notes.md

notes

to fix

1. public static doesn't work, but static public works.

2. underlining breaks for multi-line spans; just don't underline in that case.

3. types don't appear before functions for some reason, for typechecking. (ie. order becomes important)

4. enum values are not working correctly (seems to be right-shifted by 8?) (the values are not assigned)

5. defer appears to be broken:

var i = 0
while true {
	defer i += 1

	// doesn't change
}

6. "unsynchronised use of global init function!!!" -- need to figure out a way to serialise access to the global init function

7. polymorphic stuff breaks when manually instantiating

8. compiler crash:

import libc as _

struct Cat<T> {
     fn greet() {
        printf("i love manga uwu\n")
    }
}

@entry fn main() {
    let c: Cat!<Integer> = Cat()

    c.greet();
}

9. ambiguous call to initialiser of class

10. assignment to runtime variable in #run block

11. numbers basically don't cast properly at all

to refactor

3. all uses of defer() need to be audited. there are instances where an exception is thrown during typechecking as a result of unwrapping a TCResult that contains an error; as the stack unwinds, it may encounter a defer() object, which necessitates calling the code inside.

   • in general that is fine (most of the time, it's just fs->pushLoc() followed by defer(fs->popLoc())), but in some instances it is not fine --- most notably, when we push a body context and pop it in the defer.

   • since the popping of the body context asserts that the top of the stack is the same as what we expect, we might end up in a situation where exception-throwing function pushes a different context and throws before popping it, leaving the calling frame with an inconsistent view

   • the assertion will cause the compiler to terminate when it should have just printed the error message generated by the exception-throwing function.

4. instead of keeping a separate list of unresolvedGenericDefns which is pretty ugly, it should be feasible to create a sort of sst::ParametricDefn that simply contains a pointer to the original ast::Parameterisable as its only field; then, we should be able to simplify the typechecking code substantially by moving those the polymorph instantiation into that definition's typecheck method instead.

5. setting the method's containing class via the infer parameter feels super fucking dirty, and ngl it has felt dirty for the past 5 years.

to investigate

1. we rely on a lot of places to set enclosingScope correctly when typechecking structs etc. there should be a better way to do this, probably automatically or something like that. basically anything except error-prone manual setting.