ENH: Add translations of MB01XX family

[ilayn]

These tests are generated with LLM but they need modifications because I used the wrong prompt. Moreover, the translations can still have mistakes. Fun.

CC @jamestjsp if you want to stretch your test writing muscle

[ilayn]

I also added some build instructions to build and test the repo.

[jamestjsp]

My workflow is as follows.

1. select a SLICOT routine
2. Run the dependancy analysis tool Python tool
3. Create a multi-phased plan
4. Create GitHub issues and subissues based on the plan.
5. Each phase will create a PR
6. Independent routines can be generated parallel by subagents using git worktree, then the main agent will merge to the phase branch and create PR

[ilayn]

That won't work because you have to tame the LLM. Otherwise it starts spewing nonsense. Task it with the task of resolving this pr test failures and you will see it.

[jamestjsp]

Let me give a try, I shall update you soon.

[jamestjsp]

@ilayn
I just setup the machine and start investigation.

I have made a commit to a feature branch
jamestjsp@73fefe4

Changes:

• src/MB/mb01ld.c: Removed 'k' parameter from function signature
• include/SLICUTLET/slicutlet.h: Updated declaration to match
• python/extensions/extension_mb.c: Removed 'k' from py_mb01ld wrapper
• python/tests/test_mb01xx.py: Fixed both tests to remove k argument

[ilayn]

I appreciate the effort but you need to tame that LLM so that it does not do some irrelevant stuff like adding uv support or changing the formatting etc. It just needs to fix test failures by running it itself, and adjusting the wrapper and the code and then maybe offer recommendations. Typically this is only possible only by writing a very long description of the problem at hand. Mine is almost like a full page. Then I set it free so it starts iterating over those precise goals.

As far as I can see, your LLM is running wild right now and adding random stuff which is not something we want. They are terrible at precise goals without precise explanations. You have to restrict their context to the minimum

[jamestjsp]

Need to tune the CLAUDE.md file.

[jamestjsp]

@ilayn what was happening was the agent was running ruff check that was doing all those formatting and file diff became almost useless for code review. :)

[jamestjsp]

@ilayn can you review this test fix plan? It is planning to remove some tests. jamestjsp#3