Running cargo test causes a segfault.

[tekul]

The details are as reported in denoland/deno#19926. Essentially we are using deno embedded in a project and started seeing segfaults after a particular version. I noticed that I saw the same issue when running cargo test in deno, and also in deno_core (anything with v8 after version 0.74.3).

As described in my last comment in that issue, I get a segfault running cargo clean followed by cargo test with the latest deno_core 6a369d5.

Thread 46 "modules::tests:" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7ffff63f36c0 (LWP 36038)]
0x00005555569e23da in v8::internal::ThreadIsolation::RegisterJitPage(unsigned long, unsigned long) () at ../../../../v8/src/common/code-memory-access.cc:362

The machine is running Arch Linux:

$ uname -a
Linux celeste 6.5.3-arch1-1 #1 SMP PREEMPT_DYNAMIC Wed, 13 Sep 2023 08:37:40 +0000 x86_64 GNU/Linux

I also get the same error on a different laptop using the same OS.

[lucacasonato]

This is a bug on Linux on modern Intel & AMD CPUs. We are aware of the issue, and are planning to address it soon. @tekul running individual tests should succeed - it's just running the entire suite that fails.

[juli1]

Thanks for reporting it. This bug also impacts our project, which crashes on x86.

[juli1]

Bump this, this is impacting users and there is still no solution today.

[tekul]

A workaround we found if your project is using deno in rust tests is to use the ctor crate to make sure the runtime is initialized before the tests are run. i.e. in lib.rs add:

#[cfg(test)]
use ctor::ctor;

#[cfg(test)]
#[ctor]
// Make sure deno runtime is initialized in the main thread in test executables.
fn init_deno_runtime() {
    deno_core::JsRuntime::init_platform(None);
}

[Digifox03]

This was fixed by #471