A few points to consider in review:

• Do we want to backport this for a 1.4.x release, or just release 1.5 (Python 3.8 has been dropped)
• Do we want to add --reinstall for uv (current state)?

I'm alright to release a new version without backporting, personally.

I'm not sure about this. I would prefer to drop --ignore-installed than muddy the interface with a parameter that'll probably see no use outside of pyodide. Unless a good solution for PYTHONPATH can be found, I would rather simply mention it as a caveat in the docs. The DefaultIsolatedEnv could also notify the user that dependencies might fail to install if it detects that PYTHONPATH is set.

This does drop --ignore-installed from programmatic use (unless you know about this parameter), and makes sure only the CLI uses it, where we do control what is happening. I feel like it's a good balance between keeping python -m build working and keeping third party usage of our API stable.

How about we make this parameter private for now, and only make it public if someone needs it? That gives us the ability to solve this another way in the future, while still fixing pyodide and keeping previous fix (which seems to be working, at least no one has complained).

How about we make this parameter private for now, and only make it public if someone needs it? That gives us the ability to solve this another way in the future, while still fixing pyodide and keeping previous fix (which seems to be working, at least no one has complained).

That's a good idea - let's do it that way.