uv init --build-backend scikit followed by uv run --with jupyter jupyter lab has counterintuitive behaviour concerning rebuilds

[rimathia]

Question

This is probably just a documentation request. My first step is
uv init --build-backend scikit
which works as expected, next I'm starting a jupyter notebook using
uv run --with jupyter jupyter lab
which also works as expected, it builds the C++ extension and I can now create a notebook (using the default kernel) where I can load the C++ python extension. This is a massive improvement over any other tool or workflow that I've tried.
However, I haven't been able to find out the right way to invoke uv to create a jupyter kernel such that it rebuilds the C++ extension if the source files have changed. One way which does work is
uv clean && rm -r .venv && uv run --with jupyter jupyter lab
but that is certainly not the way to do it.

Maybe the uv init --build-backend scikit could document the right invocations (for example in the README.md which it creates which is empty by default) or it could print the two or three most frequent workflow commands to the command line after it has created the environment.

Platform

Darwin 24.5.0 arm64

Version

uv 0.7.19 (38ee6ec 2025-07-02)

[konstin]

Are you looking for the --refresh-package option?

[rimathia]

Thanks, --refresh-package doesn't recompile the C++ code, but --reinstall does. Given how fast the package reinstallation is, that is perfecly fine for a edit-compile-test cycle with a jupyter notebook.
I'll close this. There are certainly good reasons why a change to the pyproject.toml triggers changes to the python environment for other commands than changes to the python extension source.

[konstin]

The reason is mostly that we know that pyproject.toml files should rebuild, but we don't know which other files are relevant. You can control this behavior with https://docs.astral.sh/uv/reference/settings/#cache-keys.

[rimathia]

Thanks for the bonus tip, using
[tool.uv] cache-keys = [{ file = "pyproject.toml" }, { file = "requirements.txt" }, { file = "src/**.*" }]
I get exactly the behaviour that I intuitively expected: every change of a source file triggers a rebuild of the python extension.