The problem tht I’m trying to solve is the one of compiler construction, generally. Writing that down, though, I’m struck by how naive that sounds, despite that I feel that it’s something that is solveable.
I’m not saying that I should provide a system that hides the compelxity of compiler construction “for the masses,” but instead that I need to expose the facilities to extend the compiler in a very easy way.
I need to implement each of:
Syntax -> Parser -> IR ->
Compiler construction ->
VMR -> Assembly -> running the code
What’s an example of a compiler construction? A function definition might be one, but this does require re-elaborating what we’re doing.
Have an “expression” be a first class thing in the language, and then make the parser literally just be calling the constructors as a direct translation of the AST, and the function definition be literally just normal user code which takes expressions as input.
Yes, having all statements actually return expressions is the right way to do this. I can’t believe I’ve never thought of this before.
This is the right way to do this. And, the names literally evaluate to variables. A “variable” is then also a first class thing.
(of course it is a first class thing! how have I never thought of this before)
And, then, there’s more. But, how much more can I take?
Then, to make structures, we have basically “typed bufs” which becomes our next first class thing. I think this is a another key good abstraction, as well as storage classes.
Having allocator be a part of type! This is very important. If you can say “I have a collection of objects which have this performance characteristic in their operations,” you’re very far in your abstraction.
And, structs at the REPL should be a beautiful experience (but isn’t right now, so that needs to be improved. An “awesome” experience, I mean.)
Have a variable actually just be a “name”, which is a first class thing
So, what’s the plan, stan?
- Elaborate the typesystem and what “expressions” are
- Figure out how to do the