Pandoc for programming languages ? I like the idea, the “common core” as other posters said needs to be arbitrarily defined because concepts vary a lot across languages. But that kind of constraint is what makes those endaevours both desperate and interesting - best kind of project if you ask me. Good luck !

A bit pretentious writing style but I enjoyed the article.

Extremely ambitious project that I’d join and would try to help. I do agree that each language’s AST is kind of a distraction and if we can standardize a converter from that to the meta-AST then we’d score huge wins in code analysis. Which we desperately need recently.

One problem is that many languages don’t have a stable AST – Elixir is one of them. However, combining the language and its version allows us to pin down exact AST details. Requires more work – but it’s only done once.

Really though: amazing idea. Would love to see it bear some fruit.

Strange post.

First of all, there are no concepts which exist in all languages. Even variables don’t exist in all languages (like stack-oriented languages or assembly languages).

Second, is that purity is a higher-level construct and what pure and what is not — defined strictly by the developer. In a language like C, setting data by some local pointer can be a side effect, given that it can be used as a way to share some information with another thread. And even if you define what is pure and what is not (what will most likely be different from the developer’s understand), you can’t just write a generic analyzer, because not all languages have functions, and those who have them, usually can’t provide a guarantee that all code reachable by the function will be a part of the call graph.

Even in Elixir, this code is pure

def add_five(x) do
  x + 5
end

But so is this

def add_five(x) do
  ref = make_ref()
  caller = self()
  spawn_link(fn -> send(caller, {ref, x + 5}) end)
  receive do
    {^ref, result} -> result
  end
end

And the fn inside is invisible in the call graph. You may argue that the latter version changes the state of the system by introducing a new process. But I will say that the first version changes the current_stacktrace field of the process info. So, again, developer defines what “pure” means.

I mean, you can represent different languages in the similarly structured AST, but this would just offload semantic differences of the languages on the programs which would analyze such AST. There would be no universal algorithms, just universal representation with lots of language-specific algorithms around it