Execute ipnb File via API, resp. in Batch mode

[FlorianKummer]

For our research code, in the past, we used some C#-scripting module from Mono (Mono.CSharp.dll) as a scripting frontend, we even created our own, custom (but glitchy) notebook GUI.
Now, we want to move everything towards dotnet-interactive, resp. Jupyter notebooks.

A typical use-case is the following:
We further used these worksheets in our test-procedure:

• We developed and implemented some numerical method,
• wrote a notebook to do some calculations,
• wrote a scientific publication and then
• added the notebook to our set of tests in order to ensure that certain numerical properties still hold when the code is developed.

I allready managed to convert our custom notebooks to ipnb, using mostly members of the Microsoft.DotNet.Interactive.Notebook namespace.

Now, I also want to execute those notebooks which we use for testing as a unit test. I could theoretically use jupyter nbconvert --execute ..., but I'm wondering if there is any option which does not depend on an existing Jupyter installation.

I already managed to execute a single command by CSharpKerner.SendAsync(...), which seems to work; at least, I can fetch variables through CSharpKernel.TryGetVariable(...). However, I am missing some general way to check if the execution went well or not, i.e. a check for uncaught exceptions, syntax errors, etc.

Is there any way of executing an entire notebook and check whether some Exception or NUnit Assertion has been raised, or if any Syntax error occurs?
(Either command-by-command or the entire notebook at once?)

[jonsequitur]

This is something we're working on. You can see the related issues under this label: https://github.com/dotnet/interactive/issues?q=is%3Aopen+is%3Aissue+label%3AArea-Automation

The events returned by the KernelCommandResult.KernelEvents observable will tell you if something went wrong. The sequence will conclude with either a CommandSucceeded or CommandFailed event. Exceptions from the notebook code will be wrapped into the CommandFailed event.

[FlorianKummer]

Thanks, this already helps me a bit futher;

Which kernel would you recommend for this kind of 'end-user'-interaction?
I tried the following:

   CSharpKernel csk = new CSharpKernel();
   var obs = new MeObservator(); // implements IObserver<KernelEvent>
   csk.KernelEvents.Subscribe(obs);
   
   var t4 = csk.SendAsync(new SubmitCode("sdjhsjkahdk  error!"), CancellationToken.None);
   t4.Wait();

However, this seems to cause some sort of infinite loop, MeObservator.OnNext(...) is beeing called over and over and the task never completes.

[jonsequitur]

The Kernel.KernelEvents observable doesn't terminate. If you want an observable scoped just to the command you sent, you can do this:

var result = await csk.SendAsync(new SubmitCode("sdjhsjkahdk  error!"), CancellationToken.None);
result.KernelEvents.Subscribe(obs);

Note that in this model, you won't actually receive any events until they've all been emitted. The KernelCommandResult.KernelEvents observable is a ReplayObservable.

Another approach, if real-time updates are important and processing the command might take a while, is to subscribe to the Kernel.KernelEvents only until you receive a completion event for the command, i.e. a CommandSucceeded or a CommandFailed.