Open yuxianch opened 2 years ago
The valgrind errors seem to be bogus. The first one explicitly points at two accesses to an atomic variable. The second one seem to be pointing at the TLS key access from different threads, while it is guaranteed that it is initialized first within a call_once
before it is used in other threads.
Possible data race is found when running test util_ipc_reliable_mq.cpp in boostorg/log. See https://github.com/boostorg/log/issues/162 for more details. Valgrind shows the following errors for this test: