I discovered this while trying to call sentry_init, sentry_shutdown and sentry_capture_event multiple times from multiple threads at the same time during unit tests, not a valid use-case scenario of this library.

But I can imagine a valid use case to be when sentry_shutdown is called while other threads are still busy using sentry_capture_event.

---

Basically what happens is that functions like sentry_capture_event do this:

They check if it's valid here:

And then use it multiple times here:

The issue arises when on a different thread sentry_shutdown is called which invalidates g_options and makes access to it access freed memory.
So the missing part here to me seems a missing lock around g_options_mutex.

---

Another example is sentry_start_session:

So SENTRY_WITH_SCOPE_MUT locks g_lock:

But sentry_shutdown locks a different mutex, g_options_mutex:

Which doesn't block sentry_start_session from using already freed memory:

---

There are more offenders similar to those.
Is there an intention to account for this scenario?