Ok, we've spun in circles for quite a while.
Here are three separate things, which actually form a coherent
argument. Do not assume that these are rehashes of things I've said
before; in various ways they represent some change of mind about some
things. Please read the whole post before composing replies to any
part.
FIRST:
"Debian shall remain 100% free software". This sentence is ambiguous,
and can be interpreted in two ways:
INTERPRETATION A: "Everything in Debian shall be '100% free
software'". That is, everything in Debian shall be software, and
shall be 100% free.
INTERPRETATION B: "All the software in Debian shall be 100% free".
This interpretation leaves entirely unsaid what should be the status
of things which are not software.
The interpretation Branden seems to prefer is a third one, which I
think is untenable:
BRANDEN'S INTERPRETATION (as tb understands it): Everything in
Debian, whether software or not, must be 100% free; nothing
restricts Debian to have only software.
So I prefer interpretation B. But note that I don't think we should
be neutral about the status of things which are not software: I think
that freeness is an important goal, and we should consider it
carefully for *all* things, and not just software.
Interpretation A is a reasonable way to read the sentence, but it's
not Debian's practice by any stretch of the imagination. There have
always been many things in Debian which are not software, but are
intended as pure data.
Branden's interpretation is not an unreasonable goal to have about
what should be in Debian; I just don't think it's a correct reading of
the sentence from the DFSG.
It thus seems to me that we must make a *decision* about the status of
things not software, keeping in mind our strong bias in favor of all
things free, and not trying to say that the DFSG has already forced
our hands.
WARNING: If you disagree about what I say about the DFSG, but you
agree that there is some invariant text legitimately in main (even if
only copyright ascriptions), then you can ignore section the FIRST
here, and move on to what follows.
SECOND:
The only kind of unfreeness we are talking about is invariancy of
non-program text. So I'm only going to focus on that in what follows.
It is universally agreed that Debian packages may contain invariant
text. But exactly where the boundary lies differs from person to
person. Here is a rough scale:
A. Only copyright statements themselves can be invariant.
B. Only copyright statements and associated licenses can be invariant.
C. Only Copyright statements, licenses, giving-credit-where-
credit-is-due, and no-warranty requirements can be invariant.
D. Small amounts of text can be invariant if they are not
documentary. (A "documentary" text is one which needs to change
if the associated software changes.)
E. Small amounts of text of any kind can be invariant.
F. Any amount of text of any kind can be invariant.
Now, I don't know of anyone who holds positions A, E, or F.
Some people have said things that suggest that they like statement B.
But statement B would exclude GPL'd and old-style BSD copyrighted
software. Since those are paradigm cases of free software, I'm going
to skip statement B, and assume that people who have advocated B
really mean something like C.
THIRD:
So, if I've gotten the options well, there are really two options
floating around. (Many of you have been reading impatiently and
saying this all along, but if you read through to the end [please be
patient], you'll see there's a reason for the way I'm writing this.)
The two options are:
OPTION ONE: Allow copyright ascriptions, giving-of-credit,
no-warranty, and license texts to be invariant.
OPTION TWO: Allow small amounts of any kind of nondocumentary text to
be invariant.
(A secondary issue has been to discuss the proper definition of
"small" here, but here I take it to be something that would allow a
GNU Manifesto sized invariant text in an Emacs Manual sized document.)
When I first wrote this post out, I made an interesting error. I
omitted "no-warranty" from the list of OK things in option one. But
that has to be allowed, under the rule that of course the GPL is free,
but the GPL requires invariant no-warranty texts in general. So I
realized "oops, I left that out" and went back and added it back in.
But that's precisely the problem with option one. It's got a list of
four kinds of acceptable invariant sections and it's perfectly
reasonable to think a new kind will get thought of later.
Recognizing that option one is "brittle" might lead people to the
following:
MODIFIED OPTION ONE: Little bitty amounts of nondocumentary text can
be invariant.
The difference between option two and modified option one is that
option two allows "small" amounts, which is something like 30K in the
paradigm case, and "little bitty" amounts, which is something like
only one kilobyte: enough for no-warranty and such, but not more.
Even that isn't quite right, more modification might be needed to deal
with expanding lists of change records; still, I think modified option
one can be made to work to avoid the problems of enumeration that show
up in option one.
FOURTH:
The difference between modified option one and option two seems like
it's just "how much text", but that would be an unfair attempt to
elicit support for option two. The point of "little bitty" is to
exclude big political statements, novels, etc., on the assumption that
such things are not little bitty. So there is a real difference
between modified option one and option two, which must be kept in
mind. Just remember that modified option one is an attempt to keep
option one without suffering an enumeration problem.
(I mention this because I could have tried to argue "hey, modified
option one and option two just differ in the amount of text"; that
would be a deceptive argument on my part, and I want to make as
explicit as I possibly can that I don't intend any such argument.)
One example of a requirement that doesn't impinge freeness, but isn't
in option one, is something that says "you must preserve the notice
that the unmodified version of this software can be found at web site
foo." Such a statement is sometimes found, is not included in option
one, but would be allowed by modified option one (as it should be).
A different kind of modified option one would be to try and give a
definition of "incidentals of licensing"; I think that kind of attempt
is not likely to succeed. But if you think it is likely to succeed,
then go ahead and read further, taking that sort of definition as
"modified option one".
FIFTH:
So, the question now boils down to:
Should we allow bigger things (but still "small" amounts) to be
invariant, aside from things that seem like incidentals of licensing?
Here, I refer back to the question "why do we care about modification
of software?" and the answer is that such modifications are important
to user's abilities to *use* the software, and to have a free market
for support for the software. (Remember RMS's famous story about the
printer software he wasn't allowed to see and modify.)
I think my sentiments are entirely on Branden's side about
modifiability of text: I really do think the world should be free of
copyright and people should have an untrammeled right to copy, modify,
and share all kinds of text (of course, fraud being excluded).
However, I don't think that Debian needs to pursue that goal. Our
goal is narrower: free software. That's only one category, and I
think we should do what we can to advance that category--where indeed
we have a fair prospect of success.
Freedom to modify software is exceedingly important in the *use* of
the software. Similarly, freedom to modify documentary text is just
as important. But nondocumentary text is not as important to
software, and I don't think we have any need to inherently exclude it
all. I don't think it's compromising important principles to have
some.
However, it does still bother me to have it. That's why I think it
should be limited to "small" amounts. Importantly, we should say
"we'd rather not have invariant text at all, but we are willing to
distribute some, if it's small, if that will encourage the development
of more good free documentation for free software programs."
This is kind of like the LGPL--it's a compromise of a less-important
principle in order to hopefully get people to do more of the more
important parts.
Thomas
