Regression in type inference between stable 1.79.0 and nightly-2024-06-19

[marmeladema]

Code

I tried this code in nightly:

pub fn parse(bytes: &[u8]) -> String {
    std::ffi::CStr::from_bytes_until_nul(bytes)
        .unwrap()
        .to_str()
        .unwrap()
        .split('\n')
        .map(str::trim_end)
        .collect::<Box<_>>()
        .join("\n")
}

https://play.rust-lang.org/?version=nightly&mode=debug&edition=2021&gist=be71e0fb8185c11ac3123de70b02cee6

I expected to see this happen: it compiles successfully

Instead, this happened:

error[E0282]: type annotations needed
 --> src/lib.rs:8:10
  |
8 |         .collect::<Box<_>>()
  |          ^^^^^^^ cannot infer type of the type parameter `B` declared on the method `collect`
9 |         .join("\n")
  |          ---- type must be known at this point
  |
help: consider specifying the generic argument
  |
8 |         .collect::<Vec<_>>()
  |                 ~~~~~~~~~~

For more information about this error, try `rustc --explain E0282`.
error: could not compile `playground` (lib) due to 1 previous error

Version it worked on

It most recently worked on: Rust 1.79.0

Version with regression

nightly-2024-06-19

[marmeladema]

Minimized a bit:

pub fn parse(input: &str) -> String {
    input
        .split('\n')
        .map(str::trim_end)
        .collect::<Box<_>>()
        .join("\n")
}

[veera-sivarajan]

@rustbot label -needs-triage -regression-untriaged +regression-from-stable-to-nightly +T-compiler +A-inference

[apiraino]

Bisection points to #99969 cc @dtolnay @calebsander

Can this issue be considered an intended or rather an unintended behaviour?

I can reproduce in beta so I'll flip the regression labels

[calebsander]

I think this is expected. With the additional implementation of collect() into Box<str>, it's now ambiguous whether Box<_> means Box<[_]> or Box<str>. It's unfortunate to regress existing code that wasn't being explicit, but I think this is an acceptable cost for the additional functionality.