Project dependencies no longer available in JupyterLab launched with `uv run --with jupyter jupyter lab`

[bellkev]

Summary

My understanding from the uv + Jupyter integration doc is that running JupyterLab without adding it to the project dependencies like uv run --with jupyter jupyter lab should make the project's dependencies available in the JupyterLab environment. I see that that works in 0.7.22 but not in 0.8.0. I strongly suspect that the change comes from #14447.

As a minimal example, if I run:

curl -LsSf https://astral.sh/uv/install.sh | sh
uv python install 3.10
uv init --bare --python 3.10
uv add boto3
uv run --with jupyter jupyter lab

Then with 0.7.22 I'm able to run import boto3 successfully in a python cell in a notebook, but not in 0.8.0. In version 0.8.0, when I run the following in a cell:

import sys
print(sys.executable)
print(sys.path)

I get:

/Users/kevin/.cache/uv/archive-v0/clAeKvr1eBVfjXfG8R8TX/bin/python
['/Users/kevin/.local/share/uv/python/cpython-3.10.18-macos-aarch64-none/lib/python310.zip', '/Users/kevin/.local/share/uv/python/cpython-3.10.18-macos-aarch64-none/lib/python3.10', '/Users/kevin/.local/share/uv/python/cpython-3.10.18-macos-aarch64-none/lib/python3.10/lib-dynload', '', '/Users/kevin/.cache/uv/archive-v0/clAeKvr1eBVfjXfG8R8TX/lib/python3.10/site-packages']

While if I run a command like:

!python -c "import sys; print(sys.executable); print(sys.path)"

I get:

/Users/kevin/.cache/uv/builds-v0/.tmp2PZpfe/bin/python
['', '/Users/kevin/.local/share/uv/python/cpython-3.10.18-macos-aarch64-none/lib/python310.zip', '/Users/kevin/.local/share/uv/python/cpython-3.10.18-macos-aarch64-none/lib/python3.10', '/Users/kevin/.local/share/uv/python/cpython-3.10.18-macos-aarch64-none/lib/python3.10/lib-dynload', '/Users/kevin/.cache/uv/builds-v0/.tmp2PZpfe/lib/python3.10/site-packages', '/Users/kebel/source/uv-test/uv-repro/.venv/lib/python3.10/site-packages', '/Users/kevin/.cache/uv/archive-v0/clAeKvr1eBVfjXfG8R8TX/lib/python3.10/site-packages']

The latter contains the desired .../uv-repro/.venv project venv, I believe via a .pth file where the ephemeral .cache/uv/builds-v0/.tmp2PZpfe/bin/python is installed.

I believe the intention is for the kernel to use the same "builds-v0" ephemeral python interpreter location which has the overlay .pth file rather than the "archive-v0" --with python environment. I see that the kernelspec looks like:

{
 "argv": [
  "python",
  "-m",
  "ipykernel_launcher",
  "-f",
  "{connection_file}"
 ],
 "display_name": "Python 3 (ipykernel)",
 "language": "python",
 "metadata": {
  "debugger": true
 }
}

The jupyter_client logic is that the "python" command is interpreted as sys.executable, and judging by the output of uv run --with jupyter jupyter troubleshoot, that is the --with environment (/Users/kevin/.cache/uv/archive-v0/clAeKvr1eBVfjXfG8R8TX/bin/python in my case). (The !python and similar commands have the ephemeral env in the $PATH env var though, and thus seem to use the right python environment.)

Platform

macOS 15.5 (also tested in Amazon Linux 2023 and debian:bookworm docker image)

Version

0.8.0

Python version

3.10.18

[charliermarsh]

Hmm yeah, I think this is because jupyter is in the --with environment rather than the environment that we invoke from (which has the requisite .pth file pointing to both the --with environment and the base environment).

[AlexWaygood]

This also appears to have broken mypy invocations such as uv run --with=mypy mypy FYI.

(I know the way you're "supposed" to do this is by listing mypy as a dev dependency, but this is often the way I invoke mypy in practice.)

[charliermarsh]

That one I assume would be fixed by rewriting the shebangs in the final environment.

[zanieb]

Charlie's referring to #13327 which was a bug in the previous implementation as well. Unfortunately that breaks Jupyter.