Path: utzoo!mnetor!uunet!husc6!bloom-beacon!athena.mit.edu!wesommer
From: wesommer@athena.mit.edu (William E. Sommerfeld)
Newsgroups: comp.windows.misc
Subject: Re: Best window system & Why
Message-ID: <3004@bloom-beacon.MIT.EDU>
Date: 18 Feb 88 05:03:46 GMT
References: <1681@desint.UUCP> <42080@sun.uucp> <2994@bloom-beacon.MIT.EDU> <3146@phri.UUCP>
Sender: daemon@bloom-beacon.MIT.EDU
Reply-To: wesommer@athena.mit.edu (William E. Sommerfeld)
Organization: Massachusetts Institute of Technology
Lines: 54

In article <3146@phri.UUCP> roy@phri.UUCP (Roy Smith) writes:
>> Of course, it would be a lot easier for people to develop informed
>> opinions on NeWS if Sun were to freely release the source for the server
>Given a choice between buggy source and high
>quality binaries, I'll take the binaries any day.  (And no, I'm not saying
>that X is buggy.)

Given the choice between buggy binary distributions and doing without,
I'll do without. 

In our environment at Athena[0], we'd rather have buggy source than
"perfect" binaries because we can probably fix things we have source
for if they become a problem.  

There's _no such thing_ as a nontrivial bug-free program.  (TeX comes
close, but very few companies are as good at quality control as Donald
Knuth). 

If we have the source we can at least understand the real nature of
the bug (rather than making guesses based on its erroneous behavior,
which are typically completely wrong), have a better idea of how to
make it reproducible, and also probably be able to fix it. [1]

When you consider how hard it is to coordinate binary distributions to
two completely different systems (MicroVAX and IBM RT), this is a real
problem.  We are under a large amount of pressure from our donors (DEC
and IBM) to make the environment on both classes of workstations
identical.  Binary-only distributions and highly restrictive
non-disclosure agreements on sources only get in the way of this.

By the way, after reading parts of the NFS source, I feel real sorry
for those of you out there who run binary-only Sun sites..  I was
tempted to include a hex dump of a few "interesting" packets that will
crash your average Sun RPC based service, but discretion prevailed. 

I used to think that binary-only distributions were sort of OK, but my
work here has convinced me that it's not worth my time to have to
fight with software that I don't have the source for.

					- Bill

[0] ~700 workstations and growing: 1/2 DEC MicroVAXes running
4.3BSD+NFS, 1/2 IBM RT's running 4.3BSD+NFS, being managed by a crew
of less than 100 people, of which at most 10-20 could really be
considered "wizards". 

[1] If we at least have a symbol table, we can disassemble parts of
the binary and at least have an idea as to what is going on (for
example, when we discovered that the date conversion routine of a
certain database system[2] goes into an infinite loop on the first of the
year only _after_ we built our service management database using
it..), but that's rather painful to do in the general case. 

[2] Well, since you asked, RTI Ingres version 5.