Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!swrinde!elroy.jpl.nasa.gov!ames!mindcraft.com!karish
From: karish@mindcraft.com (Chuck Karish)
Newsgroups: comp.unix.aix
Subject: Re: Making A request to IBM
Summary: Testing fixes
Message-ID: <670399929.14940@mindcraft.com>
Date: 31 Mar 91 06:12:08 GMT
References: <1991Mar19.152638.19508@athena.mit.edu> <3326@pensoft.UUCP> <1991Mar23.144908.9009@bellcore.bellcore.com> <3351@pensoft.UUCP>
Organization: Mindcraft, Inc.
Lines: 43

In article <3351@pensoft.UUCP> robin@pensoft.UUCP (Robin Wilson) writes:
>In article <1991Mar23.144908.9009@bellcore.bellcore.com>
jona@iscp.Bellcore.COM (Jon Alperin) writes:

>>2. When I call IBM and report a bug, I sometimes get a patch number. HOWEVER,
>>it has been my experience that the patch often "fixes" more items than the
>>one I was looking for. This is "great, until something that worked breaks",
>>and I don't know where to look....Does IBM distribute just a single fix?
>>How does one request just a single fix?
>
>For right now, there is no way to get a 'single' fix; except for emergency
>fixes.  (An emergency fix, is where there is no update available yet, but the
>code has been fixed.)
>
>Think about like this: IBM sends you a new xlc compiler.  2 months later, your
>replacement (since you are really talented, and have moved up to management)
>calls defect support and reports a bug in xlc.  How does IBM remember that you
>got a special fix to xlc 2 months ago.

This last is a valid problem, but the main reason is more closely
related to the concerns stated in the question:  It's difficult and
expensive to test fixes in isolation so they can be shipped
individually.  Such testing is needed to keep even a single "fix" from
breaking something else.

Watching the way IBM provides fixes, it looks to me as if, for AIX 3,
fixes are added to a reference copy of the OS which is tested more or
less continuously.  Patches aren't shipped until the build of the OS to
which the fixes have been made proves itself by running without
newly-created or newly-exposed bugs for a while.  Doing adequate
testing individually for each fixed problem would be prohibitively
expensive.  More important, it wouldn't work; fixes for different
problems often interact to produce new problems.

Getting back to Robin's answer, try to imagine the poor support
peoples' responsibilities under the suggested scheme:  How would you
characterize the state of your system well enough to let them try to
duplicate the behavior you see?  Would you expect them to apply the
particular set of fixes you've used, to a system freshly loaded with a
down-rev version of the OS?

	Chuck Karish		karish@mindcraft.com
	Mindcraft, Inc.		(415) 323-9000