Should Aunt Tillie Build Her Own Kernels?

DeadBugs writes: "Linux Weekly News is reporting on a new linux controversy. The inclusion of a Kernel Autoconfiguration program that would make it easy for almost anybody to build a custom Kernel on their computer. Eric Raymond supports this idea saying that this will bring Linux to a wider market. Those that oppose this idea mainly think that only those educated few should custom build their own Kernels. I for one hope this gets included if only to make standard installations and upgrades faster."

Comments

Controversy???

[CrazyBrett]

Why the heck is this a controversy? It seems to me that anything that makes good technology accessible to more people is a good thing.

I'd like to hear good arguments in the other camp, though.

Re:Controversy???

[Nailer]

Eric Raymond supports this idea saying that this will bring Linux to a wider market. Those that oppose this idea mainly think that only those educated few should custom build their own Kernels.

My own personal opinion is that:

* nobody should have to ever recompile their kernel (just update their distro in the worst case)

* everyone should be able to have the option of doing so easily if they want to.

Re:Controversy???

[Otter]

I'd guess it's a "controversy" because Eric Raymond a) proposed it in a typically condescending, inflammatory way and then b) ran around publicizing the thread as a great controversy in the kernel world.

Here is a different angle on the same issue, that makes for a better debate: Should the typical user be running a precompiled, distribution supplied kernel or a customized kernel that may offer performance advantages or may be wildly inappropriate and which creates immense tech support headaches?

Re:Controversy???

[SirSlud]

Rediculous. People obviously should not mess with things they dont understand. How many motor-morons have you heard of that have screwed up their car engines, just because there is access to the hood via an extremely easy to use, easy to recognize lever?! I mean, not EVERYONE pulls the lever and starts hammering on the crankshaft once their car breaks down. People know what not to touch, because they've been allowed access to it ... the responsibility in understanding the tools they are using is placed in their hands, as it should be!

When people suggest that technological means of prevention is superior to ingraining a respect for a particular technology, I get so upset! It's not my fault that MS has brainwashed you into thinking that the consumer should be unable to fuck something up! A sufficient warning and a learned respect for your belongings (in this case, the kernel) should be the top priorities. When people are told to not worry about what their doing, that they can't screw stuff up, thats when you end up with people who do screw stuff up once they find a chink in the armour of the technological solution!

Here's a sobering stat: more people fall off cliffs with fences than cliffs without fences. Why? Because when you leave people to their own devices, they have to think and respect the power of the tools they are using or the situations they are in. When you put the blinders on them, you're only making sure that shit will get fucked once you slip up and accidentally allow them access to the tools and technologies that you were so adament to lock everybody out of.

I understand that people still have access to custom kernels regardless of this auto config tool, but this is akin to providing an easily labeled handle to your hood, or an emergency exit, or whatever. Because it's so easy to get access to, people are forced to learn and know implicitly what the consequences of pulling it are! Compare this to the newb who finds the man pages on building the custom kernels, or the HOW-TO .. you won't get enough people having to face the idea of respecting the OS and computer in large enough numbers to make the fragility of the kernel and hardware a widely known thing. Sure, there might be some sacrificial lambs when you open up things like that, but hey, I'm all about the greater good rather than the few who still need to learn the lesson of accountability and respect!

Re:Controversy???

[blitzrage]

People obviously should not mess with things they dont understand.

My simply answer to that is: Then how did you learn?

Re:Controversy???

[rseuhs]

.. unless you've pooched/rm'd the factory-kernel. I mean, you /can/ screw stuff up.

And exactly how would that be differnt from today?

I don't need an easy to use autoconf tool to rm my kernel.

That was my point.

No, that's not a point. It's like saying 10 years ago: "OH my god all these ./configure scripts make it too easy for newbies to compile software!!! THE HORROR!" and then saying "Oh well, they could also delete the software, format the harddrive or pour coffe on the mobo. That was my point."

What utter nonsense.

Knowing what NOT to touch (in this case, the factory-kernel) is a part of life.

The auto-configurator is the only one touching anything in /boot here. Aunt Tillie won't wander around the filesystem randomly deleting files because of that.

Or compiled with the wrong flags such that your machine /boots/, but munges your hardware.

