> the best achieved are simple REPLs, hot code swapping in Erlang/Smalltalk (which does nothing with state), and...direct manipulation (you can just update your objects directly, screw any code)
I think that sort of supports the point of the GP. Lisp had all the above and more (e.g. a proper REPL, live recompiling of everything in your image) since almost forever and yet not many people know about that. There's a lot of knowledge that was just forgotten by the fashion-industry IT is.
So I don't think we should stop trying - but it's really worth to go through the work done in 60s and 70s, because we could use it.
It's just not useful, they needed to go farther and wound up with a PX that some liked, but most didn't feel was worth it. Take the REPL for example, you are stuck in a linear command driven level of interaction if state is involved (and if state isn't, it is only useful as a fancy calculator). Take fix and continue in smaltalk - you can change the code, but you can't really fix the state (I.e. By time traveling), leaving your program in a seriously fragile state. And take morphic: where you can change the state (by directly manipulating objects) but not the code, so if you change something driven by code, your changes are completely ephemeral.
Each experience is serious lacking, and there are good reasons they haven't taken over the world. Chris Hancock does a better job than me in describing this, which is why I keep bringing up his dissertation:
https://llk.media.mit.edu/papers/ch-phd.pdf
I think that sort of supports the point of the GP. Lisp had all the above and more (e.g. a proper REPL, live recompiling of everything in your image) since almost forever and yet not many people know about that. There's a lot of knowledge that was just forgotten by the fashion-industry IT is.
So I don't think we should stop trying - but it's really worth to go through the work done in 60s and 70s, because we could use it.