Re: spurious connection_read()s

[Aaron Richton <richton@nbcs.rutgers.edu>]

> workaround is in place, slapd won't segfault on Solaris systems when it
> hits the bug, it'll just log that message.

Well, I never saw the behavior (on Solaris 9) that the <8 reports
described; I never recall having slapd segv in the connection handler,
multicpu or not. It seems odd that I'd be triggering a log message from a
workaround for a situation that I never encountered.

> Sun might not consider their odd handling of file descriptors a bug.  The
> condition already returns an undocumented error code, so apparently they
> think something of it.

I'm really not sure what's undocumented. Howard's message on the subject
(http://www.openldap.org/lists/openldap-software/200602/msg00127.html)
points to ITS #3567, where followup 14 (Sun's response) claims that
something is bogusly select()ing on a door and getting garbage in, garbage
out. Unless there's some data proving that Sun's contention is wrong,
"garbage in" seems to be a worthwhile theory. (Certainly there are doors
open in my slapd: I use syslog, just as the case Sun writes of.)