Wow, we are getting esoteric... Destroying hardware via software is only in very, very rare cases even possible. (for example old monitors, but new monitors won't break no matter what signal they get - but the kernel doesn't handle X anyway).

I run a dozen computers and I can't think of any device that could be harmed by the kernel. Not a single one.

But lets pretend such device would exist. The risk harming that hardware would be actually reduced because the autoconfigurator would choose the right setting and if unsure go with the safe settings.

You admit yourself that you havn't compiled a custom kernel, because of the lack of this tool, but more importantly, you're not even really aware of the risks!

I compiled a million kernels in my job (embedded systems engeneer). I said I didn't compile and optimize kernels for friends and for quite some time I don't to it for my desktops anymore neither because I'm a lazy bastard.

Re:Controversy???

[elefantstn]

People should not mess with what they don't understand.

I never would have learned anything if I thought like that.

Let the mob sort it out...

[davejenkins]

Am I reading correctly? Is this a debate over limiting vs. allowing certain behavior? What part of the Open Source philosophy got suspended while I was at lunch?

Let some distribution try this. It may take off, it may fail-- that's what it's all about...

Re:Let the mob sort it out...

[RedWizzard]

Am I reading correctly? Is this a debate over limiting vs. allowing certain behavior? What part of the Open Source philosophy got suspended while I was at lunch?

Unfortunately the story submitter felt the need to completely misrepresent the debate. The two camps arguments in a nutshell (IMO):

• Eric Raymond's view is that Aunt Tillie, who starts with a standard distribution should be able to click an icon and have a new kernel downloaded, configured, compiled, and installed. He's talking about standard (Linus) kernels here, not the distribution's kernel package.
• The other camp (which includes e.g. Alan Cox) doesn't see the need. Aunt Tillie would be better off sticking to the distro's kernel updates. If she wants to go beyond that then the resources are available for her to learn how to configure and compile a kernel using the existing tools.

So no-one is talking about limiting behaviour, that was just poor reporting. Personally I think that there probably aren't many Aunt Tillies who would find a need for the sort of tool Eric is advocating (although others my find it useful).

Exactly.

[NetJunkie]

Expect to see a lot of "This software only supported under the standard Red Hat v7.2 Kernel."

I don't blame the software companies one bit either.

Point could be moot.

[mystery_bowler]

Most Linux users are already familiar with the caveats and reprecussions of customizing your kernel. This kind of tool would just make it easier to get to.

There aren't all that many "casual" Linux users. That market is dominated by Microsoft. And if you've deployed Linux to a work environment, chances are you won't allow a tool like this to be used, because you'll probably want to lock down the configurations (making your life as a sys admin a lot easier).

Assuming Linux continues to proliferate to the consumer market, I still wouldn't be worried about people tinkering with their kernel too much. Most people, especially at the "average Joe" level, don't understand the inner workings of their OS. Heck, most of them fear their OS and assume that they'll break something if they tinker with the OS's inner settings. I wouldn't conclude that simply because the tool is there that most people would be interested in using it.

Aunt Tillie shouldn't *have* to...

[hpa]

I think this particular article is an egregorious misrepresentation of the arguments. Noone is arguing that an autoconfigurator isn't a nice thing to have, however:

• Eric has requested several in-kernel facilities solely to support his autoconfigurator. Most of these requests have been at the very best ridiculous.
• Aunt Tillie shouldn't have to build a new kernel. I can't emphasize this enough. We should be striving towards modular autoconfiguration at runtime, so you don't have to mess with your kernel because your hardware changed -- either at runtime or between boots.
• The autoconfigurator is bound to be an imperfect job, simply because a PC doesn't give you enough information to tell what exactly is in it, at least not in the presence of ISA cards. There is no magic you can do to avoid this problem.
• The kernel people are already drowning in bogus bug reports, to the point where it is very hard to avoid ignoring real bug reports. This, unfortunately, isn't likely to improve the situation.

I really can't emphasize strongly enough that I believe that if Aunt Tille has to build her own kernel, we have much bigger problems that Eric's autoconfigurator will solve.

Modularisation is the answer.

[Astral Traveller]

The need to compile custom kernels is a wart inflicted by Linux's monolithic nature. Instead of encouraging the painstaking and error-prone task of compiling custom kernels, we should be working on moving more and more kernel functionality into modules, which are loadable and configurable at run-time. It will always be easier and faster to setup a tool to install the proper modules with the correct parameters than it will to tweak a monolithic kernel config, then spend hours compiling the whole 20MB tarball worth of kernel source just to add support for a new feature.

