getargs VS actix-telepathy

Actually, I would say getargs on this one. Full disclosure, I'm a major contributor to the library, but it is also faster than lexopt, and provides a bit more control.

I have benchmarked getargs to have significant performance gains from inlining.

I'd recommend getargs over argparse, it's also very well documented (but that version is not on crates.io yet)

It pains me that getargs 0.5.0 is still not released yet. If you can deal with a git dependency, I would highly recommend it, it can do exactly what you want with no fuss.

I'm currently working on bringing up getargs, which was abandoned on crates.io over 2 years ago. It has an API design that I agree with and I wanted to make it great, so I'm giving it a visit from the PR fairy.

If it's perfectly fine then it shouldn't be terribly difficult to write it in a way that the borrow checker is happy with. The problem is reformulating your problem such that it's easy to either annotate the lifetimes in that way or have the compiler infer them for you. And it's made more difficult if your libraries have incorrectly elided lifetimes.

Well, one recent issue I came across recently is the lack of support for clusters. There are crates for parallelism on a local machine, but the crates wrapping MPI or coming up with a native solution are basically not maintained anymore. I've only found actix telepathy, which is not a complete solution tho, being an extension of Actix.

Rust is also another possibility: it's basically C++ but more modern with added features and safety. It can be tricky to write mathematical stuff in it, because you may not care too much about all the safety concerns Rust forces you to handle, but it can be useful to catch bugs ahead of times. Sadly, Rust seems to have no library for running programs on clusters of PCs, except maybe this one, which takes the Actor model implemented by Actix and runs it on a cluster. I don't know how tricky it is to use the Actor model for a scientific simulation, tho.