Install SymbolicRegression directly from GitHub repo

[MilesCranmer]

This would finally eliminate #27, as PySR would install the backend directly from GitHub, rather than through the Julia registry. No need to wait for the registry to update.

@mkitti does this look like what you had in mind?

(note that I use the temp dir for other files, so create it manually, rather than using the temp argument to Pkg.)

[mkitti]

This confuses me. What is this second io_arg argument?
https://pkgdocs.julialang.org/v1/api/#Pkg.add

[mkitti]

Ah, I see https://pkgdocs.julialang.org/v1/api/#Redirecting-output

[MilesCranmer]

Just to add: the reason I have to do the Main.eval here is because the PyObject wrapper for IOStreams doesn't seem to pass correctly from Python. It's a bit ugly but works for now. The reason I pass the full io_arg instead of something like io={io} is because the IO redirection was only implemented in Julia 1.5+, so I set io_arg="" for earlier versions.

[mkitti]

Next question: Why does this environment need to be temporary?

You could create one in the environments directory of the Julia depot. The short hand for this starts with an @. For example, @pysr.

Support for @ from the command line came with 1.7, but I think support within Pkg has been around for longer.

[MilesCranmer]

I did not know you could do that, thanks! I will switch to using that instead.

Any idea how to pass this to Pkg.activate as a variable? I see two different outputs currently.

with ]

(@v1.8) pkg> activate @pysr
  Activating new project at `~/.julia/environments/pysr`

and manual call:

julia> Pkg.activate("@pysr")
  Activating new project at `~/Documents/PySR/@pysr`

[MilesCranmer]

This seems like it shouldn't be the expected behavior so opened an issue on Pkg.jl: JuliaLang/Pkg.jl#2983

[MilesCranmer]

Nevermind, it's literally just Pkg.activate("pysr", shared = true)!

[MilesCranmer]

Okay, the environment is now given by .julia/environments/pysr-{__version__}, rather than created as a temp directory. This is much cleaner in my opinion. Thanks for the idea @mkitti!

[mkitti]

With the conda-forge version, .julia should be a folder located within the $CONDA_PREFIX.

[MilesCranmer]

Great!

I think the shared = true will automatically find that depot in conda. Let me check.

[mkitti]

Yes, it should. We set it up for that purpose.