On Nov 5,  8:59am, chuq@chuq.com (Chuck Silvers) wrote:
-- Subject: Re: CVS commit: src/lib/libc/rpc

| hi,
| 
| part of this change is:
| 
| +#ifdef _REENTRANT
| +                       if (errno == EINTR) {
| +                               sigset_t rmask;
| +                               if (sigpending(&rmask) == -1) {
| +                                       cu->cu_error.re_errno = errno;
| +                                       release_fd_lock(cu->cu_fd, mask);
| +                                       return cu->cu_error.re_status =
| +                                           RPC_SYSTEMERROR;
| +                               }
| +                               (void)sigsuspend(&rmask);
| +                       }
| +#endif
| 
| 
| why did you add this part?  if we got EINTR, then the signal was already
| delivered and thus won't be pending anymore.  plus, why would it make sense
| to set the mask to exactly the signals that are already pending?  this would
| require receiving a different signal than any already pending.  if the
| intent was to unmask any pending signals, this seems wrong too, because
| the pending signal could have been masked by the application before calling
| the RPC code, and the library shouldn't allow signals that the application
| has intentionally blocked.  I think this block should just be removed.
| 
| also, it looks like sigpending() needs a threaded version in libpthread too.
| 
| -Chuck

I thought that the signal would be still pending at this point. I will remove
it.

christos