Re: GPL vs. sweatshops (was Re: Bias)

"Deven T. Corzine" <[email protected]> wrote:
> On Tue, 11 Jun 2002, Matt Seitz wrote:
> 
> > Sorry, I didn't make my point clearly enough.  Yes, the BSD offers the
> > possibility of creating a better compiler and making a profit from it.  But
> > Brett's argument was that so long as a no-cost, "good enough" compiler was
> > available, no one would ever pay enough to justify the creation of a better
> > compiler.  In other words, even if GCC went to a BSD license today, anyone
> > attempting to sell an improved version would never get enough customers to
> > justify the cost of development.
> 
> He could be right, but I tend to believe there's always likely to be a
> market for a better compiler that runs faster and/or uses less resources
> and/or makes faster code and/or makes smaller code.  Yes, the market for a
> "good enough" compiler is much larger, but there's probably enough market
> for a profit incentive for a proprietary improvement.

This is true, and I don't think Matt correctly described Brett's argument.  
Although a BSD version would put downward pressure to compiler vendors, they
can leverage off the "free" version and provide something that is better.

> Now, to program a free improvement under the GPL, on the other hand, I'm
> not convinced there's much incentive there.  There's incentive to port to
> new platforms to leverage all that code, but to improve the compiler
> without adding features?  It's probably not worth it to most companies to
> do it for themselves alone, and without a profit incentive, why do it for
> others, either?

Exactly.  If you want a real world example of this, look at gcc on linux vs
java.  Gcc is pretty much the only game in town.  You are stuck with its
profilers, debuggers and code generation.  java, on the other hand, even
though it's free (as in beer), has a wealth of diversity (forte, jbuilder,
not to mention the small vendors) in development tools (and jvms, profilers,
compilers, etc).  Any honest person cannot dismiss that GPL has a hand in
this situation as the GPL denies any party from carving out a nitch to attempt
to make a living.  
 
> Back around 1990-ish, GCC was generally considered to have one of the best
> code generators out there; it was often better than proprietary compilers.

well, one should qualify that.  It was the best 68K (maybe VAX) compiler out
there.  

> In recent years, I've heard numerous claims of proprietary compilers that
> outperform GCC on various platforms.  Now, surely that's partly because
> they can optimize for a given target platform, but could it also partly be
> that once GCC got to be "good enough", there wasn't incentive to make it
> lots better than that?

"good enough" isn't even the reason.  Look at SPARC64 - I'm not certain if
gcc 3.x(!) reliably generates code for that (though it must have safe code
paths as there are supposedly versions of BSD and linux that run on it - though
I can say for certain that 2.95.x certainly does NOT).  The real reason is that
as CPUs get more complicated, the compilers are also more complicated to make
and there aren't a huge number of people qualified to work on it.  And, given
the lack of financial incentive on working on it, it's easy to show why the gcc
is where it is today.

-r