fat-ptr-transmutes lint doesn't work

[kmcallister]

#![deny(fat_ptr_transmutes)]

use std::mem;

fn main() {
    let x: [u8, ..8] = [0, ..8];
    let y: [u8, ..16] = unsafe {
        mem::transmute(x.as_slice())
    };
    println!("{}", y.as_slice());
}

compiles and runs with no warnings. The problem is that trans::intrinsic::check_intrinsics calls Session::add_lint but the pass that emits these lints has already run. See this comment in rustc::lint.

Please add a test case when fixing this! :)

[kmcallister]

cc @nick29581

[kmcallister]

Once this is fixed, I will submit a patch I wrote that ICEs when add_lint is called in the wrong place.

[kmcallister]

@nick29581: Any thoughts on this? I think we should fix this for 1.0 or remove the lint for now, because it never prints anything.

[nrc]

I think we should fix it :-) Do you know how hard it would be to do another lint pass after trans?

[kmcallister]

I think it's worth a shot. It would also eliminate a bunch of hacks for the existing variant size lint in trans.

[kmcallister]

If we leave these two lints hardcoded then we don't need a full pass, we just need to go through the session / AST and print out saved lints, which I think could be factored out easily enough.

[Stebalien]

Still broken. Updated test:

#![deny(fat_ptr_transmutes)]

use std::mem;

fn main() {
    let x: [u8; 8] = [0; 8];
    let y: [u8; 16] = unsafe {
        mem::transmute(&x[..])
    };
    println!("{:?}", &x[..]);
}

[steveklabnik]

Triage: no change.

[Mark-Simulacrum]

As far as I can tell, this lint is never invoked today (even in trans); it's previous use was removed by @nikomatsakis in c5edd22. @nikomatsakis, could you comment on what the reasoning there was? I feel like it might've been an accidental removal...

Looking into the code seems to suggest that the problem is that with slices, the representation (as I understand it) is *(usize, usize) so the type's size is the "same" in this case. That would explain why we don't detect this, I believe, though I may be wrong about my reasoning here.

[nikomatsakis]

@Mark-Simulacrum hmm I do think that was an accident, or perhaps a merge failure.

[asquared31415]

This lint appears to have been completely removed, and trying to use it emits a "removed lint" warning. Should this issue be closed?

[jieyouxu]

This lint doesn't exist today since it seems to be...

rust/compiler/rustc_lint/src/lib.rs

Line 387 in 002da76

store.register_removed("fat_ptr_transmutes", "was accidentally removed back in 2014");

Closing.