DFSG4 and combined works [was: Anton's amendment]

[Anton Zinoviev <anton@lml.bas.bg>]

On Fri, Feb 03, 2006 at 05:16:24PM +0200, Anton Zinoviev wrote:
> 
> Our discussion became too complicated and I am not sure on what we
> agree and on what we disagree.  I will try to explain my current
> opinion in a separate message and if we have some disagreement we can
> continue from there.

I tried to find some criterion upon which we can easy determine
whether some license makes the compilation works impossible or not.
Unfortunately I could only find a list of different cases with
different proofs.  For example all proofs I gave in the former
discussion are valid for some licenses and invalid for others.

Here I will give another proof for a case which I am intentionally
making as close to the text of DFSG4 as possible.  For reference the
text of DFSG4 is:

   The license may restrict source-code from being distributed in
   modified form _only_ if the license allows the distribution of
   "patch files" with the source code for the purpose of modifying the
   program at build time.

Suppose we have a license X that makes use of this rule of DFSG.  In
particular the X license gives us only the following permissions with
respect to the source code:

   1. Permits to distribute and build unmodified copies of the source
      of the program.

   2. Permits to distribute "patch files".  It is not necessary to
      require the distribution of the "patch files" together with the
      original source code even though DFSG seams to allow such a
      requirement.

   3. Permits to use the "patch files" for the purpose of modifying
      the sources of the program at build time.

Suppose that the works A and B are covered by X and C is a combined
work based on A and B.  Under these conditions the sources of C can
not be distributed because theyr distribution form should have
simultaneously the form sources_of_A+patches_for_A and
sources_of_B+patches_for_B and this is impossible.  (The formal proof
of this impossibility is too abstract, but I hope the following
analisys will make it clear.)

During the previous discussion Matthew Garrett and Kalle Kivimaa
proposed some ways to distribute the sources of C.  Here I am going to
analise the proposed ways:

1. Distribute a build system that applies a patch against work A which
   turns it into a patch against work B.  Applying this results in
   work C.

You are allowed to distribute the original sources of A and a patch
which turns them into a patch against work B and the license of B may
permit to use such automatically generated patches.

Obviously any patch that is automatically generated in this way is a
work based on A.  The license of A permits to use "patch files" for
the purpose of modifying the sources of A at build time.  However the
license of A doesn't explicitly permit us to use "patch files" or any
other work based on A for the purpose of modifying the sources of
another program, in our case B.

2. We distribute the sources of C in the form sources_of_A +
   sources_of_B + patch_set_to_A_and_B.

>From the license of A it follows that patches_for_A == sources_of_B +
patch_set_to_A_and_B.  Consequently patches_for_A is a work based on
the sources of B.  Because of that the license of B does not allow us
to apply these patches to anything different than the original sources
of B.

Anton Zinoviev