Auto-Generate Functions about imagej-ops

[dscho]

It is doable, alright. But our paradigm has been -- at least so far -- to try to Javassist things only as an optimization, and requiring everything to work even without Javassist (albeit slowly).

[ctrueden]

I think Function interface is actually about more than just performance. It conveys expectations about repetition and multithreading, too: the "special" input is the one which will vary when repeatedly calling the function in a loop, and/or from multiple threads. So, I do not think the single-input Function paradigm is "arbitrary" in that sense.

I am wary of clever code generation. Even the Velocity-based generation we are doing now adds a level of indirection making the codebase harder to understand. And throwing javassist into the mix seems very dangerous to me, from a maintenance/debugging perspective.

If we need a Function-like thing that has two special inputs, we can create a separate interface for that, and so on, for each case... unless you have a use case requiring some arbitrarily large constant C number of variable inputs to a compute-like method? But if so, there is probably another way to approach the problem.

[ctrueden]

@dietzc: Do you still want to pursue this idea? Or did I talk you out of it at the hackathon? I'm happy to discuss further on this thread if you wish; otherwise, let's close this issue.

[dietzc]

you actually talked me out of it. closed! 👍