FORM crashes at startup (‘Program terminating at -->’) when running parallel jobs on cluster

[maxkno]

I have a relatively large number of FORM files that I am processing in parallel on a cluster. Each job is independent (no shared state between them).
However, for some of the jobs FORM crashes at startup without producing a meaningful crash message. All I see is:

Program terminating at  --> 

Running the same .frm file manually works fine. Even running form without any input (just form at the shell prompt) produces the same cryptic message in these failing cases.
To rule out scratch file collisions, I moved each job into a separate working directory, but this did not resolve the issue.

Id appreciate any help debugging this further. I currently run version 4.2.1, build from this repo.

[jodavies]

Could you try with the current master branch, and compile with --enable-backtrace? For the best output you will need elfutils (or similar) provide eu-addr2line.

[maxkno]

Thanks for the quick response. I'm now getting

Program terminating at  --> 
Terminate called from startup.c:1611 (onErrSig)

for both setups, i.e. with and without a Form file to be processes.

[vermaseren]

If it has to do with you quick-firing those jobs, you could try what happens if you build a small delay between the startups.

…

On 27 Aug 2025, at 19:56, Max Knobbe ***@***.***> wrote: maxkno left a comment (form-dev/form#704) <#704 (comment)> Thanks for the quick response. I'm now getting Program terminating at --> Terminate called from startup.c:1611 (onErrSig) for both setups, i.e. with and without a Form file to be processes. — Reply to this email directly, view it on GitHub <#704 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABJPCETHJJAIQFOAC2JVL7T3PXWNDAVCNFSM6AAAAACE6I3QVGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTEMRZGE4DOMRYGU>. You are receiving this because you are subscribed to this thread.

[jodavies]

Terminate called from startup.c:1611 (onErrSig) this means that FORM caught a signal, can you check what it is? Crashes due to memory allocation failure etc should look different.

One thing that comes to mind: when running on clusters where not all CPUs support the same features, you need to compile with --disable-native or you'll get SIGSEGV or SIGILL from cluster nodes which don't support AVX2 (for example).

[maxkno]

Hi, thanks again for the quick help.

--disable-native indeed solves my issues, probably because the login node where I installed form indeed has different specs than some of the worker nodes.