miri: better ptr-out-of-bounds errors

[RalfJung]

For offsets larger than isize::MAX, display them as negative offsets.

r? @oli-obk

[rust-highfive]

Some changes occured to the CTFE / Miri engine

cc @rust-lang/miri

Some changes occured to the CTFE / Miri engine

cc @rust-lang/miri

[RalfJung]

I don't think it is possible to trigger the negative case in CTFE, but it is possible in Miri: rust-lang/miri#1853.

[oli-obk]

well... we can run the same code in CTFE: https://play.rust-lang.org/?version=nightly&mode=debug&edition=2018&gist=6670eb77e14af454abf6f33d1523f4b2

... but the error message is absolutely useless:

warning: unused variable: `x`
 --> src/lib.rs:5:9
  |
5 |     let x = unsafe { x.offset(-1) };
  |         ^ help: if this is intentional, prefix it with an underscore: `_x`
  |
  = note: `#[warn(unused_variables)]` on by default

warning: constant is never used: `FOO`
 --> src/lib.rs:2:1
  |
2 | / const FOO: () = {
3 | |     let v = [0i8; 4];
4 | |     let x = &v as *const i8;
5 | |     let x = unsafe { x.offset(-1) };
6 | | };
  | |__^
  |
  = note: `#[warn(dead_code)]` on by default

error[E0080]: evaluation of constant value failed
    | 
   ::: src/lib.rs:5:22
    |
5   |     let x = unsafe { x.offset(-1) };
    |                      ------------ inside `FOO` at src/lib.rs:5:22

For more information about this error, try `rustc --explain E0080`.
error: could not compile `playground` due to previous error; 2 warnings emitted

I don't know how this happened, it sounds to me like the intrinsic call is treated as a separate const evaluation?

[RalfJung]

The same test as CTFE doesn't work since offset(-1) will overflow (which is a different error message than going out-of-bounds).

But I found a test that works. :)

I don't know how this happened, it sounds to me like the intrinsic call is treated as a separate const evaluation?

The actual error span is inside the offset implementation. So the inside `FOO` at src/lib.rs:5:22 is just part of the backtrace. But I am not sure why the other span is not shown...

[oli-obk]

But I am not sure why the other span is not shown...

because we don't have the libcore source available. we should still print messages, even if the source of a span is missing. probably will get solved when we put the error message as the main message, instead of a label

[oli-obk]

@bors r+

[bors]

📌 Commit bed3b96 has been approved by oli-obk

[bors]

⌛ Testing commit bed3b96 with merge 718d53b...

[bors]

☀️ Test successful - checks-actions
Approved by: oli-obk
Pushing 718d53b to master...

[rust-highfive]

📣 Toolstate changed by #87224!

Tested on commit 718d53b.
Direct link to PR: #87224

💔 miri on windows: test-fail → build-fail (cc @oli-obk @RalfJung @eddyb).
💔 miri on linux: test-fail → build-fail (cc @oli-obk @RalfJung @eddyb).