sequence 455 has a result with no diagonal information. Please check your database.
corrupted double-linked list
Segmentation fault (core dumped)
Error: Alignment died
Error: Search step died
Error: Search died

Environment

• Conda/mamba installation of 18.8cc5c (Ubuntu 22.04.5 LTS)

nucleotide_example.zip

• Database type: Nucleotide (untranslated/nucleotide-to-nucleotide)
• Multiple databases and naming conventions (size 20kb-23mb)

Key Observations

1. Multi-iteration nucleotide searches ONLY: --num-iterations 1 works fine
2. Protein and translated searches unaffected: Same data with translated search mode works correctly
3. Lost diagonal information?: