Interleave compiler passes

[jfecher]

Problem

After the comptime interpreter inlines macro code at its callsite, the code will need to undergo name resolution and type checking. Yet the interpreter itself also needs access to type checking and thus must be after that pass.

Our current plan to resolve this is to have a loop in the compiler where after a macro is inlined, we loop back to name resolution and redo that pass and type checking for any relevant parts of the program which changed. This turned out to be non-trivial however. Most notably in converting back from the high-level IR (HIR) to the Ast since information is lost in the initial Ast -> Hir conversion done by name resolution.

Happy Case

One possible alternative is to interleave these passes together instead so that they operate in lock-step. This would be as if each pass were merged together into one larger pass which traversed the Ast only once. This way, when a macro is inserted the name resolver is still available to re-resolve the code immediately. Most importantly, the shape of the call stack is still intact so the new result can be checked against recursive call results.

This last point is important for type checking. Think of the example 3 + foo!() where foo!() inserts some macro code. When type checking this for the first time we'd see only ? for the result of foo!(). If the passes were not in lock-step, then we'd pop off the call stack and check the validity of Field + ? immediately and error. Since the passes do operate in lock-step however, the comptime interpreter will be available to immediately expand the code to e.g. 2 and update the type checker in place to return Field for that expression instead. Then, the call stack is popped and the type checker will now see Field + Field.

Project Impact

None

Impact Context

No response

Workaround

None

Workaround Description

No response

Additional Context

No response

Would you like to submit a PR for this Issue?

None

Support Needs

No response