Is the kernel error being reported correct (i.e. the most useful?)

[greazer]

See the following image. The kernel (interpreter) I selected apparently is missing access to win32api.dll somehow. Yet the error reported is a much more nebulous and late-to-the-party type message: "Cannot execute code, session has been exposed".

I think we have code to try and display to the user a better error message when a kernel can't be started, but are we getting the right one?

[DonJayamanne]

session has been disposed is a bogus error (its just an erro from us indicating that something is wrong and we don't know what).
Basically we know why things fell over, this is basically debt that needs to be cleaned up.

I have some work (pending PRs branches from a week or so ago) that'll address this. waiting for debt week to fix them. Hence assigning to myself.

[DonJayamanne]

elected apparently is missing access to win32api.dll somehow. Yet the error reported is a much more nebulous and late-to-the-party type message: "Cannot execute code, session has be

FYI @greazer we have some other suggestions on what we can do with these errors (tracked here #7902 & #7846 (comment)).

TLDR: We found that re-installing ipykernel, pyzmq and a few other packages resolves the kernel startup.
& the plan was to implement changes so that we'd re-install them for the user to try & get the kernel working.

[DonJayamanne]

I think notifying the user immediately that the kernel died instead of displaying an error message would be better (much like what Jupyter does).
Suggestions here #7399

[DonJayamanne]

@greazer I think #7399 is relvant.
I think what should ideally happen is:

• When kernel dies, automatically restart the kernel
• Do not log errors like session disposed with a long stack trace... etc

FYI - When using remote jupyter or local kernels where raw kernels don't work, kernels automatically restart. I.e. this error is only visible when using raw kernels in local scenraios.

[DonJayamanne]

Testing:

• Create a notebook with at least 3 cells (1 & 3 containing print statements)
  • Second cell should have the following code

import IPython
app = IPython.Application.instance()
app.kernel.do_shutdown(True)

* An error message should also be displayed indicating the kernel died (bottom right)
* Try to run another cell or the same cell in the notebook
* You should be presented with a modal message indicating the kernel is dead with the option to restart the kernel

• Create a notebook with the following code:

import os
os.getpid()

* Using this PID kill this process from the task manager or terminal
* An error message should also be displayed indicating the kernel died (bottom right)
* Try to run another cell or the same cell in the notebook
* You should be presented with a modal message indicating the kernel is dead with the option to restart the kernel