Charles Hannum wrote:

# ab>
# ab> 	int f = open(NULL, O_RDWR|O_CREAT|O_UNLINK, 0);
# ab>
# ab> which would give me back something that had a link count of zero.
#
# Making this a standard feature is a *terrible* idea.  If it wasn't
# politically infeasible, I would have made I/O to unlinked files fail
# years ago.  This sort of hack makes it nearly impossible to analyze WTF
# is going on when something breaks.

How is it a "hack"?  It'd be more of a hack to suddenly reclaim resources
on an unlinked file.  The separation between the in-core metadata and the
on-disk metadata was, as far as I can tell, deliberately designed into
the OS.  I/O to unlinked files was the initial form of IPC; even pipes
were on-disk buffers as opposed to in-core for the longest time.

ktrace/ktruss in the present day makes it much less nearly impossible
to analyze WTF is going on when something breaks (especially in the
verbosity provided by NetBSD -- I can't be as polite about Solaris).

				--*greywolf;
--
NetBSD: For The Network Generation