[SGVLUG] FC repository searches

[Dustin Laurence]

On Mon, Sep 11, 2006 at 02:28:34PM -0700, Jeff Carlson wrote:

[[Interesting description of Fedora]]

I'm going to take a bit of issue with your description of Debian here,
since it's worth making the points in public.

> On the other hand, Debian plods along with one seemingly monolithic
> repository that really doesn't require upgrading the entire distribution
> between releases.

It really does, the problems and risks are the same.  The only
difference I perceive is that Debian users by and large trust their
packaging tools to do the upgrade while RPM users usually don't.  (RPM
users always complain when I say that, but that seems to be the reality
of it.)  It's possible that at this point the reason is no longer
package manager capability so much as having put in all the finicky
metadata in the pool or even long memories of past pain, but I have a
hard time just assuming this when there is a consistent demonstrated
difference in trust in the tools.

That said, upgrading Debian from release to release without following
the directions for *that release* appears to be a fine way to hose a
system and I always force apt to track a release *by name* (such as
"sarge") rather than by status ("stable") so that I can't accidentally
upgrade after a new one is released.

> ...The complete distro is something like a dozen CDs,
> although it is far better (IMO) to get a single CD to do a minimal
> install and grab all other software from the 'Net.

I agree completely, and have never done it any other way.  The complete
set, frankly, is for people in Lower Swabovia who don't have broadband
(I actually mean that--I think it's most useful in Third World countries
without good broadband, such as Bangaladesh or North Dakota).)

> I've used Debian a couple times (actually installed via Knoppix) and for
> some reason it broke on me each time.  Something about apt-get
> dist-upgrade was just not functioning for me.  Maybe it was because I
> started with Knoppix, not vanilla Debian, or maybe you're never supposed
> to do dist-upgrade and nobody told me that.

I'd never start with Knoppix, for fear of precisely that.  Also, if you
want to play sufficiently advanced tricks with apt the configuration is more
confusing than a Medieval proof of the existence of God, but you
shouldn't be doing that anyway. :-)  It doesn't count if you didn't
start from Debian itself.

That said, you're not generally supposed to use dist-upgrade on a daily
basis, at least for Debian stable.  The normal routine is apt-get update
&& apt-get upgrade.

> Second, what is my beef against DEB packages?  Nothing, really.  But the
> thought of shifting Fedora to them is seemingly insane to me.  To start
> with, Fedora is, as is well known, sponsored by RedHat, and since RPM
> was invented at RedHat, shifting to DEBs just isn't going to happen.

Surprisingly, I more or less agree with what you say but disagree with
the conclusion. :-)  The *most important* thing to know is probably that
both Red Hat, Inc. and Debian have a fairly acute case of NIH syndrome
and I don't think either's ego would allow it to switch.  Note that I
regard this as a damning statement about both.

> But furthermore, such a shift would mean at some point users would be
> forced to deal with upgrading between releases with old software in RPM
> and new in DEB, and I can only think about the world's greatest train
> wrecks happening right there.

Yes, it would suck mightily.

> Essentially, there is no advantage of RPM over DEB or DEB over RPM.
> Both do the same job, and both contain just about the same meta-info.

I'm not at all convinced that this is true, and the major piece of
evidence is the fact that (based on discussions with users) it's normal
to upgrade a Debian machine from release to release without
re-installing (our own David Lawyer is an extreme example of this),
while this appears to be the exception with RPM-distro users.  When
people ordinarily just upgrade through several releases without burning
CDs and reinstalling, I'll be more convinced (take that as a friendly
challenge for the Fedora project, if you like).

> and now yum (and I haven't looked at smart).  I think yum is a very good
> tool but it needs to get better (particularly faster) and that's not
> going to happen if we quit using it.  Well, we're not going to quit

apt-get isn't a speed-demon, at least not on doing updates, so for all I
know it's already at parity on performance.  Very likely I will end up
with FC and Debian installed on the same machine and will have some more
useful opinions about the comparison.  As I said, it's a good idea for
me to see how Red Hat Linux (which is what I really think of Fedora as,
since RHL is what I ran for so many years without even considering any
alternatives) has progressed.  I haven't used it since before apt-rpm,
much less yum.

> using it (at least not going back to apt-rpm) so that's not a big issue.

Here is the disagreement: it *is* a big issue.  It is a tremendously big
issue, because there is really no standard way to package third-party
software for Linux.  For Fedora, for Debian, for Gentoo, for Slackware,
yes--but for Linux, certainly not.  I named those four because they each
use a different package manager, but of course there are several more
out there either written for a specific distro (Arch I think is an
example) or those without a distro (I've seen several but none come to
mind just now).

