Re: gnu readline / license

[Hans-Bernhard B. <br...@ph...>]

On Wed, 2 Jun 2004, Ethan Merritt wrote:

> I do not believe that the GPL restricts gnuplot from being
> linked against libreadline.  

It doesn't restrict the link as such, but it does forbid distributing
the resulting binary, because the gnuplot Copyright contains some
restrictions that would violate the readline's GPL.

> Debian apparently believes otherwise, but that is their problem, not
> ours. 

> 1) gnuplot is clearly not a derived work of libreadline. I don't
> think anyone is arguing this point.

A binary version of gnuplot linked against readline pretty definitely is
such a derived work.

> 2) It is my understanding that restrictions imposed by the
> GPL do not cross a well-defined API boundary.  Therefore mere
> linkage to  an independent (and optional) shared library does
> not by itself contravene the GPL.

That library is by no means "optional" for a binary linked against
-lreadline.  If the user doesn't have libreadline.so on his box, such a
binary would fail to start.  Such behaviour pretty strictly contradicts it
being optional, by my book.

The only exception from this might be if we dlopen() libreadline.so
ourselve, at runtime, and are prepared to live with it not being
available.

>   The most extensive discussion
> on this issue that I have followed has been on the linux kernel
> mailing list.  So far the broad (though not universal) consensus
> is that non-GPL binaries can link against the GPL linux kernel.

Careful there.  Unless I misremember, the Linux Kernel has special
exceptions for such cases, i.e. it's not pure GPL in this sense.

-- 
Hans-Bernhard Broeker (br...@ph...)
Even if all the snow were burnt, ashes would remain.