file_to_run and notebook_dir would collide, with possible dictionary
ordering randomness depending on how they were passed.
With this change, the default value of notebook_dir is pulled from
file_to_run, instead of pushing the value when file_to_run is changed.
This makes it possible to specify both and have the server behave as
expected.
- start with no number (Untitled0 -> Untitled.ipynb)
- copy of copy increments instead of adding another `-Copy` (copy Foo-Copy1.ipynb gives Foo-Copy2.ipynb, not Foo-Copy1-Copy1.ipynb)
- copy file to new folder starts with the original filename, instead of unconditional `-Copy0`
- various docstrings, comments clarified and updated
- misc typos
- fix and test creating an untitled directory via POST
- only define `message` if there's something to say
- path is full path (including name)
- name is only used for read-only convenience, and plays no role in any API
- remove 'untitled upload' (POST with content and no path)
store the Future for the initial request,
allowing subsequent requests to wait on the same pending reply.
Previously, any incoming requests that arrived while waiting for the first reply would send their own request.
- add debug statement at the very beginning of a web socket request
- add debug statement in open, indicating that the connection has been accepted
- add timeout, so failed or slow kernel_info doesn't cause the request to never get a response
- don't send kernel_info_request before authenticating the request
The last one required some icky coroutine shenanigans,
because of our subclass structure, but it should work fine.
WebSocket connections will not open until kernel_info is retrieved,
removing a race condition waiting for the reply to indicate adaptation,
which could result in a v5 message being sent to a v4 kernel.
The reply is cached, so that it need not be requested on each new connection.
