tsan.suppressions

# ThreadSanitizer suppressions file for Couchbase
# https://github.com/google/sanitizers/wiki/ThreadSanitizerSuppressions

# In memcached we choose to ignore accesses to the conn structs from
# other threads when printing connection statistics. We accept the
# possible stale / outdated information as we don't want to add
# additional locking overhead / complexity in the common case.
race:connection_stats

# The auditd unit tests trigger a heap-use-after-free error, however Valgrind doesn't
# detect anything wrong and ThreadSanitizer is known to have issues with exceptions
# so I'm pretty sure this a false positive.
race:TestCrashRecoveryGarbeledDate

# memcached ISOTime::generatetimestamp is using libc mktime which updates
# the TZ env-variable through tzset (_internal). This shows up as a data-race.
race:tzset_internal

# ThreadSanitizer appears to report a false positive data race in
# operator delete via std::string::assign for std::string when using
# libstdc++ - see discussion at
# https://groups.google.com/forum/#!msg/thread-sanitizer/dZ3QiQE49ns/j7rKGHg08foJ
# Running with LLVM's own C++ library (libc++) makes the warning go
# away, so would seem to be either a bug in libstdc++, or a bug in
# ThreadSanitizer.
race:^std::string::assign(std::string const&)

# ThreadSanitizer appears to report a data race in GCC's basic_string
# implementation on _M_refcount: specifically on operator delete via
# std::string::_Rep::_M_dispose for std::string when using libstdc++ -
# report at: https://gcc.gnu.org/ml/libstdc++/2015-09/msg00000.html
# Suppressing this for now.
race:^std::string::_Rep::_M_dispose(std::allocator<char> const&)

# In forestdb plock_lock is invoked from a lock structure that is sure to not
# overlap with that of another thread by use of is_overlapped() test.
# However since this function is not trusted by ThreadSanitizer it reports
# a lock inversion since the locks in the race are from the same struct definition
# but different memory addresses. This is hence a false positive.
deadlock:plock_lock

# The memcached unit test 'memcached_logger_emfile_test' deliberately
# consumes all file descriptors during it's run. This has the
# consequence of preventing thread sanitizer's external symbol lookup
# from failing. In turn this means that it cannot check for
# symbol-based suppressions (e.g. the tzset_internal one above).
# Therefore just suppress anything for this unit test.
race:memcached_logger_emfile_test