Migrate build system away from distutils (deprecated in NumPy)

[mesonepigreco]

Migrate build system away from distutils (deprecated in NumPy)

NumPy has deprecated distutils, which our project currently relies on to compile a significant amount of Fortran code. This code links against LAPACK/BLAS libraries, but in some environments we also need to compile against Intel MKL instead.

To ensure long-term compatibility, automated builds, and portability, the build system should be migrated to a modern alternative such as CMake or Meson.

This task is self-contained: it does not require any knowledge of the scientific functionality, algorithms, or physics implemented in the code. The scope is strictly the build infrastructure.

---

✅ Requirements

• Replace the current distutils-based build system with CMake or Meson.
• Ensure compilation works with:
  • LAPACK/BLAS backends (Linux, standard HPC clusters).
  • Intel MKL (Intel compilers).
  • Apple ARM (M1/M2) architecture (linking against Accelerate framework or equivalent BLAS/LAPACK libraries).
• Preserve automatic detection of available math libraries where possible.
• Update documentation with build instructions for the supported platforms.

---

🚀 Getting Started Checklist

1. Set up the environment

   • Clone the repository and check out the build system files (setup.py, distutils configs, etc.).
   • Install dependencies (Fortran compiler, BLAS/LAPACK, MKL if available).

2. Choose a build system

   • Decide whether to use CMake or Meson (both are acceptable).
   • Initialize the new build system files (e.g. CMakeLists.txt or meson.build).

3. Add Fortran sources

   • Configure the build system to compile all existing Fortran files.
   • Ensure that object files are correctly linked into the Python extension module(s).

4. Handle linear algebra libraries

   • Add detection logic for:
     • OpenBLAS/LAPACK (default).
     • Intel MKL (on Intel compilers).
     • Accelerate framework (macOS ARM).

5. Test the build

   • Compile and install locally using the new build system.
   • Run the test suite to confirm functionality is unchanged.

6. Update CI

   • Modify the continuous integration workflow to use the new build system.
   • Ensure CI covers Linux, Intel with MKL, and macOS ARM.

7. Documentation

   • Update the README / installation guide with new build instructions.

---

💡 Why this is a good first issue

• No deep dive into algorithms or physics is needed — the challenge is purely build infrastructure.
• It improves portability and maintainability.
• It’s a great way to contribute without spending too much time learning the internals of the code.

[diegomartinez2]

I'm working on a Meson build system. Right now it 'works' on CellConstructor, but I don't have a way to test it on macOS.
Basically I'm learning Meson by working on this, so do not expect a perfect job.

[mesonepigreco]

Wow, this is super useful! Thank you @diegomartinez2

[mesonepigreco]

When this is done and working for the full suite, we can release the next version 1.5

[diegomartinez2]

Hi Lorenzo, I need somebody to test the meson install in other computers before adding it to master. I only tested in my laptop on a conda env, but it already has SSCHA in the base system so I have doubts. I think it will work on Apple ARM (M1/M2) architecture, but I do not have way to test it.
Meson_compiler_test

[diegomartinez2]

Right now the python-sscha project can't work because the CellConstructor dependency is not only old, but apparently has a problem with the numpy library (maybe it's not in requirements.txt of the old version?). Maybe it's solved by the new CellConstructor version.

[mesonepigreco]

Fixed by @diegomartinez2 and merged into master #115