E0425: Don't recommend the use of intrinsics first

[dave-tucker]

Given the following code: https://play.rust-lang.org/?version=stable&mode=debug&edition=2021&gist=dc1180286b0a5d9e0f1dbac02ccdb7b3

fn main() {
    let sz = size_of::<usize>();
    println!("Hello, world! A usize is {}", sz);
}

The current output is:

error[[E0425]](https://doc.rust-lang.org/stable/error-index.html#E0425): cannot find function `size_of` in this scope
 --> src/main.rs:2:14
  |
2 |     let sz = size_of::<usize>();
  |              ^^^^^^^ not found in this scope
  |
help: consider importing one of these items
  |
1 | [use core::intrinsics::size_of;](https://play.rust-lang.org/#)
  |
1 | [use core::mem::size_of;](https://play.rust-lang.org/#)
  |
1 | [use memoffset::__priv::mem::size_of;](https://play.rust-lang.org/#)
  |
1 | [use pin_utils::core_reexport::intrinsics::size_of;](https://play.rust-lang.org/#)
  |
    and 3 other candidates

Ideally the output should look like:

error[[E0425]](https://doc.rust-lang.org/stable/error-index.html#E0425): cannot find function `size_of` in this scope
 --> src/main.rs:2:14
  |
2 |     let sz = size_of::<usize>();
  |              ^^^^^^^ not found in this scope
  |
help: consider importing one of these items
  |
1 | [use core::mem::size_of;](https://play.rust-lang.org/#)
  |
1 | [use core::intrinsics::size_of;](https://play.rust-lang.org/#)
 |
1 | [use memoffset::__priv::mem::size_of;](https://play.rust-lang.org/#)
  |
1 | [use pin_utils::core_reexport::intrinsics::size_of;](https://play.rust-lang.org/#)
  |
    and 3 other candidates

In cases where intrinsics have been stabilised in core|std it would be useful to suggest these to users first.
Especially as using intrinsics generates another rustc error that tells you to prefer stabilized functions in std.
This would hopefully make handling this error easier for new rust users coming from languages like C/C++ and looking for the equivalent of sizeof()

The output is the same on stable, beta and nightly

[compiler-errors]

I'd be satisfied if we filtered out intrinsics altogether except for when the core_intrinsics feature was enabled.

[LoganDark]

I'd be satisfied if we filtered out intrinsics altogether except for when the core_intrinsics feature was enabled.

I also get quite annoyed when stable rustc recommends I use unstable features (maybe that should be fixed in an RFC)

[estebank]

Checking the DefId of each suggested path to see if they are annotated with unstable should be doable.

[compiler-errors]

Checking the DefId of each suggested path

This seems to be harder than I initially thought... so since we emit this "consider importing" suggestion in LateResolutionVisitor, we do not have access to TyCtxt it seems. And I have no idea how to extract the attributes from each Res without TyCtxt... 🤔

[estebank]

We might want to consider adding yet another dumping ground field in the Res itself or in Resolver... But at that point it seems like too much mem consumption for little gain. The other option is to delay the emitting of these errors until later, but that also seems... like a lot of additional complexity :-/

[compiler-errors]

For now, I guess it wouldn't hurt just to filter suggesting paths that are literally ["core", "intrinsic", ..] (and std::intrinsic::*) instead of going thru all of the work of adding more info to the Res. I can put up a PR that does that later.