libgit2/script/thread-sanitizer.supp at main · libgit2/libgit2

26 lines (23 loc) · 1.34 KB

# In attr_file_free, the locks are acquired in the opposite order in which they

# are normally acquired. This is probably something worth fixing to have a

# consistent lock hierarchy that is easy to understand.

deadlock:attr_cache_lock

# git_mwindow_file_register has the possibility of evicting some files from the

# global cache. In order to avoid races and closing files that are currently

# being accessed, before evicting any file it will attempt to acquire that

# file's lock. Finally, git_mwindow_file_register is typically called with a

# file lock held, because the caller will use the fd in the mwf immediately

# after registering it. This causes ThreadSanitizer to observe different orders

# of acquisition of the mutex (which implies a possibility of a deadlock),

# _but_ since the files are added to the cache after other files have been

# evicted, there cannot be a case where mwf A is trying to be registered while

# evicting mwf B concurrently and viceversa: at most one of them can be present

# in the cache.

deadlock:git_mwindow_file_register

# When invoking the time/timezone functions from git_signature_now(), they

# access libc methods that need to be instrumented to correctly analyze the

# data races.

called_from_lib:libc.so.6

# TODO(#5592): Investigate and fix this. It can be triggered by the `thread`

# test suite.

race:git_filter_list__load_ext

Provide feedback