We'd all be better off if, in the end, we all used the same manager, and
it would be worth the pain of transition.

The major two formats and managers are Debian's and Red Hat's, so the
most important thing to do would be to build a system which satisfies
both.  It can't be .deb--Debian is a law unto itself and isn't likely to
accomodate anyone else, and it surely can't be .rpm--Red Hat has an
equally deserved reputation for relentless focus on it's own needs as
well.  What would be needed is a development and ownership model
something like the current X system.

Of the other formats, we can ignore Slack--Slack's method is essentially
*perfect* if you don't plan to do automagic dependency resolution, as
Slack famously doesn't, and inadequate if you do, as everybody else on
the planet will and must do.  OTOH it would be a darn good idea to
ensure that any such system has at least the full capability of Gentoo's
portage or Arch's system, i.e. is suitable for a totally source-based
system.  Ultimately this would even benefit the regular distros, if they
were capable of seeing that (for example, Gentoo's system actually
encourages a lot of experimentation because of the ease of writing an
ebuild and even overlays--neither RPMs nor debs are good enough for
installing a few bleeding-edge packages on an otherwise stable system,
though this is very hard to see until you've used a source-based
system).

"One package manager to rule them all" isn't going to happen, we can
agree, but it's a disgrace, not a feature or even an irrelevancy.  We
need it, and it's worth the pain.

> ...It isn't openly recommended for
> production servers, although I have used it in such positions on
> multiple occasions.

"Openly" is the key here.  I think the ambiguous product placement (it's
stable, er, no it isn't, er yes it is) is a place where the tie to Red
Hat is a genuine disadvantage.  Red Hat doesn't really want to admit the
degree to which Fedora is suitable for typical server jobs because it
then begins to compete with their commercial product, but Fedora's users
and foundation really want it to be server-ready.  This tension isn't
resolvable as Fedora is currently constituted, I don't think.

> I'm not sure if the Debian project has any similar and, more
> importantly, published goals.

Zow.  You're kidding.  If there's one thing Debian does, it's document
such things at excessive length.  In fact, I think they more or less
pioneered this.  Most community distro's written goals and such seem to
be direct imitations of Debian's efforts in this regard.

> ...From my own experience and perception, I
> think they have goals which push stability ahead of staying up to date.

This is an unavoidable conclusion, but it's incorrect.  That is the goal
of Debian *stable*, which after all compares to RHEL and not Fedora.

Actually, that's a very good thing to say to yourself whenever reading
about Debian: Debian Stable is Debian's RHEL.

For desktop work, the secret that isn't written on the websites is that
everyone runs Debian Sid on their desktops.  This is closer to Fedora
but only somewhat--it's really more like Gentoo than anything else in
that it is kind of an eternal rolling upgrade that stays very
bleeding-edge but has no fixed release points.  The true analog of
Fedora is Ubuntu, and Ubuntu's success is very largely attributable to
the fact that Debian had simply and completely failed to serve that
particular niche.  The pent-up demand for a desktop that had fixed
releases and wasn't quite as bleeding-edge as Sid is what catapulted
Ubuntu from nowhere to mindshare-leader.

Debian's complaining about it rarely acknowledges that the niche was
theirs to lose, and they simply refused to try.  I would have preferred
Debian do it, but I can't complain that someone else stepped up to the
plate when they were AWOL.

>  I recall installing Debian on a machine a couple years ago and being
> faced with sendmail version 8.9.something, when the current version was
> 8.11.something.  I mean, wow, that was years behind.  Probably very
> stable, but I was just glad I was only setting it up as a null
> client.

That was almost certainly on Debian Woody.  Trying to get Sarge out the
door was the release that finally broke Debian's old way, and there is
no question that Woody should never have lasted that long.  That isn't
typical of Debian--look at the intervals between Debian releases and
you'll see.

The benefit is that Sarge was such a disaster (getting out, I mean, not
the final product) that it forced Debian to improve markedly in this
area.  Fixed 18-month release cycles, I believe, which seems about right
for the goals of Debian Sarge.

> On the other hand, Debian seems like a very popular choice as a starting
> ground to build another distro on top of.  From Knoppix to Linspire,
> there are lots of examples of this.

The size of the package pool and the flexibility of the tools make it a
very good base, yes.  I'm not sure it is better than Fedora in this
regard, but it's apparently as good.

Dustin

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://www.sgvlug.net/pipermail/sgvlug/attachments/20060911/128eec32/attachment.bin