Build inline components

[itowlson]

I am not sure what to do about the duplication of manifest deserialisation stuff. It's intentional that build shouldn't break over a bad kebab etc., but I'm not loving how much this increases build's need to make "forgiving" copies of manifest code. I'm open to ideas on a way forward!

[lann]

It's intentional that build shouldn't break over a bad kebab etc

To clarify: you're saying that you want build to work when the manifest is otherwise invalid? Is that scenario worth accommodating?

[itowlson]

There is invalid and there is invalid. It does break flow when you can't build your code because you forgot whether variables are snake or kebab case or you messed up an allowed_outbound_host. "Yes but I can fix that later, I just want to get this lifetime error sorted and then I'll come back to it." Admittedly, it breaks flow once and then it's done... We can certainly reassess this, and I guess should: I have a feeling that we adopted the "forgiving subset" approach at a time when the loader crate did a lot of validation: I think we now expose something closer to a 'raw schema' stage, so maybe that's an acceptable compromise.

[lann]

I think we now expose something closer to a 'raw schema' stage, so maybe that's an acceptable compromise.

Yeah that was part of my intention with the Great spin-manifest Refactor of '23. If we feel like ID casing is something we need to be flexible with here I hope we could add that flexibility to spin-manifest by - say - making that validation optional somehow.