Slashdot Mirror
A Better Installer for Debian?
F1re writes "Linux User mag in Germany has decided to include Debian on the mag and wants to make a more user friendly installer. They are looking for help from Debian developers. More info here Linux User"
← Back to Stories (view on slashdot.org)
First Windows removes its real-mode command line, now Debian loses its undecipherable installation syntax? What will all the esoteric-knowledge gurus do?
Jouster
Progeny Debian Linux was a GREAT distribution, when it existed. It had great hardware detection, a very simple installer and some other features. Some of the best parts of it are now available.
You can use Progeny Graphical Installer (PGI) to install a nice Woody release, or download the package and create your OWN customized installer with it. This thing is GREAT. Check it out - they are pushing hard towards the 1.0 release.
http://hackers.progeny.com/pgi/
It is very nice, and has a text-mode and X-based installer (you can even do the X install remotely on another machine). This thing is great, and I use it for all of my installs right now. Thanks, Branden Robinson and team for keeping this great part of Progeny Debian Linux alive...
I think that the basic point is being missed.
From the article
Right now what we'd be most interested in is some feedback by Debian developers and users out there
The golden rule in HCI is "Developers are not target users". Sad as it may seem for some people Linux Developers are no longer the same people as Linux Users. This means that, by and large, interface designers should IGNORE THE DEVELOPERS!
Users are the ones that matter here. As a first time Linux installer I don't really care about most of the things a developer cares about.
I haven't installed Debian, but let me compare my last Linux install (Mandrake 8.something) to WinXP...
All WinXP asked me was, essentially, "What is your Country and TimeZone".
Mandrake wanted to know the intimate details of my network card, how much swap space I wanted, what make of scroll-mouse I had, what sound card I had, what video cards I had (and don't get me started on XFree's Multimon support!). All this does is serve to scare and confuse a Linux Virgin. And if you want Linux on the desktop you can either make the world smarter, or make your products smarter.
Debian should not be soliciting people in the know - they know far more than the average first time user and are, consequently, useless for developing interfaces for newbies.
Sorry for the rant/misspellling/smell.
T
If a square is really a rhombus, why aren't all triangles purple?
I recently switched from Redhat to Debian linux. I used the network install (because i have a fast connection) and i found that the most obscure part of the install was finding which NIC model i had (because they went by manufacturer code instead of human-readable names) If a prepackaged installer simply had something that detected your NIC automatically, with some simple instructions to read along with each install stage (easy ones found at www.linuxnewbie.org), then it would be a much less painful install. As a seperate note, something must be added to automatically configure USB optical mice, because as it is they are not (a huge pain for a user with limited skills).
dselect is already a good tool for choosing packages to install and seeing what is out there to install. Its interface could be improved somewhat (always going past help screens becomes a pain, and collapsable trees should be in to reduce clutter(and if they are already, why aren't they obvious))
This should put user-friendliness in, while maintaining most of the customization available in the regular install (after all, you could always ignore the advice...)
Free as in *BUUURP!*
I'm sorry, but when did "I can click on it" equate user-friendlyness?
Debian's current default installer consists of a group of sub-menus with descriptive explainations of what task each menu item would perform.
If this is an arguement of asthetic or practicality, then it should be thrown out. The only way this arguement for a better Debian installer could possibly hold water is if we're talking about the detection of hardware, which is marginally frustrating (I had no problems the first time I installed debian).
~/ssh slashdot.org ssh: connect to host slashdot.org port 22: too many beers
The Debian installer used to be awful, and you really needed to know your way around dselect to get it installed properly (or you can just install nothing and apt-get it all once you've installed).
I recently installed unstable, using the testing installer, and I was surprised to see that it doesn't seem to use dselect anymore! There's a much more friendly (although still text-based) utility it uses to select packages. Honestly, the installer really is pretty easy now (on par with RedHat, anyway, only a bit less pretty). It could be better for non-computer-literate users, but only stuff like the partitioning utility.
This was on alpha, btw. Things may be different on the x86 side of things.
dselect sucks. It's the hardest thing about getting a working debian install, akin to a purity or intelligence test. This is exclusionary, and the only way to fix it is to streamline the way a base debian gets installed. And to me, that means dselect must go. It's too hard and takes too long to get right. I've always found it much easier and faster to completely ignore dselect and add the packages I need later using apt, which is far more friendly (and actually works).
In HCI terms, you *must* understand your users. If your user base is educated professionals who have done hundreds of debian installs and can compile their own kernel without assistance, then the current installer is probably okay, but it's not where Debian needs to go. It has the developer Linux user sown up; Debian needs to add to the collection other types of users.
So we pick another user set - the Linux newbie and/or Windows refugee. These people don't want to know about installers, and you must make the interface hard for them to screw up. Remember in HCI terms, allowing the user to screw up might be powerful, but it's wrong. I'm not talking about GUIs here (even though I like 'em), I'm talking HCI and interface. You can have a very decent text installer.
Moving along... You describe to the potential newbie users why you need an installer in very basic non-prejudiced terms, so they understand the problem space but without suggesting to them potential solutions. Grab their suggestions and recommendations and experiences and write them all down. This is your specification to a certain extent. Users have a keen insight on what they like and they don't like. Ignore their advice at your own peril.
You create a first cut at an installer, constantly second guessing the users: "will my mum be able to do this?" "Do I have to do this now?" "Is this a reasonable set of defaults that don't need to be adjusted?" You want the user to make as few decisions as possible, whilst postponing as many decisions as possible to allow experienced users to customize it if they wish.
Once the first cut of the installer is done, you must get a bunch of new users, and watch them use it without assistance. Learn from the mistakes or missteps they make, and learn if there's steps you can eliminate. And of course, eliminate any bugs the users find.
