Possibility to specify the virtual environment to use

[cerna]

Hello,
I noticed that in file .pdm.toml there is specified the exact path of Python interpreter to use. I am thinking if there could be also an exact path to virtual environment to use for build and install? As there already can be use_venv = true configuration.

My use case is a bit particular:
I have a project (repository) assembled as a workspace. (In other words the modules are acting as if were up to a point independent.) The project is build with CMake and written in C, C++, Bash and Python. The Python part is comprised from extension modules for CPython wrapping calls to other C parts and pure Python modules building on top of the extensions. So everything has to be build in order and have dependencies on previous builds.

I would like for CMake to copy the Python .py files to BINARY tree (CMake has strong distinction between SOURCE and BINARY trees), template the pyproject.toml there (set version, description etc) and build and install the package using specific virtual environment so all packages (distributions really) are using the same virtual environment and thus have the dependencies already present. You would be then able to use these "installs" after activating the virtual environment.

I don't have strong opinion if it should be virtual environment or the __pypackages__, but the later would probably fail on each package being in a different folder.

[frostming]

If I understand your request correctly, it is doable with PDM:

https://pdm.fming.dev/usage/project/#working-with-a-virtualenv

[cerna]

Thank you, I didn't realize the implications of

Besides, when use-venv is on and the interpreter path given is a venv-like path, PDM will reuse that venv directory as well.

and then come upon it by accident. It's great it is not just a temporary by-product, but a feature!

Also, when I do the pdm install, it will install the package in development/editable mode and create these *.egg-link files back to the code. You can then see the package with path when running pip list. However, when I try to do pdm install to the same virtual environment in other Python project which depends on the first one, it cannot resolve the dependency. But if I install this package as built wheel via pip install dist/<name>.whl, then everything works, i.e. pip doesn't care about the dependency being installed in developer mode. Can I somehow get the same resolution with PDM?

(BTW, Poetry is the same, even though it is creating a *.dist-info folder with metadata when running the poetry install command.

[frostming]

Can you clarify your case with detailed steps to reproduce?

[cerna]

I created a simple example in this repository. The detailed steps are in README.

[frostming]

Because PDM requires locked candidates before installation. For locking, the package must actually exist on whatever index you choose(default: PyPI)

But pip install only performs installation and checks against the currently installed packages. So it can be installed successfully. This is a difference in philosophy and I am afraid PDM can't achieve the same as pip.