Re: what DOES the GPL really say?

[dewar@merv.cs.nyu.edu (Robert Dewar)]

Ronald Cole said

<<I'd recommend that you find some other license agreement that would
prevent someone from enhancing your code and then engaging in the
following exercise of the "letter of the law":
>>


Ronald complains that Richard Stallman has watered down things in the GPL
and that it does not place sufficient restrictions on people (an uncommon
complaint, the opposite of the usual one). Like many people, Ronald
started this thread under the incorrect impression that the GPL forces
you to distribute code if you make modifications.

One can imagine such a license agreement, but I don't think it would be
workable in practice, how can you force someone to distribute something?

You could also have a license that said if you distributed it to anyone
you must distribute it to everyone, but that seems equally unworkable
in practice.

I am not sure that either of these would be an iomprovement from any point
of view, since they would remove too much personal freedom. if you pick
something up off the net with sources, the whole idea is that you can
fiddle freely with it for your own use. Saying that you had to distribute
it to the world if you modified it would place an intolerable burden and
for many people be equivalent to saying you cannot modify it.

Similarly, if you want to give your friend what you have done, the GPL
makes you give your friend the source on request, but certainly does NOT
mean you have to undertake the burden of general distribution.

Of course what Ronald COle wants is that we should be forced to give him
whatever we do as soon as we do it. Sorry, the GPL does not require this,
and it is not something that is going to happen. (By the way, if you are
a bit confused by his quoted text, it is from personal email that he is
reposting without the full context -- the full context was messages from
Richard Stallman confirming that Ronald's attempted interpretation of the
GPL was mistaken).

Anyway, here is how we do things at ACT, just so it is clear to people.


There are three kinds of versions of GNAT

First. The public versions. We only make versions public when they have been
in reasonably wide use for a while, so that any problems with installation,
or any other serious problems that have crept by our own procedures are
minimized. These public versions are always distributed *with* sources.
We package the objects and sources separately, since so many people want
to pick up only the objects, and the net does not have infinite bandwidth.
All mirror sites should always pick up both the objects and sources, so that
the sources are always available to anyone getting the objects.


Second. Interim releases for customers. These are fully tested internally
using the ACVC suite, our own regression suite, and also more recently the
DEC test suite (or rather selections from it that are relevant to 
implementations other than the VMS one -- the VMS implementations of course
use the whole DEC test suite). They are clearly not widely used when we
first release them, but they are fully supported. Many of our customers move
to these releases pretty rapidly, since they contain new features and new
bug fixes that are perceived as being worth the possible disruption of moving
to a new release. It is these interim releases which eventually later on
become public releases. Like the public releases, we release these with
sources, so that any customer obtaining the objects can also obtain the
sources. These are distributed via the GNAT FTP site, and are not available
from us to other than our customers. It is one of these releases that
Ronald has been demanding that we send him, but he will have to wait until
it is publicly released to get it from us (or become a customer).

Third. Wavefront releases. We make these available to customers on a need
basis (the typical situation is that a new feature or bug fix is urgent
enough that some specific customer is willing to switch to the latest
development version that has the needed feature of fix). These versions
have also been run through the ACVC tests and regression tests, but do
not have the level of internal use or confidence that we have in interim
releases. Furthermore, we expect them to be replaced by the next interim
release, and do not guarantee long term support for the wavefront releases.
These wavefront releases are distributed using option (b) in the GPL. That
is we provide objects only, with an offer to provide the sources on CD ROM
for a copying charge. 

I certainly understand that Ronald would like to get everything we do
free as soon as we do it, but it is not the way we work. In fact we do
not think it would be helpful for the Ada community if there were a new
public version of GNAT three times a week, it would end up causing huge
confusion and version chaos.

The current version situation with GNAT is as follows

Latest public release: 3.09 for most targets. A notable exception is DOS,
where the latest public release is 3.07. There currently is no working
3.09 for DOS. There was a problem with tasking (having to do with the
changes we made to separate out a non-tasking version of delay, which in
retrospect was a mistake). We are hoping to remedy this and create a
DOS 3.10, but we are not committing to this!

latest interim customer release: 3.10a. This is in use at many sites,
and has been verified as being in good shape by a number of critical
customers including SGI and OIS (who has the latest version of their
products working with 3.10a now). So it seems in pretty good shape.

We expect a 3.10b fairly soon which will have very extensive improvements
to the debugging capabilities of GDB. 

A public release of 3.10 will probably be based on 3.10b. We do not
have a definite schedule for this release yet.