beta 1.33 seems to break tarpaulin on multithreading

[elinorbgr]

Consider the following minimal project:

Cargo.toml

[package]
name = "tarpotest"
version = "0.1.0"
authors = ["Me <me@me.me>"]
edition = "2018"

[dependencies]
futures-executor = "0.2.1"

src/lib.rs

#[test]
pub fn a() {
    futures_executor::ThreadPool::new();
}

#[test]
pub fn b() {
    futures_executor::ThreadPool::new();
}

Install tarpaulin from crates.io:

RUSTFLAGS="--cfg procmacro2_semver_exempt" cargo install cargo-tarpaulin

And run it on this small project using either stable (rustc 1.32.0 (9fda7c223 2019-01-16)) or beta (rustc 1.33.0-beta.5 (1045131c1 2019-01-31)) rustc:

• cargo +stable tarpaulin : works as expected
• cargo +beta tarpaulin: occasionally, (in roughly 15% of the runs), tarpaulin errors as the test executable segfaulted

Error: Failed to run tests! Error: A segfault occurred while executing tests

I have no idea what is at fault here, but apparently something in the last beta of rustc was changed which broke the way tarpaulin does its instrumentation. For more details, the segfault appears to only occur if at least 2 tests in the same binary involve spawning threads, and seem to be more likely the more threads are spawned.

See xd009642/tarpaulin#190 for full details.

[pnkfelix]

visiting for triage. P-high for initial investigation; assigning that task to @nagisa

[nagisa]

This will be hard for me to reliably bisect because among the failures related to sigsegv I also get

[ERROR tarpaulin] Failed to get test coverage! Error: Failed to run tests! Error: Error occurred when handling ptrace event: ESRCH: No such process

---

Bisection says that the sigsegv began in 2fba17f (#56837), cc @nikomatsakis @arielb1

The callstack looks like this:

#0  0x0000000000414e4f in futures_executor::thread_pool::ThreadPoolBuilder::create (self=0x7fffffffc170)
    at /home/nagisa/.cargo/registry/src/github.com-1ecc6299db9ec823/futures-executor-0.2.1/src/thread_pool.rs:262
#1  0x0000000000413d6f in futures_executor::thread_pool::ThreadPool::new ()
    at /home/nagisa/.cargo/registry/src/github.com-1ecc6299db9ec823/futures-executor-0.2.1/src/thread_pool.rs:78
#2  0x000000000040424d in tarpotest::b () at src/lib.rs:8
#3  0x0000000000404a4a in tarpotest::b::{{closure}} () at src/lib.rs:7
#4  0x00000000004047ae in core::ops::function::FnOnce::call_once ()
    at /rustc/7164a9f151a56316a382d8bc2b15ccf373e129ca/src/libcore/ops/function.rs:231
#5  0x000000000046105f in test::run_test::{{closure}} () at src/libtest/lib.rs:1475
#6  core::ops::function::FnOnce::call_once ()
    at /rustc/7164a9f151a56316a382d8bc2b15ccf373e129ca/src/libcore/ops/function.rs:231
#7  <F as alloc::boxed::FnBox<A>>::call_box ()
    at /rustc/7164a9f151a56316a382d8bc2b15ccf373e129ca/src/liballoc/boxed.rs:734
#8  0x00000000004cdbfa in __rust_maybe_catch_panic () at src/libpanic_unwind/lib.rs:92
#9  0x000000000047f188 in std::panicking::try ()
    at /rustc/7164a9f151a56316a382d8bc2b15ccf373e129ca/src/libstd/panicking.rs:276
#10 std::panic::catch_unwind () at /rustc/7164a9f151a56316a382d8bc2b15ccf373e129ca/src/libstd/panic.rs:388
#11 test::run_test::run_test_inner::{{closure}} () at src/libtest/lib.rs:1430
#12 0x000000000047eaad in test::run_test::run_test_inner () at src/libtest/lib.rs:1452
#13 0x000000000047d22c in test::run_test () at src/libtest/lib.rs:1471
#14 0x0000000000475c8d in test::run_tests () at src/libtest/lib.rs:1150
#15 test::run_tests_console () at src/libtest/lib.rs:957
#16 0x000000000046d764 in test::test_main () at src/libtest/lib.rs:290
#17 0x000000000046dfb2 in test::test_main_static () at src/libtest/lib.rs:324
#18 0x0000000000404276 in tarpotest::main ()
#19 0x0000000000404ad0 in std::rt::lang_start::{{closure}} ()
    at /rustc/7164a9f151a56316a382d8bc2b15ccf373e129ca/src/libstd/rt.rs:64
#20 0x00000000004bbfb3 in std::rt::lang_start_internal::{{closure}} () at src/libstd/rt.rs:49
#21 std::panicking::try::do_call () at src/libstd/panicking.rs:297
#22 0x00000000004cdbfa in __rust_maybe_catch_panic () at src/libpanic_unwind/lib.rs:92
#23 0x00000000004bcbc6 in std::panicking::try () at src/libstd/panicking.rs:276
#24 std::panic::catch_unwind () at src/libstd/panic.rs:388
#25 std::rt::lang_start_internal () at src/libstd/rt.rs:48
#26 0x0000000000404aa9 in std::rt::lang_start (main=0x404260 <tarpotest::main>, argc=2, argv=0x7fffffffdc08)
    at /rustc/7164a9f151a56316a382d8bc2b15ccf373e129ca/src/libstd/rt.rs:64
#27 0x00000000004042ad in main ()

I won’t be able to look into it much more at the moment.

[arielb1]

I suppose I'll have to investigate this then. Hope I can find enough time.

[pnkfelix]

triage: speculatively assigning to @arielb1 . They should hopefully unassign themselves if they are unable to look into this.

[Mark-Simulacrum]

Realistically I expect this to be a regression because we have ~4 days left before we need to produce the stable artifacts.

[arielb1]

had some busy time, will try to look at this quickly

[arielb1]

This appears to be triggered by the breakpoints tarpaulin places, which makes me suspect there is some race condition here we somehow create in futures. This doesn't sound good.