While 2.4's module support is excellent, and modularisation is become more and more prolific throughout the Linux architecture, there are still several important features which need to be excised from the kernel core and made available as runtime modules. Trivial features such as APM support, SMP and Unix sockets shouldn't require a full recompile to activate. Why do we insist on prolonging the life of "make config" and its brethren when we could very well do without it altogether?

Different levels of effort for different people

[Stiletto]

Ponder:

Should Aunt Tillie Build Her Own Kernels?
Should Aunt Tillie Install Her Own OS?
Should Aunt Tillie Install Her Own Applications?
Should Aunt Tillie Run Her Own Applications?
Should Aunt Tillie Produce Her Own Documents?
Should Aunt Tillie Think Her Own Thoughts?

Disclaimer

[SirSlud]

*****WARNING. USING THIS TOOL CAN SCREW UP YOUR COMPUTER, BIG TIME. IF YOU WANT TO USE IT, DONT BE MAD AT ANYONE BUT YOURSELF IF YOU FUDGE THINGS UP*****

If MS can include regedit, you cant tell me that we can't inlcude autoconf .. I mean, seriously. The only thing you could end up with is some fucked kernels (who should get along just fine with the fucked registries) and some users who will learn and be cautious, and end up having a better understanding their computers.

Re:Disclaimer

[SirSlud]

whoa, did I say autoconf?!?!?!?! Sorry ... I meant, this auto kernel config tool. Yikes, I think it's time to go home ... or maybe for a refill on the coffee. :(

Isn't that the whole point?

[Cyberllama]

I mean honestly, what claim can linux hold over windows if not that the availabillity of the source code allots the user more freedom? This is, as far as I'm concerned, what linux is all about. I am totally unable to understand any argument against making one of the most important benefits of linux more accessible to a wider market.

Good old-fashioned response

[xant]

Eric has requested several in-kernel facilities solely to support his autoconfigurator. Most of these requests have been at the very best ridiculous.

I suppose I'd have no trouble believing this. I'd still like to know what the requests are and why they are ridiculous.

Aunt Tillie shouldn't have to

But.. she does. We don't have runtime autoconfiguration that works in every case. If an autoconfigurator is easy to build, and won't impact the people working on runtime configuration, then why stop them from doing it? My computer should read my mind (or at the least, the pointer should move to the thing my eyes are looking at) but I'm not going to tell people to stop working on improving mouse support.

The autoconfigurator is bound to be an imperfect job

True enough, but this is true for runtime autoconfig as well.

The kernel people are already drowning in bogus bug reports

Kernel bugs are reported via email on the mailing list. This is described in marginal detail in /usr/src/linux/REPORTING-BUGS. Furthermore, it begins with the following dubious paragraph:

What follows is a suggested procedure for reporting Linux bugs. You aren't obliged to use the bug reporting format, it is provided as a guide to the kind of information that can be useful to developers - no more.

What this document highlights more than anything is that kernel developers are drowning in bug reports because linux kernel bugs are reported in an informal format on the mailing list. Get a proper bug tracking system and it will be much easier to keep track of real bugs. This should be done regardless of whether or not we make "kernels for the masses". I hadn't heard about the bug report problem until you brought it up, and it's frankly amazing that it hasn't been addressed in this manner already.

Re:Good old-fashioned response

[rlowe69]

I hadn't heard about the bug report problem until you brought it up, and it's frankly amazing that it hasn't been addressed in this manner already.

Actually, the bug/patch reporting problem was mentioned in a very recent article about Linux VMs. Rik van Riel complained that Linus' (rather human-based) system was prone to missing patches, no doubt because the mailing list is filled with bogus bug reports, if indeed these are the same lists. Even if they aren't the same lists, Linus would probably have to monitor both anyway.

The point is that we have clear evidence a better system is needed for bug reporting and patch submission to give the main developers some way of organising and prioritising things. Clearly a simple mailing list does not suffice when the number of people submitting gets very large. Any takers?

Granny should build custom kernels.

[rice_burners_suck]

It would be great if anyone could build a custom kernel.

