Debian, Past Present & Future

solferino writes "Christoph Lameter, a major guru in the debian project, has put up a very well written talk that he gave earlier this week that addresses debian's past, present and future. He includes a good background history of the project, some interesting sets of figures and projections (30,000 packages by the end of 2004!), a good discussion of the pros/cons of source based distros and his ideas about a new package manager he is developing (uPM). In all a very good read, whether you are just now considering dipping a toe into the debian well-spring or have been drinking from the source for a long time already."

30,000 pkgs by 2004?

[bovril]

And 100,000 by 2006!!
I guess this means that sarge will be released around 2028...

Re:source distribs

[grumbel]

For those who need an incentive to try Debian, the keyword is stability; their QA process is what make the distrib lag behind in terms of latest versions, but the benefit is a rock solid platform.

Well, I heard this quite a few times now, but my own experience is quite a bit different. Debian stable is at first old, not stable. I have run in quite a few showstopper bugs that were already fixed in upstream and in unstable but which never made it into stable, since the QA process which makes it quite hard for new upstream to ever make it into stable.

I think the main problem here is that the freeze is globally to all packages instead of local to small package groups, in a lot of cases a package is still heavily under development when the freeze happens and then for month or years it will not get updated, even if the upstream becomes a lot more usable and stable.

Re:source distribs

[autechre]

Yes, this is sometimes annoying. One example is Mozilla; for a very long time, we were stuck with M18 in Potato, while new releases were certainly an improvement.

I understand that there are more variables in Debian with all of its supported architectures, and it wouldn't be as easy to simply release updates of later versions as distributions such as Red Hat do; you can't be sure of the impact it will have everywhere, and backporting security fixes is safer.

Perhaps a "mostly harmless" package repository could be created. No, "testing" doesn't count, because the packages in there will often be built against new libraries, and you probably don't want to go there. But this could contain binaries for packages such as Mozilla, which gets updated a lot (1.1 really is much better than 1.0) and would be unwieldy to build from source). These binaries would be built on a potato system. Those who wanted this sort of thing could simply add another line to their apt sources file, and accept the small risk.

It's possible for someone to do this on their own; Adrian Bunk maintained a repository of several updated packages so that 2.4 series kernels could be used on Potato. But I think it would be nice to have this as an official part of Debian. It doesn't sound so great to say, "Oh, yeah, you can do that; just get the packages from $THIS_GUY".

I didn't make him...for you!!

[autechre]

If you got that reference, I'm really sorry.

Anyway, please read this:

http://lists.debian.org/debian-devel/2002/debian -d evel-200204/msg01343.html

Maybe this will clear up a few things. Debian is supporting these architectures because no other Linux distribution does. As the message states, XFree doesn't even support as many architectures as Debian; the Debian project is how users of those architectures get XFree86 at all.

Maybe you feel that they are not important, but I think that the people using them would disagree. Obviously, there are enough people who use each platform to do the work of porting packages to it. What makes you think that they would turn around and do some other, "more important" work instead if support for their architecture was dropped?

[And isn't this why most hardware manufacturers don't release Linux drivers? Because "most people" use Windows?]

Debian exists as it is for many reasons, and there is nothing else like it. It is not going to change into your idea of the perfect distribution. However, there are several distributions which are addressing some of the "problems" _you_ (and others) have with Debian. Most of these amount to pretty graphical installers and a few other things, and are only for x86. Since that seems to be what you want, why don't you try one of them? IOW, don't complain that Mozilla doesn't have an integrated AIM client; use Netscape instead.

Attacking debians package structure

[Anonymous Coward]

But... there's something I'm worried about in debian... it's just too simple to apt-update.
If an user has installed an "unofficial" apt-source (are there any people out there which havn't?), a hack of a popular unofficial-deb FTP site can be disastrous.
This is not inherently debian's problem, but this distro makes it the easiest for people to update...
So debian should even stronger encourage the use of signed packages etc.

The problem is right there in the numbers

[iiioxx]

If you look at that nifty chart provided at the beginning of the paper, you can easily see that the release time for stable is stretching out further and further as more packages and architectures are added to the fold. At 9,000 packages and 11 architectures, it took 2 YEARS to ship a stable release. If they are right, and 2004 will bring 30,000 packages (and probably another arch or two), how long is the release cycle going to stretch out to? And will people tolerate that long between stables? Look at the grumbling and outcry about the delay in getting Woody out the door.

I find it very interesting that the article would point out several times how difficult it is to maintain all of those packages and the diffs as they are updated, then point out how using a source-based distribution makes that kind of thing much easier. And yet, the author seems to suggest that source-based distros are somehow not as feasible as binary-based distros. He even goes on to call source-based distros "immature". Perhaps in the Linux world, but how long has FreeBSD been around? It's okay to borrow ideas from other groups when those ideas seem to be working. I think that the Gentoo project has done a great job in taking the idea of a "ports" system, addressing the shortcomings, and putting a workable source distribution system on the Linux platform.

In my mind, if Debian is going to continue scaling to 5-digit package listings, the project might want to look into the possible benefits of switching to a source-based distribution system. Look at what Gentoo has done, address any shortcomings, and develop a better source distribution system. Doing it the current way with 30,000 packages to maintain, we might not see Debian 4.0 until 2010. And there are probably a lot of people who can't or won't wait that long.

