Consider an AST validation function/pass to assist with debugging user-defined macros

[huonw]

There's a variety of invariants in the AST, e.g. the Expr in a PatLit should (AFAIK) be an ExprLit, and the Option<Expr> in an ExprIf should be either a ExprBlock or ExprIf. It would be convenient if the compiler either had a AST validation pass run after macro expansion (could include verifying that there are no macros left in the AST), a validation function called on the resulting AST after each macro is expanded or just a series of manual functions that syntax extension authors can call on the code they generate to help debugging.

(The first two should, IMO, be opt-in, since most users won't want to be wasting the compilation time on something that should never trigger if they're not writing their own macros.)

[huonw]

(@kvark: this was prompted by our discussion on IRC, but you left before I could point it out. :) )

[kmcallister]

It seems like this could be a lint, although we may need a way to run it earlier in the pipeline. That opens up the possibility of reusing some of the machinery in plugins to enforce style and code structure. Perhaps there's a nice declarative language for describing valid and invalid AST shapes.

The first two should, IMO, be opt-in, since most users won't want to be wasting the compilation time on something that should never trigger if they're not writing their own macros.

My hunch is that the compile time won't be significant, unless our invariants are really complex. Converting the lint system from one pass to several dozen increased the compile time for librustc (before it was split up) by about half a second. It's hard to do anything syntax-focused that takes significant time compared to trans and LLVM. (edit: sorry, not dozens of passes, but a bunch of virtual calls at each step of the traversal)

[steveklabnik]

Triage: I think that this solution here is HIR/MIR, @nrc?

[nrc]

This is not addressed by the HIR/MIR, but it is likely that this would be a totally different concept in an AST-less macro world. Let's see how that pans out, I guess.

[steveklabnik]

This is an interesting idea, but has gotten basically no comments in years. I don't think keeping this issue open is helpful, and so I'm going to close. If you'd like to work on fixing this, please let me know and we can re-open!