Spurious test failure `codegen-llvm/issue-107681-unwrap_unchecked.rs`

[JonathanBrouwer]

On quite a few PRs, this test failed spuriously:
#158462 (comment)
#155521 (comment)
#158493 (comment)
#152225 (comment)

More affected prs: https://github.com/rust-lang/rust/pulls?q=is%3Apr+issue-107681-unwrap_unchecked.rs

Zulip discussion thread: #t-compiler/help > Spurious test failure in `codegen-llvm/issue-107681.rs`

[workingjubilee]

To summarize our current findings from that discussion thread:

#158067 was the PR that introduced the sporadic failures

cc @TimNN @dianqk

The issue is probably due to randomize-layout in the codegen-llvm suite.

The test was made more imprecise in a way that is probably unnecessary, while being made more precise in ways that are brittle in the face of randomize-layout.

The line

    // CHECK: [[INNER:%.*]] = load ptr, ptr %x

is not resilient to the load sometimes being from %x and sometimes being from a pointer assigned to %0 that is itself GEP'd from %x

Revert posted at #158502

[TimNN]

Thank you for the revert and sorry for the trouble, I wasn't aware of the Layout randomization part.

[workingjubilee]

Honestly, it should not have been trouble, so it's quite fine. We probably need to try to figure out how to make the layout randomization not engage for codegen-llvm tests, but the Zulip thread also describes why that is not a quick 1-minute fix.