Run cells before notebook is live

[SylvainCorlay]

When a notebook just got opened, and as the kernel is still starting, it is already possible to "execute" the first cells with Shift+Enter.

However, execution requests are lost instead of being queued and the later cells executed likely error due to the missing first cells. This problem is especially visible with windows where the kernel startup time is longer than on OS X or linux.

We should probably either queue the messages or keep the current cell selected.

[minrk]

If I recall correctly, execution requests aren't lost because the kernel connection isn't established yet. Instead, they are never produced.

[SylvainCorlay]

OK, it remains an issue from a UX perspective. We should probably avoid giving the impression that cells are run.

[jasongrout]

Yes, it looks like this line is the one ignoring the execution:

notebook/notebook/static/notebook/js/codecell.js

Line 308 in 5275388

if (!this.kernel || !this.kernel.is_connected()) {

[jasongrout]

Here are a couple of options:

1. Queue up the pending cells at the notebook level and run them when the kernel is set. This doesn't prevent the cells from being modified before they are run, though.
2. Make a fake "kernel" that just queues up execution requests, and set that to the cells when they are first created. This has the advantage that the cell logic identifying it's running, etc., still works as-is.
3. Throw an error when trying to run a cell and a kernel is not active.

I think a combination of (2) and (3) would work well. The initial dummy kernel optimistically assumes there will eventually be a good connection, and just queues up execution requests. However, if there is an error with the kernel, we throw an error to the user when trying to execute the cells.

[jasongrout]

@SylvainCorlay, @minrk, comments on the above?

[jasongrout]

One question: when we delay execution, should we execute the original executed contents of the cell, or the current contents of the cell?

[jasongrout]

Okay, I think I have a minimal solution at #1002, but it's one that makes the notebook rely a bit on the CodeCell behavior, so it smells a bit, but maybe not more than other notebook code.

[takluyver]

Should this be closed by what Jason implemented in #1002, or is there more to do?

[minrk]

Just lost track. This is fixed.