Re:Why I run Debian (testimonial/rah-rah)

[Wylfing]

The desktop, for me at least, is another story. I guess that I found debian fell short.

Hm. Ok, I know I'm being that annoying Usenet guy who says "It works for me!" but here goes: Debian on the desktop works for me. On my recently-assembled work desktop I had a nice, fresh, system on which to try stuff out. I first tried Mandrake 9.0, because I assumed it would be a better desktop choice. After about a day I started over and loaded Debian Woody instead. I think it took me about 1 hour longer to get Woody up and running to a KDE 3.0 desktop, but that included such things as recompiling the kernel for my hardware! (Few things bring more geeky pleasure than making your own kernel .deb.)

Yes Debian is geekier. But there is a good reason everyone keeps using this "refreshing drink of water" metaphor for it. It's like you've never had a satisfying computer-using experience and then you "get" Debian and you go "Ahhhhh..."

Re:Debian's problems, RedHat sucks, but still use

[Anonymous Coward]

about the "horrible installer" - you only install debian once, then you use apt-get to upgrade.

so, why do you need a cute gui installer?

text is good, you are supposed to read, not drool over the cute looks of the installer.

after all, it's just an installer, not a girl.

Re:The problem is right there in the numbers

[Nerant]

Source based distributions won't solve the numbers problem, that namely you still need someone to package up the software you want in their equivalent of a .deb or .rpm. You still need people to test the building, potential library incompatabilities etc etc.

Re:The problem is right there in the numbers

[crywolf]

The biggest problem is that they have a central repository, where all packages are held. This is great for all important packages, but there's a lot of stuff that could be split off into auxillary repositories and maintained separately. Things like bash, ssh, and apache should be in the central repository. Games, IM clients, and media players are examples of extra stuff that can be updated at any time. A good division is whether or not you might use it on a server. I don't think this would overly complicate things; this kind of division would only be needed for Stable.

Re:Red Hat users - try apt-rpm

[Tolleman]

Tought, it's not the apt program that makes apt so good. its the package maintainers.

Data fraud

[epukinsk]

I like how he makes utterly false claims about his data based on his misleading graphs.

Look at his debian growth [telemetrybox.org] graph. He conveniently skips the year 2001, making it look like the growth in recent years is something other than linear. He even states "Note that the number of packages seems to be growing exponentially."

The truth is, he's crammed two years of growth into a one year slot on the graph, making it appear to be accellerating. In actuality, if you imagine that growth spread over two years (as it actually is) it looks damn linear.

I guess even volunteers without corporate agendas are subject to fradulent data analysis.

Erik

Plots (somewhat) misleading

[PSC]

The plots in chapter 2, The Past, are somewhat misleading in that they suggest a steeper growth rate lately that is actually true:

The time scale (x axis) is nonlinear!

The year 1996 is listed twice (thus making 1996 a particulary long year :-), while the year 2001 is missing (making the latest growth seem exponential).

Debian does grow rapidly, but not *that* rapidly.

Re:source distribs

[alfaiomega]

These binaries would be built on a potato system. Those who wanted this sort of thing could simply add another line to their apt sources file, and accept the small risk. (...) Adrian Bunk maintained a repository of several updated packages so that 2.4 series kernels could be used on Potato. But I think it would be nice to have this as an official part of Debian.

Since the current stable release [debian.org] is Debian 3.0 Woody, I'd suggest you apt-get dist-upgrade [debian.org].

Re:Debian users - try rpm

[lsd]

dpkg has it's own technical advantages too, such as suggests and recommends, and diversions. It also, as far as I know, integrates more closely with apt, allowing you to put packages on hold and handle systems which mix packages from a variety of sources (eg: base system from latest stable, latest version of package foo from unstable or testing).

Ignoring for the moment that IMAP4, ical, vcard and MAPI are all very different things, there's a fatal flaw in your analogy - you're equating things are are established, open standards to something that is comparitively closed. A more appropriate analogy would be to compare IMAP with POP3 - they're both open, established standards that solve the same problem in different ways, each with their own advantages and disadvantages. Trying to force everyone currently using POP3, or some other method they've come up with themselves, to use IMAP instead would cause chaos - IMAP would have to be expanded to handle functionality that currently only exists in POP3, and the changeover would be a nightmare for all concerned.

This is why the LSB states that here [linuxbase.org] that "The distribution itself may use a different packaging format for its own packages, and of course it may use any available mechanism for installing the LSB-conformant packages.". To me, alien sounds like a fine way of installing LSB-conformant RPM packages. The LSB have realised that trying to force all distributions to switch to the RPM format internally would be a fatal mistake - the toll on the Debian project would be enormous (volunteers really don't have time to repackage 8000+ packages to suit the desires of a committee), and it would destroy the usefulness of innovative new distributions like Gentoo.

Maybe I'm missing your point slightly, but I really don't see what switching Debian to the RPM format would achieve apart from causing an entire upheaval of Debian's entire package archive and development process, not to mention all of the Debian servers out there which can currently be so easily upgraded to the latest stable Debian version without so much as a reboot. You seem to say that it'd be for the greater good of Linux, but I personally think that keeping Debian around as a viable free alternative to the commercial distributions is a far more important thing to the future of Linux than some piddling squabbles over a file format.