[danilomonge]

When ext.args does not set --genomeSAindexNbases, star/genomegenerate computes it from the genome length. The STAR manual says that for small genomes the value must be scaled down to min(14, log2(GenomeLength)/2 - 1) (parametersDefault). STAR warns when the value is larger than log2(GenomeLength)/2 - 1 and recommends int(log2(GenomeLength)/2 - 1) (Genome_genomeGenerate.cpp). The module rounded log2(GenomeLength)/2 - 1 to the nearest integer (printf "%.0f"), so whenever the fraction was 0.5 or more it chose one more than the limit:

Genome length	log2(L)/2 - 1	Before	After (STAR's value)
40,001 bp (test genome.fasta)	6.64	7	6
140 Mb	12.53	13	12
600 Mb	13.58	14	13
3.1 Gb	14.77	14	14

With 7, STAR 2.7.11b logs --genomeSAindexNbases 7 is too large for the genome size=40001, which may cause seg-fault at the mapping step. Re-run genome generation with recommended --genomeSAindexNbases 6. This PR rounds down with int(), so the module uses the value STAR recommends and the warning is gone. Genomes above about 1 Gb (capped at 14) and runs that set --genomeSAindexNbases in ext.args are not affected.

The snapshots do not change. The star/genomegenerate snapshot records the index file names and the versions, and the star/starsolo outputs, the only other CPU test that uses the computed value, have the same md5 sums with 6 as with 7. The other tests that build an index with this module set --genomeSAindexNbases 9 in their configs.

#13453 also edits this gawk line (it adds $args3), so whichever of the two PRs is merged second needs a one-line merge.

Generated by Claude Opus 5.5

PR checklist

• This comment contains a description of changes (with reason).
• If you've fixed a bug or added code that should be tested, add tests! (the existing tests run the computed value; their snapshots are unchanged)
• Remove all TODO statements.
• Broadcast software version numbers to topic: versions - See version_topics
• Follow the naming conventions.
• Follow the parameters requirements.
• Follow the input/output options guidelines.
• Add a resource label
• Use BioConda and BioContainers if possible to fulfil software requirements.
• Ensure that the test works with either Docker / Singularity. Conda CI tests can be quite flaky:
  • For modules:
    • nf-core modules test <MODULE> --profile docker (at a6f93fe, on linux/amd64 with nf-test 0.9.5 and Nextflow 26.04.0: nf-test test --profile=+docker --ci on star/genomegenerate and on star/align, star/starsolo, arriba/arriba, arriba/visualisation and fastq_align_star, whose tests use it: 26 of 26 passed against the snapshots; nf-core modules lint star/genomegenerate: 0 failed)
    • nf-core modules test <MODULE> --profile singularity (not run locally; CI runs it)
    • nf-core modules test <MODULE> --profile conda (not run locally; CI runs it)