atexit(3) vs dlclose(3) about cxx-abi HOT 6 OPEN

https://bugzilla.redhat.com/show_bug.cgi?id=28625#c5 There was a discussion on updating standards.. but I don't know what standards in particular and whether they were updated.

from cxx-abi.

CC @drepper

from cxx-abi.

The topic of the subject (atexit vs dlclose) seems more like something for POSIX to sort out, not the C++ ABI standard.
Note that there are POSIX implementations where dlclose has no effect by design. Even implementations which implement library unloading cannot do so under all circumstances. C++ is particularly tricky in this regard: There could be otherwise harmless symbol interposition which prevents unloading. Implied destructor references, particularly for std::shared_ptr and TLS data, occasionally result in bug reports because they generally do not prevent unloading. Another thorny subject is the order of destructors when dlclose is called. There are so many open questions and technical difficulties I fear it's hopeless to specify the semantics of dlclose.

from cxx-abi.

C++ ABI standard mentions in https://itanium-cxx-abi.github.io/cxx-abi/abi.html#dso-dtor-motivation

The API specified below is intended to provide standard-conforming treatment during normal program exit, which includes executing atexit-registered functions in the correct sequence relative to constructor-registered destructors, and reasonable treatment during early DSO unload (e.g. dlclose).

It's barely defined in my opinion in this specification.. but not fully.

from cxx-abi.

Actually there are people who claim based on this spec that atexit(3) callbacks must be called upon dlclose(3).

Please make it clearer.

from cxx-abi.

ping?

from cxx-abi.