Join GitHub today
GitHub is home to over 50 million developers working together to host and review code, manage projects, and build software together.
Sign upIntermittent segfault in servo_config/net/profile unit tests #26192
Comments
jdm
commented
Apr 15, 2020
|
I think this must be a regression from the latest rustc upgrade, since that's the first time we observed a unit test segfault. |
|
Running
|
|
I can reproduce that result with Running |
|
|
Aha! When I run with cargo test directly without RUST_BACKTRACE=1, it works. When I add RUST_BACKTRACE=1, it segfaults. |
|
CARGO_TARGET_X86_64_APPLE_DARWIN_RUNNER=lldb yields:
|
|
When I comment out the non-cookie tests, I get this:
|
|
The only change to the libbacktrace in use in libstd between the 3/16 rustc nightly and the 4/9 nightly is rust-lang/rust@64a8c8a#diff-1a7024d46c31d728a04d5e0cafd98df8 (based on https://github.com/rust-lang/rust/commits/3712e11a828af2eea273a3e7300115e65833fbc5/src/libstd/Cargo.toml). |
|
We're still hitting this segfault in the opts tests after #26217:
|
|
I have verified that the segfaults in profile and servo_config are also triggered by #[should_panic] tests. That just leaves the openssl segfault that I previously found. |
|
In an attempt to make a testcase that is more palatable than "run servo's unit tests", I've successfully reproduced the crash by adding a panic to https://github.com/Kethku/neovide/blob/56cc29d47ccbff7cf823a738b3f7f4a31e8af26d/src/window.rs#L150. |
|
Filed rust-lang/rust#71397. |
|
@jdm issue has been solved on rust, running test with rust backtrace does not segfault on latest master. |
|
Good point! |