> OVS_RELEASES applies to both read-locks and write-locks? > > I guess that an ordinary mutex is considered a write-lock?
Yep, I suppose we could create a second #define for that. Not sure if it matters or not. > In lockfile.c, I guess that clang can only handle guarded-by on > pointers? Otherwise I don't see why one would introduce the new > lock_table variable as a level of indirection. Assuming that's true, > can we declare lock_table as a const pointer, e.g. > static struct hmap *const lock_table OVS_GUARDED_BY(lock_table_mutex) > = &lock_table__; > (That is, the pointer is const, not what it points to.) Yes, this is fairly annoying. Clang doesn't consider taking the address of a variable as a read. Technically they're right, but that opens the door to all kinds of unsafety. In my experience, this change catches bugs, so I think it's worth the ugliness. X-CudaMail-Whitelist-To: dev@openvswitch.org _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
