Tracking issue for constant prop not causing add'l errors (RFC #1229)

[nikomatsakis]

Tracking issue for rust-lang/rfcs#1229

cc @oli-obk

[eddyb]

Can we move the errors trans::consts generates to check_const?
E.g. only bar causes an error here (a special-case I added before the overflow checks were implemented):

fn foo<T>() -> u8 { 1 + 255 }
fn bar<T>() -> u8 { 1 / 0 }

foo only errors if it is monomorphized, even though the overflow exists in the generic definition.

[oli-obk]

@eddyb: that sounds reasonable as a second step, but is orthogonal to the RFC. I also don't want to create a humongous PR, so I'll address the port to check_const right afterwards. It's probably better to create a new issue (and assign it to me if you want)

[brson]

@eddyb This is ancient. Can it move forward yet?

[eddyb]

@brson Nope, this is blocked on miri!

[XAMPPRocky]

@eddyb What is currently blocking this issue?

[oli-obk]

The new const propagator needs to be fixed to actually lint these functions

1. even if they are unused (make them nongeneric and don't use them -> no lint)
2. even if they are generic (currently generic functions seem to not get linted)