Plimpton's FFT library is licensed under GPLv2 not BSD-3

[felker]

Summary of issue

The software licensing topic came up again during my development of #263, specifically regarding the prospect of adding a third-party library for Gaussian-Legendre quadrature #263 (comment).

Athena++ is licensed under the 3-clause Berkeley Software Distribution (BSD) license:
https://github.com/PrincetonUniversity/athena/blob/91421866c865c10ad053c243e7e8d63167ab1d30/LICENSE
As far as I understand, this is a requirement from some national laboratories since they refuse to work with open-source software that has a copyleft license.

However, I noticed that Plimpton's Parallel FFT Package, which we keep an (unmodified?) copy of in src/fft/plimpton/ is actually licensed under GNU General Public License (GPL) 2. In fact, we keep a copy of this license text in our repository:
https://github.com/PrincetonUniversity/athena/blob/91421866c865c10ad053c243e7e8d63167ab1d30/src/fft/plimpton/LICENSE

Even though this library comes from Sandia National Laboratory, I believe we are in violation of the GPLv2 license by distributing Athena++ under BSD-3 while src/fft/plimpton/ is contained in our repository.

@tomidakn raised the possibility of creating a new top-level directory ext/ where the source code for such external dependencies could be added via patches, but not provided with the main Athena++ repository. This might also be a better overall practice for storing source code that was not authored by Athena++ developers. @jmstone mentioned that the mixing of Steve Plimpton's source code in src/ was less-than-ideal. @changgoo, you mentioned that src/fft/plimpton/ was not verbatim Plimpton's library, but a modified version written by Nicole Lemaster?

We would not need to worry about this for other software like SUNDIALS that are licensed under BSD-3 like Athena++.

[changgoo]

@changgoo, you mentioned that src/fft/plimpton/ was not verbatim Plimpton's library, but a modified version written by Nicole Lemaster?

yes, Nicole made changes to make it only works with FFTW.

[felker]

Even worse, I am realizing that Athena++ (and a ton of files in Athena 4.2) is likely in violation of the Numerical Recipes license. See Boycott Numerical Recipes for a discussion.

The NR source code cannot be redistributed, nor can derivative works such as the code adapted to double precision in:

athena/src/utils/ran2.cpp

Lines 17 to 20 in 318b1b1

	//----------------------------------------------------------------------------------------
	//! \fn double ran2(std::int64_t *idum)
	// \brief Extracted from the Numerical Recipes in C (version 2) code. Modified
	// to use doubles instead of floats. -- T. A. Gardiner -- Aug. 12, 2003

There are no exceptions for noncommerical or scientific/academic use. At a minimum, we cannot relicense it under the BSD-3 header. I think we need to replace double ran2(std::int64_t *idum).

"recipes" appears in the comments of 6x more Athena++ files for relativistic calculations in the context of what I believe are all root-finding applications. I think these files are fine, since the source code is not pulled from NR; the comments merely identify steps of the generic algorithm. Can you confirm this @c-white?

[c-white]

I wouldn't use NR code even if it were public domain -- it's pretty terrible all around. Those are just convenient formula references for the quadratic formula (including the form no one knows about) and cubic equation roots.

[changgoo]

Just realized that the new version of the Plimpton's parallel FFT library is noe released with BSD license. If I port it, it will resolve the issue. I just tested the performance and there's no improvement on it, which gave me little motivation for porting it. But, at some point, I may need to port the new version anyway with somewhat large modifications.