Re: [GNU-linux-libre] Re: any Free BSD variant?

[Rubén Rodríguez Pérez]

> I think it would be even easier on a technical level to create and keep
> updated a GNU OS distribution based on OpenBSD than a GNU OS based on
> Debian or Ubuntu 

Well, as you will be replacing almost the complete BSD userland with GNU
programs, is not that easy.

> (a) you wouldn't need to fight against blobs in the upstream repository,

There are almost no blobs in Debian anyway.

> (b) the entire distribution is in a single source code control repository,

That is not true if you replace the userland programs, you would need to
add them to the ports system.

> (c) "make world" works, and

GOTO B

> (d) the major thing we object to, the non-free software in ports, could
> be solved just by making a copy of the distribution without ports and
> with name changes in some places to avoid pointing at upstream nonfree
> repositories.

If that's it, you can simply copy the OpenBSD repo and remove the
offending ports, and you will have a fully free system. (And just not to
piss off Theo: I mean a system that does not contain non-free pieces nor
recommend them)

> Socially it might be tricky to form a collaborative relationship

> With regard to (d) ports, the most obvious solution would be to create a
>  derivative of OpenBSD ports without the nonfree software.

You would also need to change it's name and website, but as I said, you
would have a free but pointless OS: you need to do more than that if you
want to maintain a fully free distro. If you don't, the upstream hackers
would think that you are just "teaching them ethics" by doing a tiny
tech job at the expense of their huge own, therefor they won't help you.
That would SURELY be the case with OpenBSD.

> it would be pretty trivial to create a
> tool to generate a free version of the repository, and they have
> well-documented tools to create an automated compile farm.

I think that the point of all this -despite of the technical examples-
is the question: should we build up a fully free distro from scratch?

Taking a well maintained distro and cleaning it up is easy, almost all
the bugs will be fixed upstream. But, no matter how well you automate a
farm, it will never give you a usable distro if you don't put LOTS of
work on it. Even Ubuntu is using Debian as upstream for most of it's
"extra" packages, it needs 124 people [https://launchpad.net/~motu] to
watch out for every single package to work well. And all the main
packages are maintained by cannonical itself.


> 
> * djbclark: dachary: http://nixos.org/about.html and also there is a
> useful academic paper I'll get the URL for in a sec...

This is a very interesting package manager, but it still needs you to:

* Brand and tune up every package using diffs
* Make sure that every package works well with the others -and fix them
with more diffs-
* Pick up every package from the upstream project -checking regularly
for new releases, and watching out for regressions-

That is what a distribution is all about, taking care of every package
by hand. It cannot be automated -you cannot even download the new
version of every package that way-. It can be somewhat automated if you
just include the GNU packages, put the result would be incomplete.

> 
> * djbclark: What makes me nervous about gnewsense, trisquel at all is
> that we are in most cases just trusting that the available binaries are
> producible from certain sources (since they are just copied from
> upstream; I don't know what blag's situation is).

We can check if the binaries are really made from it's sources -that can
be automated easily- although it would be easy to just ask them about
the build logs or to let us take a look at the build farm... Anyway,
that would be pointless: you would still need to check every source
package by hand.

The only way you wouldn't need to trust anyone is to avoid having a
upstream. We can do a GNU/hurd system without Xorg, but even if that
would be a fun technical exercise, it would be useless.

Anyway... Why shouldn't we trust -let's say- Ubuntu about putting fake
source tarballs that doesn't match the binary results?