For typical fullstack setup (common, js, jvm source sets) I would like to process common sources and generate new sources for all three targets. Should I use three different processors or there is some way to know which source set is being processed?

That's correct. I'm currently thinking about a possible API, which should be convenient to use, yet cover all relevant use cases. I'm currently investigating possible use cases like processors which feature • one set of generated files per source set (files are generated when their input source set is the current output source set), • expect/actual code generation with an input source set containing annotations which then trigger generation of

expect

code on the same source set level and

actual

code on leaf (target) source sets. If you'd like to, you could add your use case to https://github.com/google/ksp/issues/965. That would certainly help.