libgit2中的全局操作需要同步吗？

Question

我一直从cache_store那里得到赞的报告。

文档中说，我需要防止在每个对象之上并发访问，但根据TSAN的说法，我将其解释为对每个git_repository对象的并发访问。所有派生对象都需要使用相同的锁(例如git_commit、git_tree等)。对吗？

TSAN产出：

WARNING: ThreadSanitizer: data race (pid=28177)
  Atomic write of size 4 at 0x55bc2dc780ec by thread T14 (mutexes: write M889877933370572712):
    #0 __tsan_atomic32_fetch_add <null> (libtsan.so.0+0x6c101)
    #1 git_atomic_add external/libgit2/src/thread-utils.h:82 (analyze_remote_git+0x46e240)
    #2 cache_store external/libgit2/src/cache.c:200 (analyze_remote_git+0x46ed4c)
    #3 git_cache_store_raw external/libgit2/src/cache.c:231 (analyze_remote_git+0x46eec0)
    #4 odb_read_1 external/libgit2/src/odb.c:1095 (analyze_remote_git+0x413b73)
    #5 git_odb_read external/libgit2/src/odb.c:1116 (analyze_remote_git+0x413cdf)
    #6 git_object_lookup_prefix external/libgit2/src/object.c:221 (analyze_remote_git+0x4056ab)
    #7 git_object_lookup external/libgit2/src/object.c:252 (analyze_remote_git+0x4057d1)
    #8 git_commit_lookup external/libgit2/src/object_api.c:23 (analyze_remote_git+0x40f980)
    #9 create_branch external/libgit2/src/clone.c:41 (analyze_remote_git+0x37c582)
    #10 create_tracking_branch external/libgit2/src/clone.c:102 (analyze_remote_git+0x37c880)
    #11 update_head_to_new_branch external/libgit2/src/clone.c:124 (analyze_remote_git+0x37c94e)
    #12 update_head_to_remote external/libgit2/src/clone.c:189 (analyze_remote_git+0x37cc5c)
    #13 checkout_branch external/libgit2/src/clone.c:315 (analyze_remote_git+0x37d1b1)
    #14 clone_into external/libgit2/src/clone.c:348 (analyze_remote_git+0x37d460)
    #15 git__clone external/libgit2/src/clone.c:432 (analyze_remote_git+0x37da83)
    #16 git_clone external/libgit2/src/clone.c:463 (analyze_remote_git+0x37dbfc)
... ADDITIONAL STACK REMOVED ...

  Previous read of size 4 at 0x55bc2dc780ec by thread T13 (mutexes: write M1852):
    #0 cache_store external/libgit2/src/cache.c:192 (analyze_remote_git+0x46ec31)
    #1 git_cache_store_parsed external/libgit2/src/cache.c:237 (analyze_remote_git+0x46ef20)
    #2 git_object__from_odb_object external/libgit2/src/object.c:148 (analyze_remote_git+0x405314)
    #3 git_object_lookup_prefix external/libgit2/src/object.c:244 (analyze_remote_git+0x405747)
    #4 git_object_lookup external/libgit2/src/object.c:252 (analyze_remote_git+0x4057d1)
    #5 git_tree_lookup external/libgit2/src/object_api.c:56 (analyze_remote_git+0x40fb0f)
    #6 tree_walk external/libgit2/src/tree.c:978 (analyze_remote_git+0x3a22f5)
    #7 tree_walk external/libgit2/src/tree.c:989 (analyze_remote_git+0x3a237e)
    #8 tree_walk external/libgit2/src/tree.c:989 (analyze_remote_git+0x3a237e)
    #9 git_tree_walk external/libgit2/src/tree.c:1025 (analyze_remote_git+0x3a2545)
... ADDITIONAL STACK REMOVED ...

  Location is global 'git_cache__current_storage' of size 4 at 0x55bc2dc780ec (analyze_remote_git+0x0000011d60ec)

  Mutex M889877933370572712 is already destroyed.

  Mutex M1852 (0x7b4400019f28) created at:
    #0 pthread_rwlock_init <null> (libtsan.so.0+0x2ef82)
    #1 git_cache_init external/libgit2/src/cache.c:72 (analyze_remote_git+0x46e61f)
    #2 repository_alloc external/libgit2/src/repository.c:245 (analyze_remote_git+0x416764)
    #3 git_repository_open_ext external/libgit2/src/repository.c:829 (analyze_remote_git+0x418cfe)
    #4 git_repository_open external/libgit2/src/repository.c:887 (analyze_remote_git+0x419127)
    #5 git_repository_init_ext external/libgit2/src/repository.c:2112 (analyze_remote_git+0x41ced5)
    #6 git_repository_init external/libgit2/src/repository.c:2056 (analyze_remote_git+0x41c97b)
    #7 default_repository_create external/libgit2/src/clone.c:234 (analyze_remote_git+0x37ce3e)
    #8 git__clone external/libgit2/src/clone.c:420 (analyze_remote_git+0x37d8cb)
    #9 git_clone external/libgit2/src/clone.c:463 (analyze_remote_git+0x37dbfc)
... ADDITIONAL STACK REMOVED ...

Answer 1

文档说，我需要在每个对象之上防止并发访问，但是根据TSAN的说法，我将其解释为对每个git_repository对象的并发访问。所有派生对象都需要使用相同的锁(例如git_commit、git_tree等)。对吗？

是。git_repository及其派生的所有对象只能由单个线程使用。

git_repository是轻量级的，它的目的是安全地与其他git_repository(以及其他进程中的其他git客户机，包括命令行中git的使用)进行交互。因此，您应该可以跨多个线程创建多个git_repository对象。