Path: utzoo!yunexus!telly!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!cs.utexas.edu!uunet!odi!benson
From: benson@odi.com (Benson Margulies)
Newsgroups: gnu.gcc
Subject: Re: Free [Software] Foundation
Message-ID: <377@odi.ODI.COM>
Date: 4 Jun 89 11:42:46 GMT
Article-I.D.: odi.377
References: <1107@mailrus.cc.umich.edu> <8906040445.AA04908@hop.toad.com>
Reply-To: benson@odi.com (Benson Margulies)
Distribution: gnu
Organization: Object Design Inc., Burlington, MA
Lines: 53

Some of us, at one time or another, are in the position of building
something based on someone else's unfree source. At my previous job,
it was Unix. In this circumstance, it is a major puzzle how to satisfy
both copyleft and someone else's requirements. FSF is entitled to be
unsympathetic. It is, however, a real situation. So some of us are not
picking nits at copyleft to subvert it, but rather to figure out how
it can fit with other, conflicting situations.

What follows might be called: the slippery slope of software
recombination:

1) I write code that knows how to fork and exec gnu make. 
I don't even distribute gnu make, I just tell my customers where to
get it. I don't distribute source of the system I built.

2) I supply diffs to gnu make (in source, of course).

3) I supply diffs to gnu make, and a set of instructions for building
it as a sun shared library. My proprietary code calls parts of it
via dynamic linking.

4) I supply a makefile which modifies and compiles some parts of gnu
make and links it in with my code.

The point of this is that there is, in my mind, a significant
difference between:

*) if you make changes to a FSF codebase that are generally
applicable, you are obligated to make them generally available. 

and

*) if you use so much as a particle of FSF code in a system of yours,
you must apply GPL terms to the entire system.

My problem with the second, which currently sure seems to be the
intent of the GPL, is that its mighty hard to define a "system". the
GPL offers little guidance on the definition of this term. The
singular exception for "operating systems" clarifies nothing.
Surely, the accident of the use of an RPC protocol versus fork and
exec versus dynamic linking versus static linking can't really define
a system. But if it dosen't, then what does?

ps: I'd appreciate not receiving nastygrams from FSF followers in
response to this, especially to FSF followers who don't have the
braincells to put valid addresses anywhere in their mail. If RMS or
Len Tower or someone else central would like to supply some insight
here, fine. If not, still fine. No one else's opinion of the FSF's
position is relevant.


-- 
Benson I. Margulies