I know the locking gets exercised (both libldap and GnuTLS sides) during connection and TLS handshake, so I tested this by doing a bind/search/unbind loop on a bunch of parallel threads.

I built master with this patch in all of the following environments:

Debian 8 (jessie), GnuTLS 3.3.30 Debian 9 (stretch), GnuTLS 3.5.8 Debian 10 (buster), GnuTLS 3.6.7 Debian unstable (sid), GnuTLS 3.6.9

FreeBSD 12.0, GnuTLS 3.6.8 Solaris 11.4, GnuTLS 3.5.16, --with-threads=lwp Solaris 11.4, GnuTLS 3.5.16, --with-threads=posix

Windows 10 (MinGW), GnuTLS 3.6.8, --with-threads=nt Windows 10 (MinGW), GnuTLS 3.6.8, --with-threads=posix

and ran my test program successfully on all of them.

On each of these I also stepped into the debugger and confirmed that the expected calls happened (gnutls_system_mutex_lock calling resp. pthread_mutex_lock or EnterCriticalSection, ldap_pvt_thread_mutex_lock calling the expected API, etc).

Any further comments, or suggestions for additional tests I should run?