PyBaMM 25 breaks Pyodide/wasm compatibility

[WhyNotYT]

PyBaMM 25 no longer works with Pyodide, since solving now relies on the pybammsolvers package.
PyBaMM 24 still works because it could use CasADi directly.

I was able to get PyBaMM 25 working in Pyodide by patching 3 calls to pybammsolvers in processed_variable.py.

If PyBaMM is fully moving to pybammsolvers, it would be great to:

• Maintain wasm/Pyodide compatibility, or

• Provide a fallback to CasADi-only solving.

A wasm build of pybammsolvers would likely fix this.

[kratman]

I don't think this is a high priority unless you can provide a strong reason to go through the work of a pybammsolvers WASM build

[agriyakhetarpal]

Hi @WhyNotYT, thanks for opening this issue. I didn't know that people were interested in PyBaMM working in Pyodide! Yes, PyBaMM is fully moving to pybammsolvers as the default solver. I previously experimented with a WASM build at agriyakhetarpal/pybamm-lite#1, and I was previously stuck on some build-time parallelism handling that I now think I understand how to handle. I could take this up soon and add a minimal Pyodide CI job for pybammsolvers.

[agriyakhetarpal]

I see that you have https://github.com/WhyNotYT/pybammsolvers-pyodide and https://github.com/WhyNotYT/PyBaMM-Pyodide. Please consider merging your efforts with mine; this will be beneficial for all users, as I plan to move my repository to the PyBaMM organisation once it is complete.

[WhyNotYT]

Hi @agriyakhetarpal, thanks for your help!
I don't have much experience in building python packages or in WebAssembly. So I couldn't make any notable progress in porting pybammsolvers to pyodide based on the pyodide docs. The https://github.com/WhyNotYT/pybammsolvers-pyodide is pretty much just an empty fork.
In https://github.com/WhyNotYT/PyBaMM-Pyodide, I removed all references to pybammsolvers and rewrote three functions (idaklu.VectorRealtypeNdArray, idaklu.observe_hermite_interp, idaklu.observe) in pure python, I can merge this but I don't think it helps in packaging pybammsolvers for pyodide or with the broader PyBaMM project.

[WhyNotYT]

Hi again,

I have managed to compile pybammsolvers for pyodide/wasm32.

https://github.com/WhyNotYT/pybammsolvers-pyodide

As of now, only the casadi solver works, and casadi 0.7.0 is compiled together. Overall, it's 50% faster than using the casadi in the pyodide repos and pybamm 24.

Would it make sense to work on upstreaming this to either pybamm or pyodide? I haven't conducted much testing apart from the basics.

[agriyakhetarpal]

Hi @WhyNotYT, thank you for your work and for stepping up! This sounds great. We recently landed PEP 783 in Pyodide-land, so this is at the right time!

I didn't have it at the top of my priorities to work on refactoring the IDAKLU solver to compile to WASM fully – and don't expect to be able to help out, either, until the next few months. So I think this is quite alright for now. We are just waiting for PEP 783 to be merged into warehouse, and then we can start uploading PyEmscripten wheels to PyPI. I'd like to note that we do have a new release with a stabilised ABI for 2026 with Python 3.14, so could you please build it against this xbuildenv instead? https://github.com/pyodide/pyodide/releases/tag/314.0.0a1? We have added more details here: https://pyodide.org/en/stable/development/abi.html#what-is-the-pyemscripten-platform

After doing so, yes, it would be great if you could upstream it to pybammsolvers as a PR. We can run the PyBaMM test suite there to see how much of it works before committing to uploading PyEmscripten wheels. I'll merge it soon after if things look promising.

Thanks again!