But I would appreciate if you could run all tests on your hardware and tell me if there are any remaining issues, as I am now very familiar with what can go wrong

Okay, all the R scripts started failing with

Fatal error: cannot create 'R_TempDir'

on my Mac when they had worked just fine before. Any particular thing that changed in the images that might have affected this that you can think of?

Will try on my HPC as well

Okay, I think I've found a fix, which is to set the TMPDIR environment variable using

containerOptions

in each R module in

conf/modules.config

. It seems that the

/tmp

dir is owned by root, while the commonly used

runOptions

include linking host user inside the container, which makes

/tmp

unwritable by the user - I assume this is only because of the emulation happening on Apple Silicon. What made it harder to diagnose was that R for some reason tries to set the tempdir before even finishing initialising itself, meaning that any code I added to the template file was never run (so you can't set the tempdir inside the R template, since it'll never get the chance to run). Since the modules aren't using the

script:

directive we can't specifically export a new TMPDIR environment variable there either, while setting it in the module-specific options. Would love a prettier solution if you have one 😅 I'll make the rest of the modules with this change as well to make sure it works across the board.

Though I can certainly be wrong with not being able to set environment variables in the script part before using the template - do you know how to do this with Groovy?

Though thinking about it, I wonder what's the most appropriate... The issue here is specifically using containers, so setting

containerOptions

could be more apt. What do you think?

Also, using

beforeScript 'export TMPDIR=.'

doesn't seem to work

Maybe it does not work because of this

When the process is containerized (using the container directive), the

beforeScript

will be executed in the container only if the executor is container-native (e.g. cloud batch executors, Kubernetes). Otherwise, the

beforeScript

will be executed outside the container.

This is what it is currently:

containerOptions = '--env TMPDIR=.'

inside the affected modules in

conf/modules.config

But I am sure someone else has to have encountered this before

Would could keep the fix as-is for now and think about something else when higher priority items are cleared, for example

I have nf-test 0.9.5, but would not expect that to make a difference

Also passes

Any organization- or user-level config that might interfere?

Sorry for wasting your time 😅

No problem, but good it works now

Would still like to know if all the tests pass in your setup

Yes, I've been running on the HPC in parallell, but it takes some time (I can't parallelise there with nf-test), so we'll see, but I'll run locally as well

Seriously considering just getting a Linux machine at some point

The HPCs we have access to are... not ideal, let's say, so I actually got my current laptop specifically to be able to run almost all analyses I do locally (which is fairly common now where I work), so I' ve got 1 TB of storage and 64 GB of RAM in an M2 MacBookPro.

And I agree, the Apple hardware is hard to beat - were I to get a Linux machine I'm fully expecting a downgrade in various areas