Dev: PyData 2012 HackNight
Note this was originally a gist, moved here for reference.
security discussion (I missed the beginning of this -pi)
jason chin: it's be nice to have multiple users for
the notebook - dual-notebook discussion
david strauss: git has functionality to be file-level aware - and have it ignore portions of a file - git config - clean and smudge filter mechanisms (can be file type aware)
brian: this is crazy! no one else does this (generate rich text, embedded videos and html, generated plots, AND version control friendly)
"save stripped" for VCS-friendly notebooks - can't be default - we need In[] prompt numbers
separation between things the user
there's an sqlite database for all sent code
Format we have we keep identical - and we create a new thing that is the stripped version of it
brian: notebook save action has a hook system - that triggers things when it is called (that maybe incorporated the git clean/smudge stuff)
fernando: pipeline maybe too much to swallow for "regular users"
chain of filters / different renderers - common options are available - scripts, stripped (raw) scripts - code only, transformed ipython syntax or not, version-control friendly of the notebook, ReST. others provided by exposing the chain of filters functionality
fernando: github - getting meaningful diffs of notebooks on github (like the image diff they already have, but for whole notebooks - with different types of cells of the notebook properly rendered.
jason: teach git about cells - make each object a cell?
we can have an 'nbdiff' command - your own little thing - which uses git under the hood - or just compares two separate notebook files.
john: one way to do it would be just to do the diff on the exported clean py files.
sharing of documents: - we should just use git and github for that - save on github and a share - gist and full github for full project (a directory with .ipynb and also data csv)
brian: don't want to create a git ui on the web - github has one:
paul: but, github web ui: should be able to make new files and folders in git
david: can have separate git driver for diff - can be "lossy" - not as precise
david: you want to have both - a local good representation for diffing locally (don't want to have to push to github to get this), and a github friendly one - for sharing and collaborative
hook up "terminal" tab - or terminal cell type - only for the shell - they
don't get saved - and the ! is prepended to everything that happens there.
this is a nice entry point for other people.
brian - slippery slope - xcode has git integrated everywhere, will want more functionality, full color on the diffs, etc.
fernando: have an upload magic - "put" so that in lab environments - they don't have to scp - and then they can have their full screen codemirror tab, too. notebook could have a filesystem view so they can grab files from the kernel using notebook.
brian: very easy to do pam authentication in python. but there's this company in redmond.... punt on it.
brian: back to multi-user environment - a hotel/motel is not a single family dwelling. the single user experience of the notebook would go away - don't think it's possible to have both. (who will maintain it? you can't handle this logic in subclassing) all urls : for each - go write get, post, etc. - for each request I write a request in tornado, with each having a regular expression that will have the things that it needs to have encoded in it.
fernando: job of applications is different - one is the hotel, the other is just the hotel room. - start up a new single-user instance for each new instance. it's just so the user doesn't have to type at the prompt "ipython notebook" - one http port per user advantages: - simple, and - preserves the local, application-like thing we've got right now. - and possibility of someone taken the current experience and reuse it for something else altogether. what do we lose if we do this: - the notebook itself won't we multi-user aware. - can't do any sharing that isn't opaque to the notebook. - can't add anything to the notebook ui - port management thing is a huge concern.
brian: implementation side: for the req handlers - reg ex groups capture the parts you'd like to match - those get passed as arguments. If you subclassing - the signature of your methods change turn off authentication / specify list of users that can access it the url scheme will look more complicated
Names correspond to people who kicked off particular questions or points, not everything that is stated is attributable to the name at the beginning of a given snippet. Also putting this on Gist so that others can clarify and add information that I missed or mis-captured.