Subject: Re: mmap
To: Andreas Wrede <andreas@planix.com>
From: Tom Ivar Helbekkmo <tih@nhh.no>
List: current-users
Date: 11/13/1998 09:10:37
Andreas Wrede <andreas@planix.com> writes:

> The configure script for the Cyrus imapd server ver 1.5.14 runs a
> test program to "Verify the OS supports the memory mapping semantics
> needed by map_private".

I've been using Cyrus on NetBSD/i386 for a while, and I know for a
fact that it does things with mmap() that don't work right there.
Even if you let it configure for "reduced functionality" use of
mmap(), which it will do (at least at 1.5.10, which is what I run
right now), it won't work reliably.  I have to manually override it by
editing the config.status file to include

s%@WITH_MAP@%nommap%g

...and then rerun it.  Works fine, but less than optimal, of course.
John Friend suggested that there must be a bug in the NetBSD mmap()
support somewhere, because he felt certain that the code in Cyrus for
"careful" use of mmap() in a non-integrated vm and buffer cache
environment should be correct.

The way I tested it was to run a bunch of deliveries into the same
mailbox in parallell, and the failure mode was that not all messages
were delivered.

-tih
-- 
Popularity is the hallmark of mediocrity.  --Niles Crane, "Frasier"