Use custom devcontainer images with unexpected python instalations

[DiamondRC]

The uv setup relies on the standard devcontainer image which comes with the repo. If attempting to switch image, the default uv integration fails.

Resolved by clearing existing uv setup and installing fresh, then using $ uv venv --seed --system-site-packages to create the venv. Run programs with $ uv run python /workspaces/path_to_file/file.py.

[coretl]

So the only things I imagine you need to do are:

• Ensure that uv is installed in the container in some way
• Remove .python-version as you want to follow the version of python installed on the system, not pin it in the repo
• Add uv venv --seed --system-site-packages && to the start of the postCreateCommand
• Then the venv should be activated by default and you can just python /workspaces/path_to_file/file.py

Does that work for you? If so this can be expanded into a how-to guide

[DiamondRC]

Correct about removing .python-version, though I believe this is only required if you want to use a different python version, rather than just re-configuring uv.

Using the suggested uv venv --seed --system-site-packages && prompts the user to explicitly clear existing .venv's upon container creation, adding the '--clear' flag makes this automatic.

I'm now using "postCreateCommand": "cd /workspaces/deltabot_nn_controller && uv venv --seed --system-site-packages --clear && pip install pre-commit && pre-commit install --install-hooks"

I believe you will then need to run the file as uv run python /workspaces/path_to_file/file.py instead of just python /workspaces/path_to_file/file.py if you'd like to use the new .venv.

There's some other bits and pieces I needed to do for my exact use case (different image, use system packages, expose CUDA), so there's possibly some entanglement between that and a minimal setup.

[coretl]

I believe you will then need to run the file as uv run python /workspaces/path_to_file/file.py instead of just python /workspaces/path_to_file/file.py if you'd like to use the new .venv.

But we already set the path here:

python-copier-template/.devcontainer/devcontainer.json

Line 19 in dc6564e

"PATH": "/cache/venv-for${localWorkspaceFolder}/bin:${containerEnv:PATH}"

so it should pick the correct environment up from that

[DiamondRC]

I think you're right, I have "PATH": "${containerEnv:PATH}:/usr/local/cuda/bin", at the moment to detect the CUDA libraries.