From owner-freebsd-ports  Fri Jul 14 16:50:56 1995
Return-Path: ports-owner
Received: (from majordom@localhost)
          by freefall.cdrom.com (8.6.10/8.6.6) id QAA05313
          for ports-outgoing; Fri, 14 Jul 1995 16:50:56 -0700
Received: from silvia.HIP.Berkeley.EDU (silvia.HIP.Berkeley.EDU [136.152.64.181])
          by freefall.cdrom.com (8.6.10/8.6.6) with ESMTP id QAA05294
          for <freebsd-ports@freebsd.org>; Fri, 14 Jul 1995 16:50:48 -0700
Received: (from asami@localhost) by silvia.HIP.Berkeley.EDU (8.6.11/8.6.9) id QAA15744; Fri, 14 Jul 1995 16:50:38 -0700
Date: Fri, 14 Jul 1995 16:50:38 -0700
Message-Id: <199507142350.QAA15744@silvia.HIP.Berkeley.EDU>
To: knarf@nasim.cube.net
CC: freebsd-ports@freebsd.org
In-reply-to: <3u4ete$shi@nasim.nasim.cube.net> (knarf@nasim.cube.net)
Subject: Re: diffs and checksums
From: asami@cs.berkeley.edu (Satoshi Asami)
Sender: ports-owner@freebsd.org
Precedence: bulk

 * > Also, we should be able to save space on wcarchive (and mirror sites)
 * > by doing znew --best on all the distfiles.
 * 
 * I don't think it's worth the CPU usage using --best. But let us discuss
 * this later. ;)

I mean we put tarfiles that have been znew --best'd on wcarchive, and
every mirror site will pick them up.  So the CPU is used only on one
site (probably thud), and all the mirror sites can save space.  How's
that now? :)

 * > The only problem is, how?  I couldn't come up with a way to implement
 * > this without having to gzcat the file every time a user types "make
 * > <something>" ('cause most of the targets depend on "checksum"), or
 * > forcing a znew from bsd.port.mk.
 * 
 * Something like work/.checksum.done would cure this problem, wouldn't
 * it? Hmm, work/ may not exist at this time, then what about
 * distfiles/.filename.ok ? Again a bad idea if someone mirrores these
 * files. Create the work directory with the ckecksum target?

Um, you are confusing yourself here.  Why would someone mirror a file
like that unless I screw up and put that (as well as the distfiles) on
wcarchive? ;)

Anyway, the problem is that the tarfile may have changed without a
name change (that's really the whole point of checksumming).  Having a 
cookie won't really help alleviate this problem, because it won't
guarantee anything except that the checksum was once correct.  Think
of this sequence:

(1) Joe user sups ports/games/coolgame

(2) Joe user does a make, it fetches the tarfile, checksum is ok and
    builds the cool game

(3) Joe user sups ports/games/coolgame again (probably a nightly sup), 
    the original file has changed without name change and there is a
    new pre-install target.

(4) Joe user types make install, and it bombs because there is a
    checksum cookie but the extracted source doesn't match the
    Makefile anymore!

Satoshi