Closed nikic closed 1 month ago
We should remove this test and reopen the issue, especially if it's true that nobody understands why the codegen improved: https://github.com/rust-lang/rust/issues/111508#issuecomment-1962529985
"Nobody understands" as in "nobody bothered to run a bisect".
This test has been consistently passing with our own LLVM 19 builds, so it's pretty weird that it fails with Ubuntu's.
@nikic is it possible that a commit was merged in the past... uhh... 5 months that fixed a behavior that affects this?
I think that Ubuntu's LLVM 19 may actually be a snapshot from April.
That would certainly be bad -- where did you find that it is from April?
It looks like Ubuntu 24.10 has the final release since yesterday: https://launchpad.net/ubuntu/+source/llvm-toolchain-19/1:19.1.0-1ubuntu1 I can't find a history on their website though.
Possibly just regenerating the docker image will fix this issue.
The log from https://github.com/rust-lang/rust/pull/129974 has:
2024-09-14T18:47:39.7301296Z #10 11.43 Get:213 http://archive.ubuntu.com/ubuntu oracular/universe amd64 llvm-19-dev amd64 1:19.1.0~++rc4-5ubuntu1 [47.3 MB]
So the docker image should currently have rc4, which is close to what our fork currently uses.
ah, I misunderstood something, then. I was looking at their "llvm-toolchain-snapshot" and trying to figure out its relationships to their other llvm repos and which branch would be in use: https://launchpad.net/ubuntu/+source/llvm-toolchain-snapshot
and I was having trouble finding the version string in the log, too. it seems it isn't always printed?
I've tried running this test in a loop in the x86_64-gnu-llvm-19 docker container and it never failed :( So no idea what's going on here.
Yes, I tried similar antics right before I put up the PR to remove it. I actually probably would have left it if I could have reproduced a failure, because then I'd have any hope of it actually being solved by anything except magic.
On https://github.com/rust-lang/rust/pull/130389 this test failed, then passed, then failed. That PR modified rustc_codegen_llvm, so maybe it's a problem with the PR, but...
On https://github.com/rust-lang/rust/pull/130641 the test failed as well. And LLVM update should have no impact on the llvm-19 job, so this is an unrelated failure.
Test: https://github.com/rust-lang/rust/blob/master/tests/codegen/issues/issue-111508-vec-tryinto-array.rs