Status update: I'm working on nbconvert support. Investigating using require.js's optimizer.

Status update: r.js optimization success. Both embeded and external widget JS works with nbconvert output.

Status update: The last commit adds a new fab js command.

Status update: Widgets will now reconnect on page refresh.

So if there is a kernel running, and you refresh the page, it will reconnect with the currently-running kernel? Could there be missed messages? What happens if there were missed messages?

So if there is a kernel running, and you refresh the page, it will reconnect with the currently-running kernel? Could there be missed messages? What happens if there were missed messages?

Hi @jasongrout ! Very glad you came to check-in on this. I was going to ping you for your feedback after making another review pass- I don't want to break any of your single cell server stuff and I value your feedback with the widget stuff because it has helped a lot so far!

Yeah, that's how it's intended to work. Messages shouldn't be lost- I've added a closure that is created if the underlying socket hasn't been created yet. If that's the case, the closure(s) get called when the sockets are reconnected.

If the kernel is new, i.e. a kernel restart, I don't attempt to create any widgets in the back-end. So if you save your notebook with widgets, close ipython, start ipython, and open the same notebook, the widgets will display (with "disabled" = true) but any messages at that point will be discarded (an error will appear in the sys console saying that comm id blah blah doesn't exist).

closure here - https://github.com/ipython/ipython/pull/5900/files#diff-d66a08bcdaaf950c48f42fb08f5bc189L155

Thanks; I've been keeping a close watch on this, though I haven't been looking at the code. I hope to soon, though.

@ellisonbg when it's convenient, could you post the text from our in person review here?

notebook.js

• The disabled widgets should also show up if a notebook is loaded, but no
  kernel is started. This state will be possible in IPython 3.0 when users
  are selecting which kernel to run. Right now, this state is sort of probed
  when a kernel is restarting.
• Think about renaming WidgetManager.get_model to reflect that it will
  create a model too.
• I am not sure we have a consistent way of naming private methods in
  notebook.js: _load_widgets and
  _save_widgets.
• _load_widgets should also work if this.kernel is null.
• Look into creating WidgetManager.toJSON/fromJSON and then use that from
  Notebook.
• this.widgets -> this.widgets_data

codecell.js

• this.saved_widgets -> this.widgets_data
• CodeCell._load_widgets should work without kernel.
• Naming of methods with leading underscores
• this.widgets -> this.widget_ids or this.widget_model_ids

On Thu, Jun 5, 2014 at 4:55 PM, Jonathan Frederic notifications@github.com
wrote:

@ellisonbg https://github.com/ellisonbg when it's convenient, could you
post the text from our in person review here?

—
Reply to this email directly or view it on GitHub
#5900 (comment).

Brian E. Granger
Cal Poly State University, San Luis Obispo
@ellisonbg on Twitter and GitHub
bgranger@calpoly.edu and ellisonbg@gmail.com

TODO: #5997

It would probably make sense to not disable widgets when the comm is closed, but notify the user via a broken link icon or something like this. Indeed, in the case when certain widget models are linked on the JavaScript side (Using things like @jasongrout 's link widget), it would make sense to keep the widgets active to maintain this interaction even if the comm is closed.

I talked to @SylvainCorlay in person and I agree, I shouldn't use disabled for the reason he mentions above. How about using http://fontawesome.io/icon/chain-broken/ to symbolize the link is broken?

I am reading / testing this in the meanwhile. This is related to #6151, #6194 and #6216, in terms of lifetime of widget models and comms.

This PR will wait for the PR's @SylvainCorlay mentions. I'll rebase this after they are merged and then I'll implement a comm_alive trait or something similar and use the link icon to denote whether or not the widget is running in a detached state.

There are still a few things to be done here, in addition to the comments above:

• Fix the nbconvert content (broken after rebase).
• Make this work with undisplayed models. This will require reference counting.

Following the hangout of yesterday on this matter with @ellisonbg @minrk @takluyver @jdfreder and @jasongrout .

One of the reservations of @minrk against the persistence of model states in the notebook file format is that it seems to imply that we would need future versions of IPython's widgets to be compatible with these models.

However, the "nbconverted" notebook embeds the optimized version of the JavaScript code, meaning that it will always be usable regardless of the changes in IPython's code. Readablility of the file only depends on the ability of future browsers to execute today's JavaScript.

We would only be sensible to changes in IPython code when loading a notebook ipynb file with a different version of ipython.
In that case, the simple addition of a model version number in widget would solve the issue. If the version numbers don't match, we should not try to re-instantiate models and views from those states.

In doing so, there would not be incremental backward incompatibility due to the persistence. (Backward incompatibility is already introduced when changing the Python API of widgets anyway.)

This PR is probably going to be broken into small bites as described in the For 3.0 section of the widget Trello: https://trello.com/b/5Xaj8Z7j/widgets

Superseded by #6789