Open petrochenkov opened 5 years ago
Tests failing with debuginfo-level-tests = 2
:
// Non-codegen tests
incremental/krate-inherent.rs
mir-opt/inline-closure-borrows-arg.rs
run-pass/issues/issue-29466.rs
// Codegen tests
codegen/adjustments.rs
codegen/alloc-optimisation.rs
codegen/c-variadic.rs
codegen/call-metadata.rs
codegen/dealloc-no-unwind.rs
codegen/function-arguments.rs
codegen/issue-32031.rs
codegen/issue-34947-pow-i32.rs
codegen/link_section.rs
codegen/match-optimizes-away.rs
codegen/move-val-init.rs
codegen/nontemporal.rs
codegen/refs.rs
codegen/repeat-trusted-len.rs
codegen/repr-transparent.rs
codegen/scalar-pair-bool.rs
codegen/stores.rs
codegen/union-abi.rs
codegen/vec-iter-collect-len.rs
Tests failing with debuginfo-level-tests = 1
:
// Non-codegen tests
incremental/krate-inherent.rs
run-pass/issues/issue-29466.rs
// Codegen tests
codegen/call-metadata.rs
codegen/link_section.rs
codegen/nontemporal.rs
cc @michaelwoerister @nagisa
Since debuginfo influences what LLVM IR the compiler generates, I'm not surprised. I think that codegen tests should not inherit any debuginfo settings from the environment.
A data point: I was using debuginfo-level-tests = 1
(or the previous equivalent) for a little while on a NixOS system and was experiencing a failure on run-pass/issues/issue-29466.rs
. I posted about it on Zulip at the time.
I am somewhat doubtful this is actionable for the codegen or assembly tests but it is probably worth examining for the non-codegen tests. Interesting, none of the mir-opt tests seem to have trouble anymore.
debuginfo-level-tests = 1
debuginfo-level-tests = {1,2}
debuginfo-level-tests = {1,2}
debuginfo-level-tests = 1
debuginfo-level-tests = 2
debuginfo-level-tests = 2
Discovered while working on https://github.com/rust-lang/rust/pull/60568.
The lists of failing tests are below, testing was done on
x86_64-pc-windows-gnu
target.