Re: [RFC] make localconfig

Re: [RFC] make localconfig

[Giacomo Catenazzi]

Hi.

I have some bash code (to probe hardware), an hardware/driver database
and a python script to partly generate the database direct from
kernel sources.
It should be in http://people.debian.org/~cate/files/kautoconfigure
and in http://people.debian.org/~cate/files/gnome-os

If someone find it usefull I can continue to work on that.

ciao
	cate


Thunder from the hill wrote:
> Hi,
> 
> On Sun, 25 Aug 2002, Zwane Mwaikambo wrote:
> 
>>For this kind of thing, code talks. Otherwise no one will take heed.
> 
> 
> I can't currently supply code, but the thing I'm doing is:
> 
>  - parse the dmesg for the vital stuff (the things that were loaded up to 
>    the moment may be used very often...)
>  - Find out what code it's belonging to
>  - Configure that code in
>  - The rest is CONFIG_MODULE
> 
> So what?
> 
> 			Thunder

Re: [RFC] make localconfig

[Kerenyi Gabor]

8/24/2002 11:12:46 PM, Thunder from the hill <thunder@lightweight.ods.org> wrote:

>	Generate a .config for the local computer, so that the kernel 
>	could be built right in that moment. Therefor the local computer 
>	is being examined, probed and configured and all the devices that
>	we find go into your .config.
>
>	The version is probably never 100% accurate, it might be a good 
>	idea to manually recheck the .config (e.g. via make menuconfig)
>
>	This is supposed to be a first step into a new direction where 
>	we no longer copy vendor kernels from the vendor CD to the system 
>	in the first position, but rather configure a new kernel for each 
>	system, hoping that somewhen the boxes will be fast enough to 
>	handle it in no time.

I disagree. It can't make things better or faster if you have to recheck the
.config file manually. When you are going through the checking part you
could just easily set the options too.
People compiling kernel know their machine and they also know what and
where must be set in the menuconfig. By the way there are a lot of machine
independent things that can't be discovered using a script, like network
options, preemption, filesystems etc. So users would have to set these
things at least and at this point I can't see how it could save time for me.
Nobody knows better than me what my computer has or what I want to see
compiled in the kernel. I think it would make things less reliable and instead
of saving time it would add extra check/time and work.

On the other hand it would be a great first step towards to have a full or almost
full automated kernel compilation for those who don't know enough or anything
about their computers or they don't have the necessary skills. Sooner or later
there will be a demand for it. Just think about avarage programmers. They now
just set up their M$ thing and start writing programs without the knowladge of
computers or deeper level of operating systems. They see only source codes of
THEIR programs and therefore they run into trouble if a hardware/driver problem
arises. (Of course I disagree with this kind of behavior) It could be useful for
this case, not for saving time.
And these people are not the ones without any computer skills. They have just
chosen another field in computer science.

Gabor

Re: [RFC] make localconfig

[Thunder from the hill]

Hi,

I think this version is supposed to illustrate much better what I've had 
in my mind.

localconfig(9)	  Generating local configuration	localconfig(9)

NAME
	localconfig - generate a .config for the local computer

SYNOPSIS
	make localconfig
	make dep ... etc... pp...

DESCRIPTION
	Generate a .config for the  local computer, so that the kernel
	could  be  built right  in  that  moment.  Therefor the  local
	computer is being examined,  probed and configured and all the
	devices that we find go into your .config.

	This is supposed to be a first step into a new direction where
	we no  longer copy vendor packages  from the vendor  CD to the
	system in the first place, but rather compile new packages for
	each  system, hoping  that  somewhen the  boxes  will be  fast
	enough  to handle  it  in  no time.  It's  basically a  binary
	distribution -> source distribution transition.

AVAILABILITY
	Linux 2.7+

SEE ALSO
	make(1), kbuild(9)

	scripts/localconfig.pl

AUTHORS
	(...)
	Thunder from the hill <thunder@ngforever.de>

BUGS
	The version is  probably never overly accurate, it  might be a
	good  idea to  manually  recheck the  .config  (e.g. via  make
	menuconfig) in the first versions.

