[AstroPy] SOFA license letter (draft)

[James Turner]

> Does anyone know for sure if the LGPL is BSD-compatible? If so, that's
> another option.
>
> Cheers,
> Tom

I'm a bit fuzzy on the details, apart from not being a lawyer etc,
but I believe that (as for the GPL) there's nothing to prevent you
combining them as long as you are prepared to follow the more
restrictive terms of the LGPL for the combination. Unlike the GPL,
those terms do allow your combined "work that uses the Library" to
have a different licence, which I believe means that downstream
recipients could still make a proprietary version of AstroPy, as
you deliberately allow them to with BSD -- but they would have to
release source code for any changes they make to the LGPL library
itself (in this case SOFA) under the LGPL.

A proprietary work based on AstroPy would also have to allow
relinking to a modified version of the LGPL library, which I
presume with Python would just mean (re)distributing the wrapper
code to make the Python .so module (or some object file version of
that wrapper). There are also a couple of requirements to include
notices about the dependency.

I see that SciPy does NOT accept LGPL code though, suggesting
that the business interests behind it have thought of some less
obvious angle (maybe they don't like the obligation to provide
source for LGPL dependencies or wrapper code; their reasoning is
simply that "the license of the code should not have more
restrictions than [BSD]"):

   http://www.scipy.org/License_Compatibility

I haven't thought hard about what constitutes a derivative work in
Python generally, when there's no difference between source and
executables (other than pyc files). Under some circumstances you
might be able to satisfy the GPL itself in the same way as LGPL,
but that would need further investigation; there seems to be some
general contention and legal greyness about what makes a
derivative work where dynamic linking is involved (eg. the painful
Linux kernel graphics drivers), so that can only apply even more
to Python.

Cheers,

James.