Missed optimization on only used once allocation

[tesuji]

godbolt link: https://rust.godbolt.org/z/d_VvSo
Here are two snippets that I expect that they should have the same assembly:

pub fn foo(s: &str) -> bool {
    s != "FOOFOOLONGLONG"
}

pub fn foo(s: &str) -> bool {
     s.to_string() != "FOOFOOLONGLONG"
}

EDIT: For C version, gcc and clang also fails to optimized this case.

[tesuji]

The problem with this is that the second snippet is used rather often.
For example:

pub fn foo() -> bool {
    std::env::var("FOO").unwrap_or("0".to_string()) != "BAR"
}

In this case, if you need better code generation, you may write:

#![feature(result_map_or)]
pub fn foo() -> bool {
    std::env::var("FOO").map_or(false, |s| s != "BAR")
}

[Ixrec]

FWIW, there's already a clippy lint to tell you to replace things like unwrap_or(...) with unwrap_or_else(|| { ... }). for precisely this reason. I believe this is also one of the reasons why so many Option/Result methods have a _then or _else variation that takes a closure instead of a value.

[tesuji]

Yes, the version with unwrap_or_else has better codegen than one with unwrap_or,
but they still has worse codegen than the one doesn't allocate.