Does a GPL API infect its apps?

[Ken Arromdee]

On Thu, 21 Oct 1999, John Cowan wrote:
> > Here's a hypothetical version which does not have this problem:
> > >But when you
> > >distribute the same sections as part of a whole which is a work based
> > >on the Program, the distribution of the whole must be on the terms of
> > >this License or a license which grants permissions which are identical to this
> > >License except for the possible addition of extra permissions on the sections
> > >which are identifiable works as described above.
> This is already possible, since a distribution of the PART is not a
> distribution of the WHOLE.

I don't think so.  Call the GPL program A and the added part B.

The GPL on A only says what you can do with A+B.  It does not say what you can
do with B alone.  That *looks* like it's only restricting distribution of
the whole and not the part.

But one of the requirements of the GPL on A is that you also GPL A+B.  The
GPL on A+B restricts what you can do with derivatives of A+B (unlike the first
GPL on A, which only restricts what you can do with derivatives of A).  B 
(when obtained by cutting it from A+B) is a derivative of A+B, and therefore
B is restricted by GPL even though it contains no A code.

That is, the fact that you cannot take the section B out of the program and
use it under another license is a second-order effect.  The initial GPL on A
does not prohibit this action, but it requires that you GPL A+B, and GPLing A+B
in turn causes the prohibition.

If B is dual-licensed, that dual license got lost at the step where you
distributed A+B.  You must distribute A+B under GPL-only, which makes that
particular copy of B GPL-only too.  Any dual-licensed copies of B must be
obtained separately.