Question: Type Inference is no more

[oneearedrabbit]

Hi. Love this project, thank you for building and sharing it! I started studying source code and noticed that Type Interference has been removed after version 0.4.0? I'm curious about the reason behind this decision, did you find anything incompatible?

P.S. I apologize for opening an issue--I couldn't find another way to get in touch with you.

[calroc]

Thank you very much! Aren't you kind! :)

(Technically this GitHub repo is a mirror of the main repo at OSDN: https://osdn.net/projects/joypy/ and the forums or mailing list hosted there would have been the place to post, however I don't mind at all that you opened a ticket here! You're only the second person to ever ask me anything about this project at all. I'm delighted! I only mention it to be complete. (And anyway I just noticed that they're having problems over there: the repos are empty right now. Hmm.))

Type Interference has been removed after version 0.4.0? I'm curious about the reason behind this decision, did you find anything incompatible?

It's a funny story. I was learning the Unification algorithm as I wrote the type inference/checking code from studying miniKanren, and I realized at some point that this was very Prolog-shaped code. There's a note to that effect at the bottom of the Notebook describing the code, eh?

Then I wrote an interpreter in Prolog and it was beautiful. Then I wrote a type inferencer in Prolog and it was the same code as the interpreter! So I deleted one copy of the Prolog code, and the Python code. I really liked that code but as the writers say, you sometimes must "kill your darlings". It was inferior to the Prolog code, it's more elegant to "shell out" to a Prolog interpreter than to keep that junk around.

Also, looking to the future, it's easier to use the Prolog code to drive compilation and other fun things, at least I suspect so.

Warm regards, and thank you again!

[oneearedrabbit]

That is fascinating, thanks! I may need to rebuild my mental model a bit, i.e. I was considering a barebone Forth/Joy implementation as a core (whatever that means, but something even simpler than stoneknifeforth, e.g. no comments or multichar words out of the box); surprisingly, it is already extensible enough to implement a decent Forth and then bootstrap a simplistic Prolog. My assumption was Prolog is great during the next stage boot to extend Forth/Joy with additional features, yet it is not the core component. To some extent repeat the experiment from "Designing and Programming Personal Expert System" by Carl Townsend and Dennis Feucht, yet start from a different foundation.

[calroc]

I'm not sure I get all that you're saying, but it sounds good!

One interesting observation is that, if you model the pending expression as a stack-of-stacks (rather than just a simple stack) then the "get next term" function begins to resemble a Forth system very closely.