[Python] pathlib: Improve compatibility with python

[acampove]

Check duplicate issues.

• Checked for duplicates

Description

I see all over the place problems when I try to use ROOT with python. This time I see that there is no support for Path objects.

Reproducer

import ROOT
from pathlib import Path
import numpy as np

root_file = ROOT.TFile("example_data.root", "RECREATE")
tree      = ROOT.TTree("myTree", "Example Tree")

x = np.zeros(1, dtype=float)
y = np.zeros(1, dtype=float)
z = np.zeros(1, dtype=float)

tree.Branch("x", x, "x/D")
tree.Branch("y", y, "y/D")
tree.Branch("z", z, "z/D")

for i in range(10):
    x[0] = i
    y[0] = i * 2 + np.random.normal(0, 1)
    z[0] = x[0]**2 + y[0]
    tree.Fill()

tree.Write()
root_file.Close()

file_path = Path("example_data.root")

df = ROOT.RDataFrame("myTree", file_path)

ROOT version

6.36.04

Installation method

mamba

Operating system

almalinux 9

Additional context

Any good out there uses Path, not strings when handling paths.

[acampove]

Another place where we need this changed is:

RDF.Experimental.FromSpec
TFile

and I believe it must be like this all over the project, everytime we use a path.

[guitargeek]

Excellent suggestion! We never thought about that so far, but this Pythonization makes complete sense because with std::filesystem::path, there is a corresponding C++ type that can map to the Path from pathlib. So we can support this without having to write pythonizations for each class and function that accepts a file path!

The way forward here would be to implement a custom cppyy converter for std::filesystem::path from pathlib.Path, and then we "just" have to change out C++ interfaces to use std::filesystem::path const & instead of generic strings for function arguments that represent file paths. That would also improve the C++ interface in general, because it will be more explicit and typesafe. Win-win!

I presume we could work on this early next year only, because we are already about to do a large upgrade of the implementation details of the ROOT Python bindings for ROOT 6.40. Then, ROOT 6.42 would be a good next target for implementing new pythonization-related features like this one, if we don't find the time to do it earlier.

Here is an example of how this interface change would look like:

• [RF] Use std::filesystem::path in RooWorkspace #20495

PS: I would not call this a bug, but a missing feature 🙂