Preferring Lmod-based initialisation

[ocaisa]

Using Lmod-based initialisation would allow using other shells and also opens the door to being able to switch EESSI versions within a single session.

The approach is being hardened in EESSI/software-layer-scripts#153 and there are new features that could be leveraged.

Personally, I would really like to move away from using the bash script initialisation. Right now innovation has to be implemented in both places which is a waste of effort.

[ocaisa]

It's also possible to do version-less initialisation with the Lmod approach

[bedroge]

Can/should we start adding deprecation warnings to the old init scripts, and maybe even just (automatically, unless overridden somehow with an env var?) redirect them to the new ones?

[ocaisa]

There's CI to check that the two approaches give identical final environments, so in theory redirecting them should not have an effect.

However, as they are right now this is only true where the host does not have Lmod (and has not called the module command yet in the session). EESSI/software-layer-scripts#153 is intended to solve those problems, and more aggressively reset Lmod. Once that is merged I think replacing the old script by a symlink should be the next step.

[trz42]

While I agree that the initialization via Lmod is preferable the current implementation has some shortcomings:

• it doesn’t show what’s going on, which makes it harder to explain in demos
• the prompt isn’t changed, so it’s harder to know what the current environment is
• the last line in the script initialization has a nice touch “… have fun”
• the name/path of the script says “…/init/Lmod/bash”, but it does more than that (also loads an EESSI module)
• it may use/support different variables, and the logic may not be exactly the same as the one via the traditional init script

[boegel]

The init script is way too noisy, that's definitely a downside.

I would say eessi init should have an option like --mode, so you can pick which one you want to use, even if we'll deprecate the init scripts soon.

Default for --mode should be Lmod, and we can also add a --verbose so more info is printed after loading the module (like which CPU/GPU was detected, etc.)

[bedroge]

I wouldn't necessarily call them shortcomings, some users may even find the quite lengthy output of the other script a shortcoming 😉
Also I don't think they need to be exactly the same before we start deprecating the old one. If we don't immediately use a symlink, but instead a wrapper that prints some information about the deprecation and then automatically redirects you to the new one (with a way to still disable that, so users can at least still use the old one and inform us of why the new one doesn't work for them), we can explain what's going on and why we're doing this.

We can also still modify the new scripts and make many of those things that you mention optional using environment variables (e.g. do you want to have your prompt changed? do you want verbose/debugging output? etc). That way they will be more similar, but maybe with new defaults.

[ocaisa]

I wanted to have the prompt change as part of the EESSI module but this is really hard to get right for all the available shells. As part of those scripts it is not too bad, you just do it individually for each shell (if that is possible).

What I was looking for is that you get an EESSI prompt when you have an EESSI module loaded, but this is very hard to do for all shells (PS1 is not universal, they have different methods for the prompt).

[ocaisa]

Also, you can make the intialisation module more noisy with export EESSI_DEBUG_INIT=true, so with the @bedroge approach you can

• Set EESSI_DEBUG_INIT
• Add the PS1 change
• Still have fun

Regarding the logic, we have a check in CI to ensure that init/bash and init/lmod/bash give the same final environment so I don't think how they get there is really that relevant.

[ocaisa]

For PS1, I guess what we could do is update it only if it exists, that will solve everything except csh and fish shells which is probably good enough.

[trz42]

Fully agree that the init script is too noisy, but Lmod init is way too silent.

For me it would be enough if it said something like detected CPU xyz + GPU abc … setting up environment accordingly… for more information run with EESSI_INIT_DEBUG=1 … have fun

[ocaisa]

I think the scripts can do something like this (it is really a personal preference), but the module should be silent.

[bedroge]

We can have another module on top of the EESSI one that enables all the verbose output: module load NOISY/2025.06? 😁