Re: [H-source-users] [PATCH 6/6] Add Guix system with GNU/Hurd to the distro list

[Denis 'GNUtoo' Carikli]

On Mon, 25 Jul 2022 23:39:51 -0400
bill-auger <bill-auger@peers.community> wrote:

> On Tue, 26 Jul 2022 05:02:38 +0200 Denis wrote:
> > I also wonder what would happen if we have strange reports about
> > intel wifi working on these distributions. 
> > On Trisquel 10 for instance we can be pretty confident that it
> > doesn't ship nonfree firmwares
> 
> my point was that no such guessing or assumptions should be
> necessary - the list of known libre kernel modules is relatively
> small
> 
> the general purpose of h-node is to determine which hardware
> works with libre software - to answer that question, the distro
> name is quite irrelevant - it is at best, a heuristic - a
> trisquel user can easily install the ubuntu kernel; which would
> foil the heuristic completely
> 
> even without mentioning any distro names, all truly relevant
> information is captured deterministically, in the kernel version
> and the kernel module in use
> 
> even the distinctions between HURD, linux-libre, the debian
> linux, and the upstream linux are not actually relevant
In theory you're completely right that it should be possible to
actually characterize the system without having to ever record the
distribution name, and in ways that are more precise. 

What I'm trying to express here is that keeping the information about
the distribution and relying on users using known 100% free software
distributions (so without nonfree repositories) gives us some safety
margin that is completely gone with the system you propose.

With known distributions, even if there are issues (like if nonfree
firmwares are in Parabola at the time of the test for instance), there
is enough transparency to find about it retrospectively. In these
distributions, if a package containing nonfree software is removed, we
should have a trace of it in at least one git repository (in Parabola
that would be the blacklist and/or abslibre repositories).

For instance if we have a printer that work and that we forgot to
record the CPUS versions (because we were focused on the kernel) and
each CPUS driver version and to also record the precise patches that
these packages use, then we might have bogus data. And the issue here
is that we cannot retrospectively asks the users who added the entry
to do it again because some crucial data was missing.

In contrast, with if we still somehow also rely on the distribution
version, we can investigate and find out if there could have been some
issue at the time of submission (for instance by looking at Parabola
actual packages and/or their history in git).

So I think it would be a really good idea to start doing something like
you suggest (characterizing more precisely the systems being used) but
I don't feel comfortable at all with either stopping to record
the distribution or to remove the requirement for running a fully free
distribution.

It would also make sense to also whitelist some 100% free software
repositories like Jxself's repository of linux-libre kernel packages.
And the more precise characterization you propose could also enable to
take that into account.

Denis.

pgpxVTu14RH1_.pgp
Description: OpenPGP digital signature