Detecting Python.h when using an uv venv to setup a project

[jacopoabramo]

Sorry for the probably very noobish question but it's my first time using nanobind in a project together with CMake (and also am very rusty with CMake into VScode itself).

I'm using VScode and developing in an uv-generated virtual environment to setup my project. nanobind is installed via

uv pip install nanobind

To tell VSCode where to find the headers I'm setting the c_cpp_properties.json file as follows:

{
    "configurations": [
        {
            "name": "Win32",
            "includePath": [
                "${workspaceFolder}/.venv/Lib/site-packages/nanobind/include",
            ]
        }
    ],
    "version": 4
}

This is a trimmed-down version but it just to show it correctly finds nanobind.
The thing I'm missing is how to tell VSCode where to find the Python header files. I tried telling VSCode to use CMake to detect where the files are via:

{
    "cmake.configureOnOpen": true,
    "C_Cpp.default.configurationProvider": "ms-vscode.cmake-tools"
}

But that didn't help: VSCode still can't find Python.h. Any suggestion?

[nicholasjng]

Hey! Although it does not use vscode's CMake integration, there is an example project that shows how to set up CMake to build nanobind extensions with scikit-build-core: https://github.com/wjakob/nanobind_example (in short: it uses FindPython)

Otherwise, to get Python's include path, you can use python3 -c "import sysconfig; print(sysconfig.get_path('include'))", which should be static unless you change your venv's Python version frequently.

[jacopoabramo]

Otherwise, to get Python's include path, you can use python3 -c "import sysconfig; print(sysconfig.get_path('include'))", which should be static unless you change your venv's Python version frequently.

@nicholasjng thanks, I did exactly that. I also understood that for my case using ms-vscode.cmake-tools was the wrong choice because I simply needed to see the source code of the included libraries and I didn't need cmake for that. But thanks for writing it, I'm marking it as solution in case others encounter my same situation.

[jacopoabramo]

@nicholasjng by the way the problem with the first approach is that uv virtual environments are not shipped with the include folders, but they're cached somewhere in a user directory. I'm not exactly sure who would be responsible to adjust this (if nanobind or scikit-build-core) but it would be nice to support the detection out of the box with the first approach

EDIT: example on Windows using a virtual environment created via uv

>>> python3 -c "import sysconfig; print(sysconfig.get_path('include'))"
C:\Users\jacop\AppData\Roaming\uv\python\cpython-3.10.15-windows-x86_64-none\Include

[nicholasjng]

Oh, interesting. Thanks for sharing! This might really be a CMake / scikit-build-core issue then.