OK. Indeed there seem to be quite some ambiguities and inconsistencies and the knowledge we are basing our discussion on is a bit fragmented and "abstract".
Sorry for being a bit pushy but I think we are not doing enough as a community to make life our users easier, and we sometimes seem to proceed without fully understanding of consequences for them - downplaying the importance of it. And for public API I think this is a very important topic. I think there is an easy way to get a better understanding - simply use the OpenAPI generator to generate both - clients (in a few languages) and server stubs (in python) for Airflow and become the "user" and try to use the client + server and see where we get from there. This will haunt us for a long time if we make the wrong decision here. For me - this is one of the most important factors on how easy it will be for someone to use the API once we release it. And being able to use generated clients in several languages is pretty much non-negotiable requirement for the API. I will spend some time over the weekend trying it out and see whether the problems that I pointed out are really present in our specs and if they are - whether can be solved quickly - and if there are problems and they need some workarounds or customizations - for me it's super important that we are aware of those problems and we let the users of our API know how to solve them I will be posting some results while I do so - I count on your help QP and Ash if I see any non-obvious obstacles. J.
