Should we instead make sure dlerror() is reset appropriately by Async-Profiler? I believe this kind of failures could also happen in different parts of the code, and is not specific to this test

Failure here: https://github.com/async-profiler/async-profiler/actions/runs/14613212540/job/40995410369

I think a better fix (and a more canonical way to check for errors) will be not to call dlerror() unless the previous function fails, i.e. returns NULL.

I think a better fix (and a more canonical way to check for errors) will be not to call dlerror() unless the previous function fails, i.e. returns NULL.

If you mean that we should not always call dlerror() in the test code, the dlerror buffer can be full even with no detectable failure for the code calling Async-Profiler API, or even with no failure at all, e.g. when calling asprof_execute for a non-JVM process. Am I missing something?

Should we instead make sure dlerror() is reset appropriately by Async-Profiler?

I would think we can log such errors, or ignore them purposefully by calling dlerror from async-profiler. Ideally, returning from asprof_execute should not leave a dlerror that was already handled or ignored.

@krk @apangin moved the fix in Async-Profiler main source set: f5ae5ce

I mean, calling dlerror() alone is not a proper way of checking errors. dlerror() should be only called if the function that you are testing fails. So I'd modify ASSERT_NO_DLERROR() and its usages instead.

How do you determine if a dl function is failed, without calling dlerror? The man example always calls dlerror: https://linux.die.net/man/3/dlopen

How do you determine if a dl function is failed, without calling dlerror?

By checking return value. dlopen and dlsym return NULL when fail. If they return non-null, they've succeeded, even though dlerror may return junk.

That sounds good, but in the example we can see that dlerror is called regardless of dlsym return value:

   *(void **) (&cosine) = dlsym(handle, "cos");

   if ((error = dlerror()) != NULL)  {
        fprintf(stderr, "%s\n", error);
        exit(EXIT_FAILURE);
    }

@krk @apangin what's the conclusion here?

We can change the tests not to be too strict about dlerror return values, as long as dlopen/dlsym functions do not return null.