PySR should check SymbolicRegression.jl version

[MilesCranmer]

SymbolicRegression.jl now has a PACKAGE_VERSION constant – this should be used by PySR to verify that the correct version is being imported. In practice, this probably not needed since PySR is managing the installs, but a user might have SymbolicRegression.jl installed in their base env, which might get imported instead. better to verify the version explicitly.

[tecosaur]

Why don't you check the version using Pkg ?

[MilesCranmer]

They are only adding it to Julia 1.9: JuliaLang/julia#45607. For LTS versions I think I need this constant parameter. What do you think?

[tecosaur]

Hmm, how long has Pkg.Types.read_package been around for? If that's somewhat stable you could do Pkg.Types.read_package(abspath(pathof(SymbolicRegression), "..", "..", "Project.toml")).version

[MilesCranmer]

Here’s the current approach: https://github.com/MilesCranmer/SymbolicRegression.jl/blob/cce9fda3d1fc6493b9d41ee92c6c630cb1944b95/src/SymbolicRegression.jl#L114. I guess the question is whether you’d want to do this inside PySR or SymbolicRegression.jl (and then just check the constant).

[tecosaur]

I'd be tempted to suggest this should be checked on the python side, since Julia already has its ways of dealing with version requirements.

[MilesCranmer]

Implemented a while back so closing. Cheers!