GitLab

with fixing Conflicts to:
	src/bin/auth/auth_srv.cc
	src/bin/auth/auth_srv.h
	src/bin/auth/datasrc_clients_mgr.h
	src/bin/auth/main.cc
	src/bin/auth/tests/auth_srv_unittest.cc
	src/bin/auth/tests/command_unittest.cc

update the test case and log message accordingly.

the users of this method only need the set behavior, and the previous code
actually didn't swap them as expected anyway.

This reverts commit e6ba20b0, reversing
changes made to c9ba8373.

This reverts commit 81239246.

This should have gone to a branch, not to master. Reverting it until it
gets reviewed properly.

Unlike what the ticket suggests, it takes concrete origin parameter, not
some yet unknown config value, for two reasons:
 * The format of config is not known.
 * Leaking some details about what memory segment or whatever is being
   used for the cache through the interface looks like a bad idea. Even
   leaking the existence of the cache is enough.

as discussed in the ticket comments

This is necessary because run_dbutil.sh sometimes loses its execute
(+x) mode bit when regenerated, and tests fail due to it.

tryLock() is necessary inside InterprocessSyncFile. It also doesn't
make sense afterwards to make a dynamically constructed Mutex::Locker
that lives after the basic block is exited. So lock() and unlock()
have also been made public.

I've added comments to discourage the use of these methods directly
and use the Mutex::Locker where applicable.

This test fails now because there's no mutual exclusion among threads
in the implementation.