[Numpy-discussion] Does float16 exist?
Wed Jan 9 06:43:15 CST 2008
Charles R Harris wrote:
> On Jan 8, 2008 11:26 PM, David Cournapeau
> <firstname.lastname@example.org <mailto:email@example.com>>
> Charles R Harris wrote:
> > The C++ name mangling can be worked around.
> name mangling is just the top of the iceberg. There are problems
> wrt to
> static initialization, exception, etc...; ABI compatibility is much
> easier to break in C++ than in C. And then, there is the problem
> of C++
> ability of compilers.
> One would naturally have to be very careful about which features to
> use. In particular, don't use anything that throws, and I suppose one
> would want to avoid having to link to the stdc++ library also. I don't
> know if that is possible.
Avoiding linking to the stdc++ would be quite difficult, I think (from
the build process, at least). Certainly doable, but difficult.
> I agree on this point (custom code generators use). Note that
> there are
> standard tools to generate simple templates in C (autogen).
> Unfortunately, autogen is GPL and I suspect NumPy would be regarded as
> a derived work. Are you familiar with any other standard, maintained
> program of that sort with a BSD style license?
I don't think generated source files should be considered as GPL
(otherwise, you could not build non GPL source with gcc either: I don't
see where the difference would be between object code generated by the
compiler and generated source
by autogen). If you read the point 0 of the GPL, it looks quite clear to me:
"Activities other than copying, distribution and modification are not
covered by this License; they are outside its scope. The act of running
the Program is not restricted, and the output from the Program is
covered only if its contents constitute a work based on the Program
(independent of having been made by running the Program). Whether that
is true depends on what the Program does."
Once you get the generated source file, you do not need autogen at all
anymore, which is a pretty strong hint that it is not derived work,
IMHO. Of course, IANAL, blah blah blah.
More information about the Numpy-discussion