Slicing after [u8]::first performs unnessessary bound check

[youknowone]

Playground link: https://play.rust-lang.org/?version=stable&mode=release&edition=2018&gist=2e2d548aa3a7b89bfb074456d971450d

pub fn boundcheck(mut lit: &[u8]) -> Option<&[u8]> {
    if lit.first() == Some(&b'_') {
        lit = &lit[1..];
        // lit = unsafe { lit.get_unchecked(1..) };
    }
    Some(lit)
}

This slice always exists because of the branch, but it performs bound check.

Real case: RustPython/RustPython#2077

[Frizi]

Interestingly, following code seems to be properly optimized

pub fn boundcheck(mut lit: &[u8]) -> Option<&[u8]> {
    if let Some(&b'_') = lit.first() {
        lit = &lit[1..];
    }
    Some(lit)
}

[rodrimati1992]

Another way to write that today:

pub fn boundcheck(mut lit: &[u8]) -> Option<&[u8]> {
    if let [b'_', rem @ ..] = lit {
        lit = rem;
    }
    Some(lit)
}

Before Rust 1.42.0:

pub fn boundcheck(mut lit: &[u8]) -> Option<&[u8]> {
    if let Some((&b'_', rem)) = lit.split_first() {
        lit = rem;
    }
    Some(lit)
}

[nikic]

Based on a local test #74533 fixes this issue as well.

[tesuji]

Closed by #74533 .
The new asm generated by nightly (ceedf1d 2020-08-08):

check::boundcheck:
 mov     rax, rdi
 test    rsi, rsi
 je      .LBB0_1
 mov     rdx, rsi
 cmp     byte, ptr, [rax], 95
 jne     .LBB0_4
 add     rax, 1
 add     rdx, -1
.LBB0_4:
 ret
.LBB0_1:
 xor     edx, edx
 ret

[youknowone]

Nice to know this will be fixed. Thank you all.
And thanks for the advice @rodrimati1992