hisat2 version 2.2.3 failing with SIGILL on old infrastructure

[cat-bro]

Galaxy Australia's training pulsars run on old hardware with Intel Sandy Bridge CPUs and these have been OK for all Galaxy tools until bioconda build of the the latest hisat2 (9d56719). The latest hisat2 fails with SIGILL on the old hardware. This may be because of the addition of the flag -march=x86-64-v3 in CXXFLAGS for this version.

[bgruening]

@cat-bro that looks correct. The new version assume the newer instruction sets.

ping @mencian

[mariabernard]

Hi,

I encountered same problem with the vsearch recipe. since version 2.30.1 a simple vsearch --version return Illegal instruction (core dumped) error. The error is reproducible locally but also on our cluster with Galaxy server.

What solution do we have (other than using only older builds) ?

[bgruening]

Can you use containers? Docker or Singularity?

[mariabernard]

This really isn't my area of expertise. We’re trying to update the “compilation instructions” (I really don’t know what I’m talking about ;-)).

My main concern is this: as a developer of the FROGS pipeline, should I pin the vsearch dependency to an older version, or stay up to date and note that user may need to revert to an earlier version to ensure compatibility with their old “compilation instructions”?

[bgruening]

Imho you should stay up to date, this error should only occur on very old hardware. For people with this hardware they would do conda create -n foo frogs vsearch=1.2-old-version

[mariabernard]

ok thank you Björn!