Executables installed via `uv` cannot be run from a subprocess on Windows

[AlexWaygood]

Minimal repro in PowerShell:

PS C:\Users\alexw> uv venv env
Using Python 3.12.1 interpreter at C:\Users\alexw\AppData\Local\Programs\Python\Python312\python.exe
Creating virtualenv at: env
PS C:\Users\alexw> env\scripts\activate
(env) PS C:\Users\alexw> uv pip install black
Resolved 7 packages in 395ms
Downloaded 1 package in 218ms
Installed 7 packages in 106ms
 + black==24.2.0
 + click==8.1.7
 + colorama==0.4.6
 + mypy-extensions==1.0.0
 + packaging==23.2
 + pathspec==0.12.1
 + platformdirs==4.2.0
(env) PS C:\Users\alexw> python -c "import subprocess; subprocess.run(['black', 'coding/typeshed'])"
C:\Users\alexw\AppData\Local\Programs\Python\Python312\python.exe: can't open file 'C:\\Users\\alexw\\black': [Errno 2] No such file or directory

The same commands work fine if I use the stdlib venv module to create the virtual environment and I use pip to install black. Invoking black directly from the command line seems to work fine; it's only invoking it via a subprocess that doesn't seem to work.

A consequence of this bug seems to be that it's impossible to run coverage -m pytest on Windows inside a GitHub Actions workflow if you've installed coverage using uv. Here's my attempt at switching typeshed-stats to use uv:

• https://github.com/AlexWaygood/typeshed-stats/pull/189

CI is now green everywhere, except on Windows, where invoking coverage -m pytest fails with the message:

 Error: The system cannot find the file specified.
 (from CreateProcessA(null(), child_cmdline.as_mut_ptr(), null(), null(), 1, 0,
    null(), null(), addr_of!(*si), child_process_info.as_mut_ptr()))

(Running python -m coverage works, but one test in my suite still fails, because I have a test that asserts that running typeshed-stats as a standalone executable works, and I invoke the standalone executable using a subprocess for that test...)

[MichaReiser]

I think that should be fixed by https://github.com/astral-sh/uv/pull/1523

[AlexWaygood]

My local repro is now fixed -- thanks! However, I'm still seeing the strange errors over at https://github.com/AlexWaygood/typeshed-stats/pull/189 (which is now using uv==0.1.4). I thought those errors had the same cause as the local repro I posted above, but perhaps not.

[AlexWaygood]

Since the original repro at the top of this issue is now fixed, I'll close this and open a new, more targeted issue

[AlexWaygood]

This is the new issue:

• #1639