RFC: Keeping a stack of expressions in trans

[steveklabnik]

Issue by cmr
Friday Mar 28, 2014 at 05:12 GMT

For earlier discussion, see rust-lang/rust#13175

This issue was labelled with: B-RFC in the Rust repository

---

For rust-lang/rust#13146, adding the lint is easy, but getting a span to a useful place is very difficult. I propose we keep a stack of expressions being translated by pushing the ast::Expr to a task-local stack in expr::trans and expr::trans_into. Then trans_intrinsic could look up that stack for an expression with a useful span, and could even print out everything (including type params) that led to the substitution failure. I think this will be helpful if #8 gets merged, too.

[Stebalien]

@cmr is this still relevant after MIR?

[eddyb]

The original problem (generic proxy of transmute) no longer exists, nor do we want to have monomorphization errors/lints, at least nothing like this.
And yeah, with MIR this would have a different design anyway (among other things, it would have to be done during the collector, not the regular trans, which is no longer recursive).