Boost logo

Boost Users :

From: Cliff Diamond (cliff.diamond_at_[hidden])
Date: 2022-12-21 17:10:55


Hi

I'm using boost::detail::spinlock (boost-1.81) in code that compiles on both Visual Studio and gcc. I've hit a problem on the gcc build where the underlying spinlock member (v_) in spinlock_gcc_atomic.hpp isn't an atomic variable and isn't explicitly initialised (BOOST_DETAIL_SPINLOCK_INIT is defined, but there's no initializer and no constructor).

v_ is declared as a non-atomic unsigned char and atomic operations use gcc intrinsics in that file. So v_ needs to be explicitly initialized to 0. When the lock method is called, it just spins as a result.

This may be work-in-progress or am I misunderstanding? It seems such a glaring omission. The STL version of the spinlock uses std::atomic which does initialize v_ because it's wrapped up in the atomic template.

Also there is a scoped_lock nested class within the boost spinlock. Why have an additional lock guard? Surely std::scoped_lock should be sufficient (or is it there just for testing - if so please comment it).

I can overcome the problem (without modifying spinlock_gcc_atomic.hpp) by trying to acquire the lock when I create it and releasing it regardless.

Is this a bug in that file?

Thanks

Cliff



Boost-users list run by williamkempf at hotmail.com, kalb at libertysoft.com, bjorn.karlsson at readsoft.com, gregod at cs.rpi.edu, wekempf at cox.net