First Look at Debian's Next Generation Installer

An anonymous reader writes "Over at LinMagAu There is an interesting look at the new beta version of the Next Gerneration Debian Installer. Putting aside the fuss around Ian Murdock, Progeny and Anaconda, this is how Debian is constructing the future of what is known to be it's Achilles heel. It's a well done beginning." While still not a graphical installer (and the article does a good job of explaining why that's not a priority) the installer now autodetects hardware, streamlining module selection, which was previously one of the more confusing parts of the install for newbies.

Comments

It's about time

[bsharitt]

I don't really care about a pretty install, I'm just glad they finally got hardware detection.

Re:It's about time

[awgriff279]

I can think of one good reason for Linux to have an easier installation process. I've wanted to switch from Windows for several years now. However, not having any Linux experience makes a proper installation very complicated. Consequently, I'm still using Windows until I have time to figure everything out. Unfortunartely, it's hard to learn about an OS when one doesn't have it to use. I love the idea of Linux, but until the learning curve drops on a free version, I'll probably keep using WIN98se.

Graphical?

[yoshi_mon]

While still not a graphical installer (and the article does a good job of explaining why that's not a priority)...

Who ever said we needed a graphical installer? There is absolutly nothing wrong with a good text installer. And for installing small footprint it's always best.

And besides, this is the logical progression. First you do the text installer, then you move on to a graphical installer if you so desire. Not the other way around.

Re:Graphical?

[Kjella]

Who ever said we needed a graphical installer? There is absolutly nothing wrong with a good text installer. And for installing small footprint it's always best.

Average consumers. There's nothing *wrong* with a command prompt either, but they don't like that either. Neither the cryptic C:\> prompt in DOS nor [root@mypc root]# in Linux/Unix. That is, if you want Linux to be interesting to average consumers, but I'd say having a market share that'd at least make companies take Linux users into consideration would benefit all.

And besides, this is the logical progression. First you do the text installer, then you move on to a graphical installer if you so desire. Not the other way around.

Mostly true. But considering that just about everyone except those installing a headless server would prefer to use the GUI if there was one available, it's not exactly a small "add-on" for a small special interest group. Particularly if you ever hope to convert Microsoft "point-n-click for almost everything" powerusers...

Kjella

Re:Graphical?

[TheRaven64]

I run FreeBSD on two boxes, neither of which has a screen or keyboard. When I installed them, I did so using a serial console. The only improvement to the FreeBSD installer I would like to see is the ability to run it over an ssh session (since serial ports are becomming less common). A graphical installer would add nothing for me.

In general, I feel graphical installers for operating systems are a bad idea, since you really should not be installing an OS unless you know what you're doing. The FreeBSD text installer has the advantage of being easy to use while still looking intimidating to the kind of person who shouldn't really be installing an OS, and encouraging them to either get help or read the documentation.

Re:Graphical?

[swb]

Text mode installers are fine, but Sysinstall (FreeBSD's) ain't that great.

The user interface isn't terribly consistant or easy to navigate, although it may be curses fault as much as FreeBSD's. It's also a major fuckaround if it fails someplace -- there's no recovering, despite the fact that the installer sticks around.

Personally I think it needs major rework to improve the UI. I'd like to see fewer seperate screens and more expandable hierarchical menus. They do seem to be kind of stuck on the two-floppy size limitation, which I'm not sure makes much sense anymore outside of die-hards that insist on doing floppy-started network installations.

I'd also like to see it capable of doing installations for network booted systems. This might seems contradictory, but think of an installer you run on the master system that lets you fill in the blanks and generate an image for bootable floppy or .iso that would then be net bootable, or on the net-booted system itself if the HDD was to be the boot source.

While it's been a usable install screen, it could use some UI and functionality help, all of which would require ditching the 2.88MB barrier.

A good graphical installer...

[ca1v1n]

A good installer for a vanilla desktop user would take advantage of all the hardware on their system. It should detect your sound card, and then play a sound that says "hey, we found your sound card!" and it should let you use your USB mouse, show all this stuff on your display in such a fashion that acknowledges the existence of the video card, etc.

Basically, it should be more like Knoppix.

Now, I wouldn't want to lock the user, who may not be a vanilla desktop user and may not even have a mouse or video card on the machine, into this setup, but it sure would be nice to have the option, wouldn't it?

Knoppix is wonderful and all, but it leaves behind some artifacts of the live CD setup that can make package upgrades (which users ought to be able to do graphically, and with little pain) very painful. If we could get stuff like this in the base Debian distribution, we'd be a lot closer to Debian being sufficiently user-friendly that we could hand a disc to grandma without fear.

*prepares for the "get redhat" flames*

Re:A good graphical installer...

[AKAImBatman]

> *prepares for the "get redhat" flames*

Try "RTFA". They state that the installer needs to work for every type of screen output from a GeForceFX to a serial cable. Being that the serial cable is the lowest common denominator that they had to support, they designed the installer as a simple text mode interface.

