SLICOT C and Rust Translations

[jamestjsp]

There are some efforts to translate F77 code to Rust and C
https://github.com/jamestjsp/slicot-rs
https://github.com/ilayn/SLICUTLET

[andreasvarga]

Hi,

First, I have no knowledge of Rust, but I am convinced that a translation of SLICOT to Rust is not trivial! I inspected just one of translations, namely of the AB01MD.f routine, and I have some questions:

1. How this translation has been performed?
2. How much has been done automatically (e.g., using Claude) ?
3. Since the code structure has been changed, I wonder if this was done manually ?
4. Somewhere in the translated code (see here), there is a comment

"// System is unobservable"

This is a complete nonsense, the function being dedicated to analyse controllability! So, I wonder how such a message can appear in this code (sorry if I oversee something)?
5. How much testing has been done?
6. Finally, are Rust-based implementations more efficient than the Fortran versions?

[jamestjsp]

@andreasvarga It was done using Claude Code almost 100% except some tests. The Claude family of models tend to write more comments throughout the codebase (can be tuned using user prompts).

[ilayn]

I am the translator for the C version mentioned above. My plan is to write things into plain C and add some tests slowly as I do the translations. Once M family is done, I'll circle out to A,T and S families. I don't know if I will ever get to 100% but at least for python-control and a few other projects I would like to get the basics included.

The progress is so far is given in ilayn/SLICUTLET#1 but as a summary, (as of today) MA, MB01, MB02 families are done and tested via Python bindings. Some remarks;

• The main reason why I am translating is due to the difficulties folks are having on different architectures and BLAS/LAPACK linkage annoyances. In comparison, C is well portable and easier to handle. I am familiar with the algorithms hence it makes it easier to translate. LLM supported translations unfortunately did not turn out to be too fruitful.
• Python ecosystem and in particular pytest module is orders of magnitude superior compared to Fortran or C testing tooling convenience. Hence it is trivial to generate numerous tests with LLMs and immediately tested on GitHub CI/CD for any pull requests. Some random test example here https://github.com/ilayn/SLICUTLET/blob/main/python/tests/test_ma02xx.py for testing ma02xx family.
• I switched to meson build system for its simplicity and since I already have the Python bindings.

I will report back if I encounter anything along the way.

In the meantime, if you do have similar efforts or plans, please let me know and I can happily hand over the repo. I don't necessarily feel any strong ownership over the translated code and the credit should rightfully go to you. I am just trying to close the gap for contemporary workloads.

[grisuthedragon]

@ilayn can you drop me an email, so we can discuss it during the next weeks.

[ilayn]

Sure, where can I find your email? Otherwise, if you want to keep it undisclosed, drop one to ilhanpolat on the famous google mail service.

[jamestjsp]

@ilayn I found the email in spam, it must be from the domain @mpi-magdeburg.mpg.de