Username collision

[Ph0tonic]

Hi,
Thank you very much for your work.
As previously discussed here (jupyterhub/oauthenticator#459 (comment)), we have a username collision issue to manage.

I mainly see 2 ways of managing it.

1. Handle it in Jupyterhub
2. Add a prefix in the authenticator's wrapper

I managed to make the second option work and opened a PR to comment #6.
For this second solution, I saw 2 main options to implement the fix, first by overriding the function normalize_username. However, this function might be called multiple times due to the implementation inside OAuthenticator. Therefore, we can't directly add a simple prefix. We should first check if a prefix exists and be 100% sure that it's indeed a prefix. The second solution was to override authenticate, check_allowed and check_blocked_user to have the prefix added or removed.

The drawback of the prefix is that it is displayed on jupyterhub page in the top right corner with the prefix.

I had a look at how to handle the name collision in Jupyterhub directly and if we want to do that, I think that we should change the database to store in the user table an additional field named authenticator. And during the authentication, instead of returning a single username or an object with 2 fields name and admin (to be check). We could also return an additional field named authenticator.

Jupyterhub code to check user duplication :
https://github.com/jupyterhub/jupyterhub/blob/29e954c407c017c045fc534410ffcd1ec6318727/jupyterhub/apihandlers/users.py#L333

[manics]

I had a look at how to handle the name collision in Jupyterhub directly and if we want to do that, I think that we should change the database to store in the user table an additional field named authenticator

This change would have to cascade down to everywhere a unique identifier is used, e.g. http://<jupyterhub>/user/<authenticator-prefix><username>/, and would therefore be a big breaking change for other authenticators and spawners. What do you see as the benefits of doing it in JupyterHub instead of in this authenticator?

[Ph0tonic]

This change would have to cascade down to everywhere a unique identifier is used, e.g. http://<jupyterhub>/user/<authenticator-prefix><username>/, and would therefore be a big breaking change for other authenticators and spawners. What do you see as the benefits of doing it in JupyterHub instead of in this authenticator?

Hum, the main advantage of being in Jupyterhub is that it would not require any other library and would therefore be an official feature. But I agree that it might have deep impacts if the id used is the username, we would be a breaking change to generate a proper id and it would be much easier to use a simple prefix for now.

[sgaist]

Fixed by #6