[Py] Manage parameters like matplotlib

[nbrucy]

Is your feature request related to a problem? Please describe.

The current mechanism to set parameters is a bit confuse and verbose. When setting up a simulation, parameters are intertwined with instructions to set up initial conditions, which is not ideal for clarity.

Describe the solution you'd like

A mechanism similar to matplotlib's rcParams would allow to set up clean parameter files while leaving the possibility to change the parameters in the python script for flexibility.

[tdavidcl]

Thanks for the issue !

I was hesitating before merging #1637. What should we do if there is additional parameters to the function.
For example say that there is a riemann solver with a pressure fix and thus we have to select a threshold. In that case the selection originally would have looked like

cfg.set_riemann_solver_hllc()
# or
cfg.set_riemann_solver_hllc_pressure_fix(threshold = ..../)

Now it would be something like

cfg.set_riemann_solver("hllc")
# or
cfg.set_riemann_solver("hllc_pressure_fix", threshold = ....)
# or
cfg.set_riemann_solver("hllc", pressure_fix = True, threshold = ....)

The issue is that now the parameters need to be parsed within the binding implementation.

.def(
      "set_riemann_solver",
      [](TConfig &self, std::string name, bool pressure_fix, f64 threshold) {
          // parse
      })

But then the enumeration becomes really complex as some riemann solver will not have specific options and the handling of all the cases could end up being a hot mess.

I'm not sure then what is the best course of action ...

[tdavidcl]

Side note: i just stumbled onto this in the pybind11 doc https://pybind11.readthedocs.io/en/stable/advanced/functions.html#accepting-args-and-kwargs

[nbrucy]

Depending on the situation, the parameter could be separated

cfg.set_pressure_fix(threshold = ....)`

with some checks to verify that the final combination is sound.

If you can deal with kwargs then that would help quite a lot, meaning that you could use a structure
like this

cfg.set_riemann_solver("hllc", pressure_fix = True, threshold = ....)

and crash / throw a warning is a kw args is not used by the selected solver.

[tdavidcl]

I would rather find an option that fiddle with kward rather than one that change the state. I think that it would be very error prone otherwise e.g. what if you do

cfg.set_riemann_solver("hllc")
cfg.set_pressure_fix(threshold = ....)

vs

cfg.set_pressure_fix(threshold = ....)
cfg.set_riemann_solver("hllc")

[tdavidcl]

Actually a better way would be to bundle the kargs in a json internally which gets fed to the json parser of the config (which i use anyway in the dumps). Then actually the config would be set using something that is guaranteed to be compatible with the dumps.

[nbrucy]

Yes that's kinda the original idea of the issue :)