Looking at https://nf-co.re/docs/developing/components/gpu-modules, I became unsure - what is the correct way for a user/institutional config to flag the lack of GPUs?

If the pipeline uses the

-profile gpu

way, then we can fast fail on that. If it enables GPU parameter based, then I guess it will have to error at task run time ( let the executor accounting handle that I guess ).

But I don't want to fail 🙂

I've emailed support to see what they say about this situation

My recollection (and my reading of that page) says that dual-capability modules should check

task.accelarator

to see if they can expect to have the resource available (and also choose the right container image).

withName '.*' {
    accelerator = null
}

will do that, but then the module itself may break

Right, thanks, maybe that's usable. We can expect pipeline configs not to use the full names?

No, they will, but this is equivalent to a full name, and will override those too.

My reading of https://docs.seqera.io/nextflow/config#selector-priority says there's no priority difference between different kinds of full names, and these all get included from the same context. I guess the actual order of inclusion might play a role?

it will according to my understanding. If they use the full name in the modules.config, then I don't think the catchall override will work, as it's in the same class, but my hope is that pipeline devs will use the simple name to set accelerator out of simplicity.

I forget, are pipelines expected to touch

modules.config

? My hunch is that the config I worry about is in

base.config

which should be read before institutional. Oh well, I'll see if I can make something up to try, thanks for the reminder.

Pipelines set a whole bunch of stuff in the modules.config, primarily prefixes, resource overrides, and I guess also parameter based enabling of gpu's