Can we deal with user code (packages) causing kernel failures

[DonJayamanne]

• We know tensorflow incorrectly installed could cause the kernel to crash
  Such as importing tensorflow See Kernel crashes when importing tensorflow #7600 , Tensorflow import Error #4645
• Similarly other packages
  • Kernel crashes when using fastparquet #9368

Misc:

• Kernel dies when running some code #9283

Suggestions

• Improve error message dispalyed in the popup
  • When kernel crashes provide a more meaningful message about what happened, users might not know what a kernel is, hence might ignore the error message or wouldn't be able to connect execution of cell to an error message about kernel failure
• Include error message in cell output
  • This way they know what cell caused the failure
• Should we pay specific focus to packages such as tensorflow, numpy, fastparquet (and provide a link to a Wiki)
  • Note: We do this today for kernel startup failures (e.g. if we have a DLL exception, we provide error information and a link to a WIKI)
  • Provide instructions in the error output specific to some packages (with links to WIKI)

[rchiodo]

@DonJayamanne do you have verification steps for this issue?

[DonJayamanne]

Run a cell with the following code (cause the kernel to crash)

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

Output should contrain a link to wiki as follows:

[rchiodo]

So this is the same fix as #9588.

/verified