office-gobmx/tsan-suppress.txt

#
# This file is part of the LibreOffice project.
#
# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at http://mozilla.org/MPL/2.0/.
#

# Use via environment variable TSAN_OPTIONS=suppressions=.../tsan-suppress.txt

# There looks to be a lock ordering problem here, but I can't see how it could
# actually be hit in practice.
deadlock:cppuhelper::ServiceManager::disposing()
deadlock:cppuhelper::ServiceManager::loadImplementation
deadlock:AffineBridge::v_callInto_v

# Ignore stuff in external DBUS library.
# Some kind of dbus lock issue when we call it from psp::CUPSManager.
deadlock:_dbus_lock
race:g_dbus_connection_signal_unsubscribe
race:g_path_get_basename
race:g_date_time_format_iso8601

# inside an assert so I don't care.
race:cppu::OWeakConnectionPoint::acquire
race:AffineBridge::v_enter
race:__vsnprintf_chk

# right now, I'm not interested in deadlocks at all, too many false+
deadlock:

# This is checking SAL_STRING_IS_STATIC, which is safe because that is written at compile time.
# Depending on the optimisation/debug level, the call stack may have different leaves, hence
# specifying this in more than one way.
race:rtl::str::acquire<_rtl_uString>
race:rtl::str::release<_rtl_uString>
race:rtl_uString_acquire
race:rtl_uString_release
race:rtl_uString_assign

# I've convinced myself this is a false+, caused by ping-ponging the buffer between two
# threads, but I might be wrong.
race:XBufferedThreadedStream::getNextBlock

# I am not interested in stuff the embedded JVM does.
race:libjvm.so

# I think this is OK, because at this point we are doing
#  if (nRefCount > 1)
# and we know from our callers that the refcount must be at least one
# so there is no failure mode.
race:ireallocSequence

# TODO There appears to be a race here, initialising the
#   ::com::sun::star::uno::Sequence< T >::s_pType
# field. But no idea at all how to fix it.
race:cppu::getTypeFavourUnsigned


# This is all inside GIO/Glib, no idea what it is doing.
#
race:slab_allocator_alloc_chunk
race:g_source_destroy_internal
race:g_source_unref_internal
race:g_task_finalize
race:g_socket_send_message_with_timeout
race:g_idle_source_new
race:g_slice_alloc0
race:g_str_has_suffix
race:g_str_equal
race:g_strsplit
race:g_variant_store

# Python build, lock order inversion
deadlock:take_gil

# Not interested in deadlock issues when they involve the shutdown path 
# - the shutdown path is special, so if we see a lock-ordering here that conflicts with
# a "normal" lock-ordering path, that is highly unlikely to result in a real issue.
deadlock:DeInitVCL

# Sometimes tsan will report data-races in these, who knows why, they should be thread-safe
race:malloc
race:free
race:close

# There is a lock-cycle here, together with the mutexes acquired in
#    SvtSysLocaleOptions::SvtSysLocaleOptions
#    SvtSysLocale::SvtSysLocale
# but I can't see how it could lead to a problem in practice
deadlock:ItemHolder1::impl_addItem

# ignore warning about nRefCount in sal_Sequence.
# <sberg> noelgrandin, we generally assume that sal_Int32 is small enough, and hardware sane enough,
#      that reading it will always give a sane value, i.e., if the read happens in parallel with a write,
#      the read produces either the old, original value, or the newly written one.
race:cppu::createEmptySequence
race:uno_type_sequence_reference2One