You were against a "special case" patch for dev-null - which I agree is not needed now that I know the "one true way"(tm)

What I was pondering was a general patch which would make communicating with filter daemons (listening on file sockets or char devices) possible - and that would be to skip the locking in these cases where it has no point - it would have the side effect of allowing /dev/null to work, but that's not my point here... It would allow direct delivery to things like spamd without a stub spamc program to manage the socket.