It's about time

[Fux the Penguin]

Personally, I've never had good luck with Debian. I know lots of people love it, and bully for them, but I have never been able to get a Debian system up and running to my satisfaction. I believed this was a personal failure until I succeeded two times with Gentoo, which is to Debian as Alaska is to Montana, in terms of frontier cred. Anyway, I agree that things that are dumb about the Debian installer could be improved, but I'm still a little worried that an installer my mama could run isn't right around the corner...

As everyone knows, Debian is maintained by an organization of volunteers. When people working on the distribution support users, it takes away from the time that they could be spending to improve the distribution. Therefore, it makes sense for them to not make Debian open for anybody to install. If someone can't make it through an installer that requires some attention and knowledge on the part of the user, then they should probably be using a commercial distribution that offers support for money or whatever. That's one of the things I like best about Gentoo's root shell installer. It immediately gets rid of people that are intimidated by that sort of thing, and prevents them from sucking up tons of attention on mailing lists or forums. The difficulty of the installer should be like those little signs in front of rides at amusement parks: "You must be this tall to ride."

The target audience of Debian doesn't need a graphical installer, so there's really no reason to put one in. If you want the easy graphical installer, perhaps you should ask yourself why you chose Debian in the first place. Besides, with distributions like Debian and Gentoo, using the installer is more likely than not a one time thing, because you can upgrade the version of your operating system without bothering with the installer. I'm all for installer improvements that save time for the core users of a distribution, but revising the installer to open the distribution to a new class of users should not be entered into lightly

Reinventing the wheel..

[spectrokid]

I don't want to be a troll, but I thought the whole idea about open source is you can copy from each other and not reinvent the wheel. If Mandrake has a really good hardware detection, then why are these dudes writing something from scratch?

Re:Logical progression?

[DavidNWelton]

Good theory, but the article explains why it doesn't work in practice: Debian has to be installable on 11 different architectures, and they have gone for a lowest common denominator approach (instead of, say, having a different, graphical install for x86 and maybe ppc).

I agree with other comments. I can do without the graphics, but it's nice to whip through hardware detection without opening another console and digging through /proc.

Not just for newbies...

[jdreed1024]

While still not a graphical installer (and the article does a good job of explaining why that's not a priority) the installer now autodetects hardware, streamlining module selection, which was previously one of the more confusing parts of the install for newbies.

While seleting modules by hand may not be confusing for non-newbies, it's still annoying. Sure, I know exactly which modules I need, and I could select them all by hand, but I shouldn't have to. One of the great things about RedHat's installer (I know, I know, RedHat is dead) is the kickstart option. I can put in a disk, kickstart a net install, take the disk out, and move on. And barring any unusual hardware, I'll come back to a fully installed system. This is great for bulk-installing machines.

I'm glad to see Debian has moved closer to this goal by doing module auto-detection.

Re:not yet graphical?

[Arker]

There are a number of good reasons not to do the install in Graphics mode. It's not necessary. It would introduce unecessary complexity in a crucial operation (installation) that doesn't require such complexity - that alone is good reason to veto the idea. Setting up the video properly is one of the most difficult things to do, and when you have a graphics mode installer a failure in setting that up properly on auto becomes a fatal error rather than a minor inconvenience. Plus a lot of Linux installations don't use graphics mode anyway - why go to the windows way of requiring a graphics card on machines that should be running headless and accessed via telnet and/or console cables only? Plenty of people use linux on machines that don't have a graphics system of any kind, and that's fine, in many cases it's a good thing. Why make an installer that won't work on a sizable portion of the machines that will run the software you're installing? How much sense does that make?

If it ain't broke don't fix it is an axiom for a reason - and making a graphic mode installer would be a great example of fixing something that isn't broke. The Debian installer could certainly be improved though, and from the article it seems they've made excellent choices in deciding what needs to be improved - and what isn't broken and shouldn't be fixed.

Re:Logical progression?

[the uNF cola]

Installation difficulties != graphical installer. Installation difficulties are due to a bad user interface.

OpenBSD has a great installer for the tecnically inclined, while dselect is just plain annoying. You have so many keys to remember just to select stuff, and the screen's view keeps changing.

Redhat's text mode interface is quite nice, 'cept it doesn't provide all the right questions or choices all the time. If i select something and its dependencies aren't met, it should ask me right away, "Do I want to add this or forget about my selection." I shouldn't have to think, I selected some packages before, and these are the missing dependencies over all of them, now I can go back and guess which ones i fooked up on.

The autodetect and what not is important too, getting rid of stupid questions like, "do you have a 3 button mouse." If there was a way it could figure it out, do it damnit. And this project at LEAST strives for the autodetection. Hopefully, it'd streamline the package selection process and what not.