[ivan.moony]

Can feature constraints replace the need for variables?

[infurl]

Can feature constraints replace the need for variables?

That depends on what you're using variables for. I'm certainly not using any variables for parsing.

[ivan.moony]

Can feature constraints replace the need for variables?

That depends on what you're using variables for. I'm certainly not using any variables for parsing.

For example, to calculate factorial or Fibonacci numbers (happens in recursion)?

[infurl]

It sounds to me like you're searching for The One True Hammer(tm) when you should just be using the best tool for any given job.

[ivan.moony]

Yeah, the mother of all the hammers sounds just about right! 

[ivan.moony]

I tidied up the parser output a bit and introduced `expected: ... at position ...` type of syntax error messages. Syntax will loose a clear distinction from semantics in later project phases, when the parser will be able to detect i.e. if a variable is declared before assignment using only exp-log patterns.

[ivan.moony]

Current project status:
Still working on logical programming language for describing other formal languages, based on extension of original parsing technology. Anytime soon, there will be completely functional general CFG parser which then has to be extended in two steps, first to be Turing complete, then to include appropriate logical operators (I observe productions as implications).

Last insights about program synthesis:
It should be possible to send arbitrary input and output pairs along the lambda calculus definition. From just that, the system could construct a set of possible specific lambda functions that turn provided inputs to provided outputs. If instead of lambda calculus we provide a definition of some other language, other kinds of source code may be synthesized. I'm still wondering about troubles with speed that may pop up, mostly around recursive functions. Anyway, simply typed LC should work well if I'm not mistaken.

Last insights about AI:
The same procedure for program synthesis should work as a replacement for what artificial neural networks provide. NNs are trained by pairing inputs to outputs, and in between, neuron set does all the future magic of firing outputs upon recognized inputs. With completely formalizing a Turing complete environment, instead of general neuron set, we may put 100% accurate synthesized pieces of code. Although the general idea should work, I'm afraid that details about AI would be pretty tricky to grasp.

What competition has to show on this subject:
Here is interesting video about business use - very insightful - https://www.microsoft.com/en-us/research/group/prose/