Hi all,
I am running GSNAP version 2025-07-31 on macOS Sonoma 14.4.1 on an Intel Mac Pro with 128 GB of memory.
I'm aligning paired-end data against HG38, I'll attach the run parameters below
On two occasions now I see a core dump with the error
rungsnap.sh: line 4: 73620 Illegal instruction: 4
So far I can find no useful discussion threads about this problem
I 'solved' the problem the first time by dividing the input into ~50 paired files of a million reads each, and running them all separately. I discovered a single 'bad' file pair, and just discarded that data and merged the rest.
Now I have seen it happen a second time. This is in the course of processing dozens of files of similar size, so this is really an isolated problem.
Anyone else run into this? To really track this down I would need to build gsnap from source and run in the debugger, or else continue narrowing down the 'bad' file to identify the problematic content, would first like to verify that no one else has seen this.
Thanks in advance,
Randy
Settings :
-d hg38 --nthreads=16 --use-splicing=hg38.introns.iit -n 24 -Q --format=sam --force-xs-dir --gunzip
1 answer
Data-dependent SIGILL like that is almost certainly a real bug in the aligner rather than anything about your machine - an instruction-set mismatch would die on the first read every time, not on one pair out of fifty files.
Once you've narrowed the bad pair down (binary search inside the file gets there a lot faster than splitting into 50), it's worth sending to the GMAP/GSNAP list. Thomas Wu does act on reports that come with a reproducer, and one read pair that reliably crashes it is about as good as a bug report gets. I haven't hit this particular one myself so no idea what's in the read.
Log in to answer this question.