Roadmap to pip-based runtime + ctf dependencies

[balbasty]

This thread aims to discuss ways of packaging the matlab runtime and the spm CTF in ways that

1. ease installation and update
2. ensure compatibility between the matlab runtime and the spm CTF
3. makes pip the only entrypoint (no need for people to figure out how to install the runtime themselves)
4. lets spm-python find the path to the runtime in a robust way

This follows from this discussion.

Related issue: #18

Solution 1: matlab runtime as a pip package

The nicest thing from an spm-python point of view would be to have the matlab runtime be fully shipped by pip, which would allow users to do

pip install matlab-runtime=r2024b

and spm-python could find the runtime through something like

# pre 3.9
import os.path
import matlab_runtime
runtime_path = os.path.dirname(matlab_runtime.__file__)
# ...

# post 3.9 (supports zipped packages)
# NOTE: I am not sure whether the zipped runtime can be executed directly or need to be unziped
import importlib.resources
runtime_files = importlib.resources.files('matlab_runtime')

However, matlab is definitely not publishing something like this, and we are not licensed to build and distribute such a package (I think).

Solution 2: matlab installer as a pip package

Another solution is to build and publish a runtime installer. One would do

pip install matlab-runtime-installer

which would have thr entrypoint

matlab_runtime_installer --version r2024b

This installer would need to either:

1. always install the runtime in the same place (e.g. $HOME/.matlab/runtime/<version>), or
2. let the user choose the installation path (with a default location as 1.), and
3. write the location of the installation in an environment variable -- which means editing the rc file (on mac/linux) or $PROFILE (on windows) -> we'd need to see how conda deals with this across platforms.

If no matlab-runtime is installed at python runtime, we should open an interactive context that lets the user effectively run matlab_runtime_installer.

Note: we cannot install the runtime inside the matlab-runtime-installer package, as it can be a zipped package, and it is anyway bad practice.

Questions:

• should the matlab-runtime-installer package be versioned per-matlab (i.e. pip install matlab-runtime-installer=r2024b means that matlab_runtime_installer can only install r2024b and the --version flag does not exist)? Or should any version of matlab-runtime-installer be able to install any version of the matlab runtime?
• If the latter, how do we let spm-runtime[r2024b] specify that it needs r2024b in its dependencies?

Solution 3: keep things as they are

I.e., ask people to install the runtime themselves, and make sure they have the correct version.

---

Any other idea? Details I missed?
@johmedr

[balbasty]

@johmedr I have started a tentative matlab runtime installer package (only tested on mac so far, and not with all versions).

https://github.com/balbasty/matlab-runtime-installer

[johmedr]

Amazing! Any chance this tool can also have a detect function that we can call from "spm-python" to automatically add Matlab to the path? (or print nice help to the user?)

[balbasty]

Yes there's a guess_prefix() function that looks at either the MATLAB_RUNTIME_PATH env variable (which I made up), or if it is absent, the default installation prefix of the runtime.

Then, it's just about checking if the target version is present under that prefix, and if not, run the install function to add it.