To: new-httpd@hyperreal.com
Subject: Re: AddModule patch
References: <199609130029.UAA10731@shado.jaguNET.com>
From: Tom Tromey <tromey@creche.cygnus.com>
Date: 12 Sep 1996 19:00:25 -0600
In-Reply-To: Jim Jagielski's message of Thu, 12 Sep 1996 20:29:29 -0400 (EDT)
Message-ID: <m1u3t3w7o6.fsf@creche.cygnus.com>
Lines: 38
Sender: owner-new-httpd@apache.org
Precedence: bulk
Reply-To: new-httpd@hyperreal.com

Jim> I like the concept of these patches, but I (being devil's
Jim> advocate) have to question how much they would really be
Jim> used. The main advantage I would see, as has been mentioned,
Jim> would be for the precompiled binaries. However, I doubt if the
Jim> serious webmaster would keep the "bloated" binary, but would
Jim> eventually simply recompile with whatever modules needed to
Jim> reduce the memory footprint.

Actually, I tend to agree with you!

However, some people here have argued that the bloated binaries won't
be a problem in practice, because modern OSs will be smart about not
paging in the sections of code which go unused.  Anyone have a
perspective on this?  Any war stories about a bigger-than-necessary
executable causing a real problem?  I'd like to know: I actually favor
the mod_dld-style approach (despite what I say below), and this would
provide me with some useful ammunition.  Unfortunately my guess is
that in practice they are right, and it doesn't matter.

>From a binary distribution perspective, this is one of two workable
(meaning: no recompile necessary; see below) solutions.  The other,
using something like mod_dld, is more of a pain (to put it mildly) to
get working everywhere.  (If there are other solutions, I'd like to
hear about them!)

I believe we anticipate that most of our clients will not recompile.
I think this has been our experience with the compiler side of the
company -- and our compiler clients are probably on the whole a lot
more technically savvy than our internet clients.

In terms of bloat/maintenance cost, my changes don't really add much:
a couple new arrays, a few new (small) functions.  Not the most
convincing argument... does anyone else doing binary distributions
find this feature desirable?

Tom
-- 
tromey@cygnus.com                 Member, League for Programming Freedom