Hang when hovering over constant

[DaniPopes]

rust-analyzer version:

• 0.3.1885-standalone (b6d1887 2024-03-17)
• 0.4.1889-standalone (5e276ae 2024-03-20)
• 0.0.0 (8d74705 2024-03-21)

rustc version: rustc 1.79.0-nightly (a7e4de13c 2024-03-19)

relevant settings: N/A

repository link (if public, optional): N/A

code snippet to reproduce:

pub enum AA {
    BB,
}

pub fn hang() -> AA {
    const CONST: AA = AA::BB;
    CONST
}

Hovering over either CONST starts the hang and rust-analyzer becomes unresponsive, consuming about ~200MiB RAM per second in this case, though in other cases it can be much faster.

The server cannot be killed through hovering on the rust-analyzer button on the lower bar or with rust-analyzer: Stop server, and must be killed through the command line. It is either non-responsive or doesn't kill the process.

I can reproduce this consistently with recent rust-analyzer client and server, but only after updating rustc from nightly-2024-03-19 (rustc 1.79.0-nightly (3c85e5624 2024-03-18)) to nightly-2024-03-20 (rustc 1.79.0-nightly (a7e4de13c 2024-03-19)) or later with rustup (override works too).

[Veykril]

rust-lang/rust#122037 might be at fault here. Presumably this broke something in our mir eval, commit range diff rust-lang/rust@7f9830b

[HKalbasi]

rust-lang/rust#122037 indeed needs some action, we need to consider function annotated with #[rust_intrinsic] or #[rust_safe_intrinsic] as intrinsic in addition to extern "rust-intrinsic" functions. But it is not clear to me how this simple example needs to call intrinsics. @DaniPopes did enum AA in your example has Debug implementation?

[Nikaidou-Shinku]

I encountered the same problem, in my case const FOO: [&'static str; 1] = ["bar"] would trigger the same problem.

[Noratrieb]

I'm taking a look at fixing this by handling the new intrinsics correctly.

[Noratrieb]

Uh, r-a's implementation of panic_fmt calls std::fmt::format... I guess that then in turn panics or something like that. I think it needs to call const_panic_fmt, just like rustc. #15222 introduced this with no information whatsoever 🤔 @HKalbasi
I suspect what happened is that something now cause std::fmt::format to panic, leading to pathological escpaing.
Because all the code is doing is escaping the panic message, and then escaping that, and escaping that etc, in just a few seconds we get gigabytes of backslashes, lmao.

[DrewMcArthur]

@Nilstrieb in case it's helpful at all, i added this comment to a closed issue before i realized that this and #16934 were both still open. memory is leaking for me whenever vs code tries to format-on-save, which sounds like it's related to your last comment here.

#16140 (comment)