Waiting for ipykernel to be installed in terminal

[DonJayamanne]

As discussed with @rchiodo @karthiknadig
Scenario:

• User starts notebook
• Runs a cell
• ipykernel is not installed in current environment
• We prompt user to install ipykernel and user clicks ok
• The installer sends the python -m pip install ipykernel (or similar) to the terminal
  https://github.com/microsoft/vscode-python/blob/5c3433b6364db03fadbf58010937b65a23d106e5/src/client/common/installer/moduleInstaller.ts#L45
• Installer completes and resolves.
• Calling code assumes ipykernel has been installed
• We attempt to start notebook, and kaboom.

Problem

• installer code returns even before the installation has completed

Cause

• We have no way to tell whether terminal/shell has completed executing our request

@int19h @karthiknadig Doesn't this sound familiar.

Proposed Solution

• Create a new subclass class that overrides the existing sendCommand method in TerminalService terminalService.sendCommand(pythonPath, args);,
  • Fortunately we have a factory. So we can switch between the two in the factory class.
  • That's one place to make a simple decision
• Ensure we have a generic Python file (~/pythonFiles/execInShell.py) as follows:

# Note: The command and arguments are hardcoded as a sample.
# These commands could be '/usr/desktop/bin/conda install ipykernel -y' or other

# print command to output in terminal so user can see what is being executed.
# These values will be passed as arguments to the file.
print('python -m pip install ipykernel')
try
    # Signal completion of command with success state
    open('started.log')

    subprocess.check_call(["python", "-m", "pip", "install", "ipykernel"], stdout=sys.stdout, stderr=sys.stderr)

    # Signal completion of command with success state
    open('success.log')
except:
    # Signal completion of command with error state
    open('error.log')

• Next, send the command python <fully qualified path>/execInShell.py python -m pip install ipykernel to the terminal (all arguments passed to execInShell.py will be executed in subprocess.
  Note: Terminal is has already been activated (current behavior).
• Extension code will use the files to check whether the process has started, completed.

Notes:

• This allows us to send any script to the terminal and wait for it to run to completion.
• E.g. python execInShell.py /usr/desktop/bin/conda install -n env1 ipykernel
• E.g. python execInShell.py c:\wow\conda.exe install -n env1 ipykernel

Obviously the args will be escaped properly.

[DonJayamanne]

Not sure whether to use files or a socket to do the necessary synchronization.

• With a file, we'd be polling it regularly.
• With sockets, we'd need to setup a socket server (seems pretty heavy handed, and need to consider port forwarding, etc...)

[DonJayamanne]

FYI - Going with FIle option as this is simpler. Doesn't impact perf as installation is happening in terminal and that's what's slow and what the user is waiting on.