From mboxrd@z Thu Jan  1 00:00:00 1970
To: 9fans@cse.psu.edu
Subject: Re: [9fans] bug fix to oct 14 release
From: "rob pike" <rob@plan9.bell-labs.com>
MIME-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
Message-Id: <20001019154715.63AB2199DF@mail>
Date: Thu, 19 Oct 2000 11:47:13 -0400
Topicbox-Message-UUID: 1795159e-eac9-11e9-9e20-41e7f4b1d025

The previous, 1995 release was put together by an elaborate script
that pulled things out of the main tree.  It was difficult to maintain
and required people to stop working as we neared the release date.  We
can't just tar up our tree because much of the stuff in our system is
proprietary or unreleasable for one reason or another; we must do an
intricate slice of the tree, eliminating some parts, adding others,
and making substitutions for still others.

This round, we tried a different and I think ultimately more
satisfactory method: we built a complete tree on the side, in the
final form for the distribution, and copied stuff over as it
developed.  Once it's ready, we just bundle it up and ship it.  This
has several advantages: it's easy; people can work in the main tree
while the distribution is being built; once the distribution tree is
ready, it is preserved in its final form; we can do full builds in the
test tree to make sure the distribution builds itself (very difficult
to do with the old method without installing the distribution package
from scratch on a private system); and so on.  The biggest advantage
of all is that this method makes it much much easier to update the
distribution; the prior method works well only for a one-off.

Scripts check the contents of the distribution tree with the main
tree, modulo various modifications, and catch most things that need
updating.  However, newly created stuff in the main tree must be
copied by hand, which is why most of the glitches you see are for
minor aspects of the system that are embodied in new files.  (The
tcp25 error was just sloppy.)

Building the newest update from the July one took maybe two hours of
my time to assemble and another 6 or so to test.  Compared to the
previous method, that's instantaneous.

-rob