Quick fix: make opt_si its own target to avoid collisions.

[eplondke]

Longer term we need to reduce recursive make.

[bryanb-h2]

So the rule now is that the T for each variant has to be distinct? That defeats the purpose of T, which is to specify which of opt or ref to build for a give target.

[eplondke]

Is opt_si different from opt?
Yes ==> build a separate build
No ==> why have it?

The collision was from two things that get built both for opt and opt_si. I think practically the problem is that we're using recursive make, and whatever different or updated opt_si is needs to be a part of the global dependence graph.

For now this should resolve your issue: by building everything twice since something might be different?

[bryanb-h2]

Yes, opt_si is different. See lines 161-163 below. This is a build that needs to be installed under pkw along with the normal opt.

"build a separate build" is exactly what is supposed to happen; the question is how to specify the parameters of the build.

Each $(TARGET) is a $(T) with particular extra build options. $(T) is either opt or ref, but $(TARGET) can be any number of things.

Seems to me that the structure should be artifacts/v$(ARCHV)/$(TARGET), not artifacts/v$(ARCHV)/$(T)

We could also inline all the $(TARGET) configs in this file into separate make targets...but that's where what we have now came from, and it was messy.

Either way, I don't think T := opt_si is the right way to go here.

[bryanb-h2]

This works.

[bryanb-h2]

Please close.