Currently by creating the entire the entire domain name, our way of constructing urls is not compatible with a bunch of proxy setups.
Because we're not changing domain/origin with our links we can just use paths to define our redirect location and our api endpoint for the POST.
This switches us to use that logic.
As I went through making that change I discovered that our tests were not actually testing appropriately for the way that base_url and default_url on the Jupyter server work.
Because our tests would break when I switched us to using paths, I added the logic needed to match the behaviour expected by the Jupyter server & the Tornado application.
This primarily consisted of removing any cases where self.base_url did not begin and end with "/". That required
Currently by creating the entire the entire domain name, our way of constructing urls is not compatible with a bunch of proxy setups.
Because we're not changing domain/origin with our links we can just use paths to define our redirect location and our api endpoint for the POST.
This switches us to use that logic.
As I went through making that change I discovered that our tests were not actually testing appropriately for the way that
base_url
anddefault_url
on the Jupyter server work.Because our tests would break when I switched us to using paths, I added the logic needed to match the behaviour expected by the Jupyter server & the Tornado application.
This primarily consisted of removing any cases where
self.base_url
did not begin and end with "/". That requiredbase_url="/"
to all of our test config (to match the code in notebook/notebookapp.py#L1048-L1062base_url
that did not begin and end with "/".