Repeat ad nauseam until it's hard to get a bodged unrecoverable install.
Developers are truly the worst people to ask to do this. They *know* the right answers, and will not even think that there might be other possibilities.
A good OS installer is like the old A/UX 3.0 installer - it literally was a one button install if you had a disk ready for it.
Other OS's with decent installers are NetBSD (with the possible exception of the very confusing disk partitioner) or WinXP (very few questions indeed).
Andrew van der Stock
Restoring from a backup is so much easier than trying to get your system to the point where u destroyed it.
/home (which is, of course, a seperate partition from the rest of the system ;P ).
/home, a trip through the Debian installer to install the base system, install my kernel, then:
Sure, unless it's so completely inconvenient to pull a proper backup that a reinstall is more manageable. For "modern" machines, this generally isn't an issue, as most have some easy removable mass-storage available (eg: CD-R/CD-RW, etc...) - but some older machines, particularly ones like my laptop, fall into a bit of a weird area.
Sure - I suppose I could transfer a couple gigabytes out over the network to do a proper backup -- but it's quicker and easier to just pull:
# dpkg --get-selections >selections.txt
...store that file, along with a custom kernel package (older laptops like mine need rather exotic hardware support that isn't in the default kernels) somewhere down
Then, if/when I do something stupid, it's a quick reformat of all but
# dpkg --set-selections <selections.txt
# apt-get -u dist-upgrade
...and I'm pretty much right back where I started. I can get the whole process done in about 1/2 hour - which is a bit shorter than backing up and restoring everyhing =)
Now, if I had a CD-R on the laptop, that would be a different story. Sure, I could get an external one - but that money (IMHO) would be better spent on a better laptop a few years down the road.
I wonder why they haven't posted anything on the Debian mailing lists...
The Debian people are by the way already working on a better installer. Woody will be the last release that uses the current one. This new installer will use aptitude instead of dselect for example.
Also, Linux User only plans for i386 support. Check this page. Debian supports many architectures, and the installer should work on all of them. Also, remember that Debian is being ported to non-Linux kernels. The Hurd is coming along nicely, and will probably released in Woody+1, and people are starting on a port to NetBSD. Again, the installer should support these kernels.
Finally someone who doesn't want to re-invent the wheel! For all the inherent benefits to Open Source and code reuse, the amount of code-duplication (and therefore time and effort wasted) in the Open Source world amazes (and disappoints) me.
And no I am not talking about Gnome vs. KDE. I am talking about things like having 10 different ICQ clients, all with different implimentations of the protocol. Sure, a different GUI and different features is worth making a new program for. But why not borrow the code for the network stack from someone else who already has that part tackled? Same with filters for MS Office. What is the big deal about KOffice, Abiword, and OpenOffice coming together and making some nice libs that translate .DOC into an XML format they can all interchange?
Simmilar things can be said about other softwares as well. Let's work together people! No need to re-invent the wheel!
I use Debian, I absolutely love it. I don't think the installation process is that bad, but I should not fail to mention that I didn't get it right the first time I tried it.
Of course, it wasn't the first Linux distro that I installed. I started with Redhat 6.2. I got frusterated with that because I couldn't figure out how to compile a kernel in Redhat, and all of the docs I found that were Redhat specific said don't recompile your kernel unless you ABSOLUTELY know what you're doing.
Then I tried Mandrake, and after using that for awhile I managed to compile my first kernel, but I still wasn't completely happy because I found it hard to configure.
I tried Slackware, and oh did I like that. The config files in /etc were super easy to modify, but the package management system left a little to be desired.
This whole time I was learning, and becoming better at using a Linux OS. I was seeing the different types of packages out there, rpm, tgz, and the one I hadn't used yet, deb. So I had to give debian a try. Like I said, I didn't get it right the first time, although the installation is easier than Slackware. Once I did though, and I discovered apt-get, I was hooked. I now have three machines running woody and one running potato and I'll never switch to anything else.
It's not the distro for beginners. It doesn't have to be. It's a good distro, perhaps the best, and it's not for beginners. There is nothing wrong with that.
My Karma was at 49, then they switched to words. All that work for nothing!
Very good point.
A quick correction, if I remember correctly (my Debian box is at home) I think it should be
# apt-get -u dselect-upgrade
^^^^^^^
rather than "dist-upgrade".
-- don't discount flying pigs until you have good air defense
I almost kick myself everytime I read one of these "re-inventing the wheel with open source" tidbits. I'm not sure what your experience is but it doesn't sound like you have a foot in the free software community.
:).
Because it is a community and the community isn't hiring people from Universities with CS degrees, rather within the community people are learning to code. They are getting experience. Some of us are rather new to coding so you will see "Hello World" re-implemented thousands of times. You'll see hundreds of the most routine shell scripts. And you'll see dozens of IRC clients all from a different code base. Why?
Its part of the fun of computing. Honestly, it is less fun (IMO) to start from someone else's program than from starting from scratch--especially when someone else's program already has all the features you want. The beginning stages of a software project are probably the most exciting.
Of course there are other things. Like it is more difficult to grok a large code base than a smaller one. And sometimes more experienced coders pull tricks that newbies don't quite grok yet--so decide to use more simpler and apparent methods.
Free Software isn't going away and I think you're going to see a lot more of this. Programmers going through different stages of experience and writing software that demonstrates different levels of skill.
One thing I've noticed is that software is becomming more and more complex. We may see what I call generational programming. Basically, instead of one programmer understanding a code base or even an entire community understanding a code base we may get to the point where several generations are needed to understand and contribute to a code base (or it may be a conveniant excuse for the TUNES project