LTO triggers type debuginfo unification error in LLVM

[luqmana]

This is with a recent rustc that includes #14819 and a new snapshot. Exact line used to build rustc:

../configure --disable-docs --disable-valgrind --prefix=/scratch/laden/local --target=i686-unknown-linux-gnu,arm-unknown-linux-gnueabi,arm-unknown-linux-gnueabihf,arm-linux-androideabi --android-cross-path=/scratch/laden/android-dev/standalone --enable-debug
RUSTFLAGS="-g" VERBOSE=1 make -j32

-> % rustc -v
rustc 0.11.0-pre (bb06790 2014-06-19 09:26:24 +0000)
host: x86_64-unknown-linux-gnu

fn main() {
    println!("Hello World!");
}

-> % rustc foo.rs -Z lto
rustc: /scratch/laden/rust/src/llvm/lib/CodeGen/AsmPrinter/DwarfUnit.cpp:938: llvm::DIE* llvm::DwarfUnit::getOrCreateTypeDIE(const llvm::MDNode*): Assertion `Ty == resolve(Ty.getRef()) && "type was not uniqued, possible ODR violation."' failed.
[1]    18332 abort      rustc foo.rs -Z lto

cc @michaelwoerister

[michaelwoerister]

Thanks for the report and for making sure that it occurs with a recent version. I was already able to reproduce it and will take a look at it next week. I suspect source file and line information of inlined items to be the culprit. But only closer investigation will show the real reason...

[michaelwoerister]

A little update on this: This is probably fixed by PR #15199 but another issue (#15156) prevents testing this properly (one has to manually fix LLVM in order to work around that). Let's leave the issue open until the other issue is fixed.

[michaelwoerister]

Update

While the original issue indeed seems to have been fixed with #15199, there's another problem occuring now (60e7317) with the same test case:

rustc: /home/mw/rust/src/llvm/lib/CodeGen/LexicalScopes.cpp:179: llvm::LexicalScope* llvm::LexicalScopes::getOrCreateRegularScope(llvm::MDNode*): Assertion `DISubprogram(Scope).describes(MF->getFunction())' failed.

Note that this is the same error as in #17201. It seems that LTO still triggers the problem, even if all crates involved contain debuginfo.

[michaelwoerister]

As the original problem has been fixed, I'll close this issue and open a new one for the new problem. I also renamed the issue to better reflect what the original error was.