Issues with Thread Waiting on Mac OS

[JerryHyun]

Problem Description

When testing the eCAL library on mac OS, I noticed that sometimes it will execute the pthread_cond_timedwait_relative_np() function and get stuck for a very long time (I don't know the exact reason).

"ecal/ecal/core/src/io/mtx/linux/ecal_named_mutex_impl.cpp"
"ecal/ecal/core/src/ecal_event.cpp"

#ifndef ECAL_OS_MACOS
          ret = pthread_cond_timedwait(&mtx_->cvar, &mtx_->mtx, ts_);
#else
          ret = pthread_cond_timedwait_relative_np(&mtx_->cvar, &mtx_->mtx, ts_);
#endif

Looking at the source code above, eCAL on mac os uses the pthread_cond_timedwait_relative_np() function.
Why are you using pthread_cond_timedwait_relative_np() and not the pthread_cond_timedwait() function? (I have confirmed that the pthread_cond_timedwait() function exists on mac OS as well).

How to reproduce

I'm not sure how to reproduce it.

How did you get eCAL?

Custom Build / Built from source

Environment

• eCAL: v 5.12.1
• OS : mac os v 14.1.2

eCAL System Information

No response

[KerstinKeller]

@hannemn do you remember?
I tried to track down the original commit that first added MacOS, but that was from a time where we synched two repos, and I was unable to pin it down. Most likely related to monotonic clocks or similar.
Does it work reliably, if changed to pthread_cond_timedwait ?

[JerryHyun]

I haven't tried replacing it with pthread_cond_timedwait yet.
I'll try to change it and test it.

[hannemn]

Hi @JerryHyun, thanks for the finding! I just had a quick look into the implementation of pthread_cond_timedwait_relative_np and seems to be case that we utilize the function in the wrong manner. Even though Apple declared the time parameter in the function signature as const struct timespec *abstime which suggested the argument is interpreted as absolute timepoint, it has to be actually a relative timespan according to the implementation. I will try to fix this in a generic manner to keep code section further on platform independent and will keep you posted once I have a first bugfix available.

[hannemn]

Can you please crosscheck with this branch https://github.com/eclipse-ecal/ecal/tree/hotfix/timedwait-relative-mac-5-12 (it's based on 5.12.x) if the blocking issue at pthread_cond_timedwait_relative_np still appears on your mac? I have unfortunately no mac hardware here available to test my code changes.

[JerryHyun]

Hi @hannemn
Yes, I see.
I'll test it and let you know if there are any issues.

[KerstinKeller]

@JerryHyun can you provide feedback on this issue? does the hotfix solve the problem?