Wrap setting of kernel_id with method that can then be overridden in …

[kevin-bates]

…subclasses.

A recent requirement for Jupyter Enterprise Gateway is for clients to be able to
specify the kernel_id for new kernels. Although jupyter_client.start_kernel() will
honor client-provided kernel_ids, Notebook's override of start_kernel() changes
the semantics of a non-null kernel_id in the argument list to mean an existing
(persisted) kernel should be started. As a result, applications that derive
from the kernel management infrastructure beyond Notebook cannot influence the
derivation of the kernel's id via the existing argument list behavior.

By introducing the determine_kernel_id() method, subclasses are able to derive
the kernel's id however they wish.

With the ability to know the kernel's id prior to its invocation, a number of
things can be done that wouldn't be possible otherwise. For example, this
provides the ability to setup a shared filesystem location possibly pre-populated
with data relative to what the request (i.e., kernel) is going to need.

[minrk]

Maybe simpler verb 'new_kernel_id'?

[kevin-bates]

I'm okay with changing the verb in the method name - please see next comment.

[minrk]

Should determine_kernel_id be called if kernel_id is specified? It seems like it should only be called if kernel_id is not specified.

[kevin-bates]

@minrk - thank you for taking a look at this.

I was originally thinking that the override should be called unconditionally with the base implementation honoring the existence of the parameter, but that probably leaves too much room for mistakes, so I'd be okay with calling the method only when kernel_id is None.

Not sure if its my Java background (very new to python), but I don't like using new as a verb. Does something like init_kernel_id(**kwargs) sound any better?

Would you mind suggesting the appropriate python way for constructing the conditional call based on the pop result?

[minrk]

I would write it like this:

if 'kernel_id' in kwargs:
    kernel_id = kwargs.pop('kernel_id')
else:
    kernel_id = self.new_kernel_id()

By Python (or maybe just Jupyter) convention, I like new better than init, because init implies a stateful change. I would expect .init_kernel_id() to set self.kernel_id, not return a value I'm free to use. The verb I would use for a stateless callable that returns a new kernel id is 'new'. 'generate' works, too, but it's more verbose than I would like.

[kevin-bates]

Thanks for the explanation on new vs. init - that makes sense. Regarding the example (and a better understanding of python args), I figured things would be more compact and something like the original would have been suggested:

kernel_id = kwargs.pop('kernel_id', self.new_kernel_id(**kwargs))

I'd prefer to push this version, but would be happy to submit the expanded form if you feel strongly about it.

[minrk]

It doesn't make much of a difference here, but I find clarity is typically preferable to compactness. The if/else is also ever-so-slightly more efficient, since new_kernel_id is only called if needed, rather than this PR, where it's called unconditionally, even if the result is just discarded.