When python-mode is activated for a buffer, at lines [25829-25738](https://gitlab.com/python-mode-devs/python-mode/-/blob/master/python-mode.el#L25830) the `py-complete-function` option doesn't add as `local`, which can interfere with non-related buffer's capfs. (this looks like a regression from [6.2.3](https://gitlab.com/python-mode-devs/python-mode/-/blob/6.2.3/python-mode.el#L25917) ) if I then set `py-complete-function` to be `nil` (the [default](https://gitlab.com/python-mode-devs/python-mode/-/blob/master/python-mode.el?ref_type=heads#L1027) specified in it's declaration), then cond forces `py-shell-complete` to be added, regardless of `py-do-completion-p`s value, so capf-functions still gets polluted. If capf is triggered by [py-indent-or-complete](https://gitlab.com/python-mode-devs/python-mode/-/blob/master/python-mode.el#L21992), because there isn't a comint process associated with a normal python buffer, the first completion condition fails, but the second one will pass. This condition doesn't wrap `completion-at-point` with `ignore-errors`. This results in the windows and buffers changing inexplicably when I press tab. I'm on a weird frankenstein of 28.2 and `doom` 3.0 with stock python 0.28 being use-packaged activating your python-mode 6.3.1, but that shouldn't be changing the core of what's happening above.