Use controller option in bake

[jedevc]

🗒️ Marking as draft until we merge #1671.

This PR continues the move to attempt to merge the build.Options struct into the controllerapi.Options message.

To do this, we extract all the input parameters into a dedicated message (adding the missing ones, except for the InStream parameter which will require some additional fiddling around). We also rework the NamedContexts to allow containing States (by transmitting them as DefinitionOps), and adding a Linked field to the common options.

With these changes, we can have bake generate controllerapi.BuildOptions, and then convert those to build.Options using the controller/build package.

This is an intermediate patch, designed to allow us to clean up some shared logic between both build and bake. The next step will be to modify bake to use the controller API directly, and completely skip the build.Options generation step.

[ktock]

Thank you for working on it.

[ktock]

Thanks, LGTM.

[crazy-max]

Linked is internal. I'm a bit on the fence to have it in the proto def:

buildx/build/build.go

Lines 93 to 94 in 4a73abf

	// Linked marks this target as exclusively linked (not requested by the user).
	Linked bool

[jedevc]

Agreed - but based on the current usage of it, I think we need to. It's used in BuildWithResultHandler which will be called from the controller side, while it's generated from targets before we make the controller request.

I think it's only used here?

buildx/build/build.go

Lines 688 to 690 in 4a73abf

	if !opt.Linked && len(opt.Exports) == 0 {
	noOutputTargets = append(noOutputTargets, name)
	}

I'm quite sure how else we can do it, though open to suggestions 🤔

[crazy-max]

Ah right let's keep then 👍 (with a comment in the proto)

[jedevc]

Actually, I think we could get rid of it, if we could move the default outputs onto the client side of the controller (which would probably be a better design anyways).

Might be slightly tricky, since we only load the driver on the controller server (which we probably want to change anyways), so maybe worth returning to this with that refactor?