Trivial vec![] creation then iteration doesn't optimize out the allocation

[bluss]

Rustc used to be able to replace vec![1, 2, 3].iter().sum::<i32>() with a constant when optimizing, not at the moment. It simply looks like a Layout method that is not inlinable.

(This is a regression strictly speaking — stable & beta can do this optimization, but not nightly.)

Code to reproduce:

playground link

pub fn sum_me() -> i32 {
    vec![1, 2, 3].iter().sum::<i32>()
}

[nodakai]

Are there any ways to write automated tests to check for "regressions" like this? Assertions against LLVM bitcode?

[kennytm]

@nodakai Yes codegen tests can be used for this, e.g. https://github.com/rust-lang/rust/blob/master/src/test/codegen/alloc-optimisation.rs. Assuming this is run against the bitcode after optimization, that is.

[Mark-Simulacrum]

cc @arielb1 -- seems LLVM related

[arielb1]

Nope. Just Layout::repeat (

rust/src/liballoc/allocator.rs

Line 210 in 0565653

pub fn repeat(&self, n: usize) -> Option<(Self, usize)> {

) that is not marked as inline.

[arielb1]

cc @alexcrichton

[arielb1]

I didn't confirm that making Layout::repeat as inline fixes this, so please confirm that/add a test before closing.