BUG: uv tree --invert --package ... sometimes also displays packages I didn't ask for

[neutrinoceros]

Summary

When I run uv tree --invert --package ... against astropy, I sometimes get more packages to show up in the displayed tree than I asked for, which seems like a bug.
This is obvious when requesting a single package as root and limiting the depth to 0.

❯ git clone https://github.com/astropy/astropy
❯ cd astropy
❯ uv tree --invert --package setuptools -d 0 --no-config
Resolved 166 packages in 8ms
asdf-astropy v0.7.0
setuptools v75.8.0
❯ uv tree --invert --package six -d 0 --no-config
Resolved 166 packages in 8ms
asdf-astropy v0.7.0
six v1.17.0

In all cases I found that reproduce this problem, asdf-astropy seems to be the one package that show up uninvited. However, it may or may not show up depending on the package I ask for:

❯ uv tree --invert --package numpy -d 0
Resolved 166 packages in 1ms
numpy v2.2.2

The output seems consistent (no randomness I can see) for any given command, however I don't have a clear idea why --package numpy works and --package six doesn't. My experiments suggest it has nothing to do with numpy being a direct and hard dependency to astropy, as I also hit the bug with --package astropy-iers-data

❯ uv tree --invert --package astropy-iers-data -d 0
Resolved 166 packages in 1ms
asdf-astropy v0.7.0
astropy-iers-data v0.2025.2.3.0.32.42

I wasn't able to reproduce this with a simpler package, though I admit I only tried a bottom-up approach; I did not attempt to simplify astropy's dependencies.

Platform

macOS 15.3 arm64

Version

uv 0.5.29 (ca73c47 2025-02-05)

Python version

Python 3.12.9

[charliermarsh]

Thanks, we'll take a look.

[charliermarsh]

I think this is because there's a circular dependency in the graph, whereby asdf-astropy depends on astropy, but astropy depends on asdf-astropy -- at least, as far as I can tell? So, like, it's trying to show you that setuptools is required by astropy-sphinx-theme which is required by sphinx-astropy which is required by astropy, but then astropy is both required by and requires asdf-astropy, which is hard to display.

[neutrinoceros]

asdf-astropy depends on astropy, but astropy depends on asdf-astropy -- at least, as far as I can tell?

I can confirm it does, and sorry for not thinking of this earlier !

I think this is because there's a circular dependency in the graph (...) which is hard to display.

Granted. Maybe the cycle could be indicated as ..., inspired by the Python repr for self-referential objects, as in

>>> L = []
>>> L.append(L)
>>> print(L)
[[...]]

?

[JadenMajid]

Hi, can i work on this?