pthread_rwlockattr_setkind_np incomplete, has impacts

[Jeff Johnson]

There are problems with pthread_rwlockattr_setkind_np.

First, the manpage refers to three possible settings, but these settings are
absent from the LSB Data Definitions, and thus from the SDK headers, rendering
the function useless. From upstream pthread.h:

enum
{
  PTHREAD_RWLOCK_PREFER_READER_NP,
  PTHREAD_RWLOCK_PREFER_WRITER_NP,
  PTHREAD_RWLOCK_PREFER_WRITER_NONRECURSIVE_NP,
  PTHREAD_RWLOCK_DEFAULT_NP = PTHREAD_RWLOCK_PREFER_READER_NP
};

Second, the LSB manpage (which was also submitted to the Linux Manpages project
in 2010, although I don't see the page even in recent releases, or in the git
tree at git://git.kernel.org/pub/scm/docs/man-pages/man-pages.git) fails to
define all three types - there is no description of the behavior of
PTHREAD_RWLOCK_PREFER_WRITER_NP.

Third, since the behavior of thread read-write locks is changed by manipulating
the attributes object in this way, the POSIX description of the behavior of
read-write locks, which is referenced directly by LSB through the various
pages, is no longer completely accurate, and LSB should have a delta manpage
indicating the difference. I have not looked at all of the pages, but
certainly the combined pthread_rwlock_rdlock, pthread_rwlock_tryrdlock page
should have a note on the impact of changing this "kind" attribute. Indeed,
the Linux manpages should reflect this difference as well - and as far as I can
see the entire read-write lock set is only documented by way of imported POSIX
pages.