Imagine this... Let's say it's 5 years down the road, and the hot new computer is the 72 ghz Apple Pentium G7 with 64 gigs of on-chip ram. Hard drives have been totally eliminated because new, memory based permanent storage technology has been invented and proven over the past 2 years. An entire meg can be recorded in under 1 microsecond. The only remaining mechanical component of a computer is the standard Glass-RW drive (the 2 terabyte recordable successor to DVD), so the whole computer is now a small single board, and most of the electronic hardware is inside the main processor, an inch square in size. In fact, the plugs on this board take more room and cost more than the computation hardware.

Now imagine that a build world takes 4 minutes to complete. Here's how installation of FreeBSD 9.8-RELEASE takes place. (Yeah, I know this was a Linux thread.) You pop the Glassdisk in the drive, choose a few options, and all your software is configured, optimized, checked for security vulnerabilities, compiled and installed within 2 or 3 minutes.

In order for that to happen in 5 years, Granny needs the ability to custom configure her own kernel right now.

Maybe not good for Aunt Tillie, but good for me

[Jack William Bell]

Aunt Tillie doesn't need this. But, as a computer consultant and VAR, I need the ability to easily make these kinds of changes based on what my customers need.

Sure, I can do this myself the old-fashioned way. But this is the kind of thing I prefer to delegate to someone with a lower billing rate so I can focus on the things that really bring in the bucks. It is easier to train someone to use Eric's AutoConfigurator than it is to explain make files and such...

Jack William Bell, who likes the KISS method in most things.

This is a no brainer!!!

[3seas]

Programming is the act of creating automations of complexities that are made up of simpler things.

Does the programmer re-write open() every time they need to open a file?

There is not only nothing wrong with making it easier to build a custom kernel, but in fact there should be a growing interest in doing this sort of simplifying, given the GNU Hurd is about not only modularity but about servers/transltors and creating such, even custom as is needed.

This can be taken even further in that autocoding tools can be and should be built for the GNU users.

In a hundred years from now, how do you suppose programming will be done (given programming today is only about 50 years young)?

As things are being done today, it is not possible to do such a program of complexity as can be imagined of what would be a holodeck program (And we do have such virtual reality cudes today in university labs).

It won't be untill the general programming field realized the need to genuinely and honestly address and do the automation of the field of programming. Certainly everything else can be automated, including human balance and movement (segway).

It's fooling to continue the illusion that programming is not itself automatable. And to begin making it happen, where better than on higher level like autoconfiguration system that allow custom kernels to be done? (Or at least one place for it to begin)

A recent research paper on autocoding presents the current/recent mindset on autocoding. It's worth reading to see how young and admitedly immature the field is. Open system and Open Source Software such as the GNU efforts (Linux, the Hurd, etc..) with their open community has far better ability to do what needs to be done than any private effort which will be biased away from doing the things that need to be done.

Soooo, anything that automates computers and their use is inherently a good thing, for iot will allow us all to reach and achieve much more advanced systems and the benefits of.

My Crazy Idea

[graveyhead]

I want a distribution that has a similar GUI installer that RH and Mandrake have, but instead of invoking "rpm -i" for each package, it would build all the install packages from source drops. The "installer script" could be a large XML file that describes how to compile each package, what its dependencies are, and provides a mechanism for tweaking the packages configuration. Most of the packages out there can have their runtime configuration configured via their 'configure' script (wow that's a lot of "configures"), making it a fairly uniform approach. In addition, at the beginning of the install, it would be neat to see controls for your *exact* hardware configuation that get turned into CFLAGS like -march=i[my]86 and -O3, etc.

The only drawback I can see is that it would increase install times by a *lot*. However, in the end you would end up with a *highly* optimized distribution.

The idea came to me while building my own.

Aunt Tillie rocks

[eyeball]

I think if Aunt Tillie can create a swap partition during installation, pick a window manager, download compile + install the latest mozilla browser update (or maybe she prefers Opera), configure her firewall, and set up lpr for her printer, she can recompile her kernel. I just don't want to be around when she starts looking for "Freecell."

what's really wrong with this

[by2]

I have problems with Eric Raymond's scenarios. Forget about if it's a good idea to make it easy for anyone to build a custom kernel, my question is, why should you need to recompile the kernel just to install a device driver ? That's just stupid. Installable drivers, that's the way to go.