Critical failure with `jupyter-core` 5.8.0 on Windows

[krassowski]

Description

Seen on jupyter/notebook#7653

jupyter_server/jupyter_server/extension/serverextension.py

Lines 22 to 42 in 4ee6e1d

	def _get_config_dir(user: bool = False, sys_prefix: bool = False) -> str:
	"""Get the location of config files for the current context
	Returns the string to the environment
	Parameters
	----------
	user : bool [default: False]
	Get the user's .jupyter config directory
	sys_prefix : bool [default: False]
	Get sys.prefix, i.e. ~/.envs/my-env/etc/jupyter
	"""
	if user and sys_prefix:
	sys_prefix = False
	if user:
	extdir = jupyter_config_dir()
	elif sys_prefix:
	extdir = ENV_CONFIG_PATH[0]
	else:
	extdir = SYSTEM_CONFIG_PATH[0]
	return extdir

assumes that SYSTEM_CONFIG_PATH is not empty, however it was allowed to be empty and now by default is empty on Windows starting with 5.8.0 release.

Reproduce

• Run on Windows with jupyter-core 5.8.0
• Run jupyter labextension list

Expected behavior

No error

Context

None

[minrk]

This can be fixed by using ENV_PREFIX again if system path is empty.

[minrk]

We could also protect against this by putting env prefix in system path in Jupyter-core, ensuring its non-empty, even though we've never promised this. Clearly it is assumed.

[minrk]

I think jupyter/jupyter_core#430 will fix this

[minrk]

Fixed by jupyter-core 5.8.1, but I think it's also appropriate here to check if SYSTEM_CONFIG_PATH is empty and handle that.

[krassowski]

Should we yank jupyter-core 5.8.0?

[minrk]

We could. I'm not sure there's a point, since 5.8.1 is out. To me, yanking is only for latest releases when a fix is not ready yet and can't be waited for.