Re: Non-LLM example where we do not in practice use original training data

Simon Josefsson <[email protected]> Tue, 06 May 2025 22:44:57 +0200

Newsgroups	gmane.linux.debian.devel.vote
Message-ID	<[email protected]>

Russ Allbery <[email protected]> writes:

> Andrey Rakhmatullin <[email protected]> writes:
>> Simon Josefsson wrote:
>
>>> I don't think Debian is perfectly consistent in applying that
>>> principle:  for example, the text of the Developer Certificate of
>>> Origin (DCO) is included in Debian packages (in 'main') and has a
>>> clearly non-free license, and IIRC sometimes not even in
>>> debian/copyright.
>
>> Same for the text of GPL.
>
> License texts have always been a special exception, and I kind of wish we
> would amend the DFSG to make that clear. Not because I think the status of
> license texts is somehow in question, but because having one undeclared
> exception makes people think we should have other undeclared exceptions. I
> would much prefer to take the time to enumerate all of our major
> exceptions.

Are you suggesting that the DCO is a license text that has to be part of
the licensing information for a piece of work, and mentioned in
debian/copyright?  Clarifying that would be good.  To me the DCO reads
as information intended for contributors to some project, to govern
policies around contributions, and I don't find it particulary relevant
for debian/copyright in the same way I don't find CLA texts relevant for
inclusion.  I think the DCO content is similar to some of the
Non-Variant GFDL sections rejected by general principle by Debian today,
explaining matters related to contributing to a project.

To me this handling feels like having a generic rule ("no non-free
texts") and applying the rule in different ways ("DCO text" vs
"Non-Variant GFDL text") depending on what outcome we want.  I find that
worse than having a (known incomplete) number of specific rules that are
applied consistently.

/Simon

signature.asc (application/pgp-signature, 1.2 KB)