That's all I'm getting done for now, will probably have to fix more tests. Definitely needs updated whatsnew docs, because this is a more impactful change than just having imports succeed with relative paths in sys.path, but I think it's for the best.

Do you think changes to ntpath should be split off as a separate bpo issue that's just for 3.10? I think some developers want the current behavior that classifies "/spam" as absolute, so, for future reference, it would be good to have an issue that's more focused on that change.

For resolving bpo-43105 in 3.8, nt._path_splitroot can use PathSkipRootW. It supports rooted paths, absolute drive paths, and UNC paths, including device paths such as "\\?\C:\", "\\.\DEVICE\", and "\\?\UNC\server\share\". One caveat is that PathSkipRootW returns NULL for drive-relative paths such as "C:spam". That case is simple enough to handle in C to make it consistent with PathCchSkipRoot.

Git is getting really messed up here, I'm going to do a reset and rebase to make sure I only have my changes in here.

Assuming all the tests pass, this change doesn't affect ntpath and so can be backported. Then I can easily add the ntpath changes back separately (later, because they rely on the changes in here).

The Azure Pipelines failure is the flaky asyncio test, so unless @eryksun spots something in the next hour or so, I'm going to hit merge.

Thanks @zooba for the PR 🌮🎉.. I'm working now to backport this PR to: 3.8, 3.9.
🐍🍒⛏🤖

Sorry, @zooba, I could not cleanly backport this to 3.9 due to a conflict.
Please backport using cherry_picker on command line.
cherry_picker 04732ca993fa077a8b9640cc77fb2f152339585a 3.9

Sorry @zooba, I had trouble checking out the 3.8 backport branch.
Please backport using cherry_picker on command line.
cherry_picker 04732ca993fa077a8b9640cc77fb2f152339585a 3.8

GH-25237 is a backport of this pull request to the 3.9 branch.

@zooba It is a shame there are no direct tests of os._path_splitroot, and no good documentation what the code is doing. What does "Removes everything after the root on Win32" mean? What is defined as "the root"?

Adding a new method to os in this way makes it hard for alternative interpreters like PyPy to catch up, especially since the function is required in order to bootstrap the interpreter.