When to use state vs return type in Process framework?

[hansmbakker]

In the Process framework, process steps can have stronly-typed state. Also, their logic is implemented using KernelFunctions which can return an object.

I couldn't find a clear explanation in the samples when to return an object from the KernelFunction versus when to return using the data parameter when calling EmitEventAsync.

[markwallace-microsoft]

@esttenorio Can you help here?

[esttenorio]

Hi @hansmbakker, return ->OnFunctionResult is recommended when your function only returns the same object type when executed [no conditional logic or complex logic that may need to branch out to different process paths]

Making use of EmitEventAsync allows custom mapping of events from a same function. Example:

if (someCondition)
{
    await context.EmitEventAsync("ConditionSucceded", correctlyGeneratedObject);
} 
else 
{
    await context.EmitEventAsync("ConditionNotMeet", reasonsObjectFailedToBeGenerated);
}

correctlyGeneratedObject and reasonsObjectFailedToBeGenerated can be different object types and yet be emitted by the same function. If using onFunctionResult probably a struct with both object types would be needed since the function would be able to return always the same object type.

We have a sample with a similar scenario here:
https://github.com/microsoft/semantic-kernel/tree/main/dotnet/samples/Demos/ProcessWithCloudEvents

In particular you can check the step ProofreadDocumentationStep

[hansmbakker]

Thank you 👍🏻