Uninitialized bytes get copied to the buffer when creating an uninitialized tuple

[MSxDOS]

In this example:

use std::mem::MaybeUninit;

extern {
    fn fill(p: *mut u16);
}

pub unsafe fn test() {
    let (mut buffer, mut buffer2): ([u16; 512], [u16; 256]) = MaybeUninit::uninit().assume_init();
    fill(buffer.as_mut_ptr());
    fill(buffer2.as_mut_ptr())
}

there is a call to memcpy for one of the buffers that copies uninitialized data to it.
https://rust.godbolt.org/z/z1zcor

[MSxDOS]

cc @nikic

This started after #82806 was merged.

Also, this is a regression from stable to nightly.

[nikic]

Upstream fix: llvm/llvm-project@5556660

[pietroalbini]

This slipped up to stable 1.52.0, apparently we didn't label it regression-from-stable-to-beta losing track of it :/

[nikic]

@pietroalbini We don't plan to address this prior to the next LLVM upgrade anyway. It's just an optimization issue.

[apiraino]

Assigning priority as discussed in the Zulip thread of the Prioritization Working Group.

@rustbot label -I-prioritize +P-high +T-compiler

[nikic]

Fixed by #87570, new IR does not have memcpy:

define void @_ZN5test24test17h0962739e00df972bE() unnamed_addr #0 {
start:
  %buffer2 = alloca [256 x i16], align 2
  %buffer = alloca [512 x i16], align 2
  %0 = bitcast [512 x i16]* %buffer to i8*
  call void @llvm.lifetime.start.p0i8(i64 1024, i8* nonnull %0)
  %1 = bitcast [256 x i16]* %buffer2 to i8*
  call void @llvm.lifetime.start.p0i8(i64 512, i8* nonnull %1)
  %2 = getelementptr inbounds [512 x i16], [512 x i16]* %buffer, i64 0, i64 0
  call void @fill(i16* nonnull %2)
  %3 = getelementptr inbounds [256 x i16], [256 x i16]* %buffer2, i64 0, i64 0
  call void @fill(i16* nonnull %3)
  call void @llvm.lifetime.end.p0i8(i64 512, i8* nonnull %1)
  call void @llvm.lifetime.end.p0i8(i64 1024, i8* nonnull %0)
  ret void
}