[SciPy-User] Central File Exchange for SciPy

Fernando Perez fperez.net@gmail....
Sat Oct 30 21:32:09 CDT 2010

2010/10/30 Jochen Schröder <cycomanic@gmail.com>:
> Secondly, the argument that most Python code is already BSD, one could
> just as well make the argument that most OSS code is GPL so use GPL.

It's not an argument of majority, it's one of free flow of code across
projects and of reciprocity and fairness.

The GPL has an asymmetric relationship re. BSD code: gpl projects can
incorporate all the bsd code they want, but bsd projects can't
incorporate gpl code (without relicensing, which is often impossible
when there are many copyright holders, and is in any case a major
burden on a project).  This asymmetry is at the heart of this
discussion: numpy, scipy, matpotlib, mayavi, ipython and most of the
open source projects around here are BSD-licensed and it means we can
all freely share code across all of them (and we do, very often,
freely copy pieces from one to the other as needed, this is not a
hypothetical statement).  In fact, I relicensed ipython from its early
LGPL license (the one that I'm probably happiest with *personally*) to
BSD precisely based on this argument of free flow of code across
projects, made by John Hunter at the time.  And I'm glad I did, as
we've been able to copy code at various points in time across projects
without any worries.

When an author takes a piece of BSD code, modifies or builds upon it,
and makes the new work available as GPL (something I've sadly seen
done many times), he's most certainly *not* behaving in a spirit of
reciprocity towards the author of the original BSD code.  The BSD
author can no longer benefit from the improvements to his code:
despite the fact that those improvements remain open source, they are
no longer available to him unless he relinquishes his original license
terms and switches to the GPL.  I find that practice actually worse
than building proprietary extensions on open source code, because when
this is done typically companies at least are doing some other
business-related stuff that the open source developers are unlikely to
engage in.

> Furthermore your argument also ignores the fact that if you're using
> (ctypes, cython) wrappers around C-code you will probably be bound by
> the licence of the C-library so some code might not have a choice.

In this case obviously there's no choice and no argument either, but I
don't think anyone here is ignoring it, as it's the most basic ground
truth of any licensing discussion.

> Finally the biggest problem I have is with the notion that forcing a
> specific OSS choice onto developers is ok. If someone chooses a licence,
> they have a reason to do so and it is their choice. The funny thing is
> that the "free software crowd", often gets accused of this, however I've
> found that often the BSD crowd is a lot worse, and often quite hostile
> towards GPL licensing. Anyway I don't want to start a licence flamewar.

Nobody is *forcing* anything onto anyone.  A community is free to say:
if you want to use our tools, these are our terms.  This is a
community that shares code under the terms of the BSD license and sets
up a website for that purpose.  The rest of the whole internet is
available to anyone who wishes to publish GPL improvements to Numpy
and Scipy, just not on the Scipy servers :)

My personal opinion is that in the long run, it would be beneficial to
have this 'file exchange' have BSD-only code (or public domain, since
employees of the US Federal government as far as I understand must
publish their codes under public domain terms).  The reason is simple:
snippets put there, when good, are prime candidates for integration
into numpy/scipy proper.  It would be a shame, and frankly somewhat
absurd, to have a bunch of great codes sitting on the scipy server
that we couldn't integrate into scipy.  At least it seems so to me...



More information about the SciPy-User mailing list