Linux build system		Aug 25, 2002		localconfig(9)

			Thunder
-- 
--./../...-/. -.--/---/..-/.-./..././.-../..-. .---/..-/.../- .-
--/../-./..-/-/./--..-- ../.----./.-../.-.. --./../...-/. -.--/---/..-
.- -/---/--/---/.-./.-./---/.--/.-.-.-
--./.-/-.../.-./.././.-../.-.-.-

Re: [RFC] make localconfig

[Tomas Szepe]

> 	This is supposed to be a first step into a new direction where
> 	we no  longer copy vendor packages  from the vendor  CD to the
> 	system in the first place, but rather compile new packages for
> 	each  system, hoping  that  somewhen the  boxes  will be  fast
> 	enough  to handle  it  in  no time.  It's  basically a  binary
> 	distribution -> source distribution transition.

You might also like to take a look at Gentoo Linux, which I hear does
precisely this, except with all the packages the user chooses to install.
(Hopefully I've not malformed the name of the distro too much.)

T.

[RFC] make localconfig

[Thunder from the hill]

Hi,

No code available yet, but while doing the compras I've physically crashed 
(on my bike) into an idea:



localconfig(9)	    Generating local configuration	    localconfig(9)

NAME
	localconfig - generate a .config for the local computer

SYNOPSIS
	make localconfig
	make dep ... etc... pp...

DESCRIPTION
	Generate a .config for the local computer, so that the kernel 
	could be built right in that moment. Therefor the local computer 
	is being examined, probed and configured and all the devices that
	we find go into your .config.

	The version is probably never 100% accurate, it might be a good 
	idea to manually recheck the .config (e.g. via make menuconfig)

	This is supposed to be a first step into a new direction where 
	we no longer copy vendor kernels from the vendor CD to the system 
	in the first position, but rather configure a new kernel for each 
	system, hoping that somewhen the boxes will be fast enough to 
	handle it in no time.

AVAILABILITY
	Linux 2.7+

SEE ALSO
	make(1), kbuild(9)

	scripts/localconfig.pl

AUTHORS
	(...)
	Thunder from the hill <thunder@ngforever.de>

BUGS
	Well, how could I tell yet?!

Linux build system	    $EPOCH+$x			    localconfig(9)





Now I want your comments...

			Thunder
-- 
--./../...-/. -.--/---/..-/.-./..././.-../..-. .---/..-/.../- .-
--/../-./..-/-/./--..-- ../.----./.-../.-.. --./../...-/. -.--/---/..-
.- -/---/--/---/.-./.-./---/.--/.-.-.-
--./.-/-.../.-./.././.-../.-.-.-

Re: [RFC] make localconfig

[Toon van der Pas]

On Sat, Aug 24, 2002 at 08:12:46AM -0600, Thunder from the hill wrote:
> Hi,
> 
> No code available yet, but while doing the compras I've physically crashed 
> (on my bike) into an idea:
> 
> 
> 
> localconfig(9)	    Generating local configuration	    localconfig(9)
> 
> NAME
> 	localconfig - generate a .config for the local computer

Yow!  The return of Aunt Tillie compiling a kernel!   :-)

-- 
 /"\                             |
 \ /     ASCII RIBBON CAMPAIGN   |  "Who is this General Failure, and
  X        AGAINST HTML MAIL     |   what is he doing on my harddisk?"
 / \

Re: [RFC] make localconfig

[Thunder from the hill]

Hi,

On Sat, 24 Aug 2002, Toon van der Pas wrote:
> > localconfig(9)	    Generating local configuration	    localconfig(9)
> 
> Yow!  The return of Aunt Tillie compiling a kernel!   :-)

Oh, please. This was not my intention. I was just suggesting something 
like this could be done in the real world, also don't I have these ideas. 
I just think it might save you some time when you've accidently rm'd your 
.config, and all you want to have is a .config for the local box. Also 
wouldn't I load all the modules one by one, possibly.

If you think it's a _deadly_bad_idea_to_do_ please tell me. It is, after 
all, just an RFC, means I request you to comment on this. I could even ask 
for kernel protection for the mice.

			Thunder
-- 
--./../...-/. -.--/---/..-/.-./..././.-../..-. .---/..-/.../- .-
--/../-./..-/-/./--..-- ../.----./.-../.-.. --./../...-/. -.--/---/..-
.- -/---/--/---/.-./.-./---/.--/.-.-.-
--./.-/-.../.-./.././.-../.-.-.-

Re: [RFC] make localconfig

[John Levon]

On Sat, Aug 24, 2002 at 03:25:58PM -0600, Thunder from the hill wrote:

> If you think it's a _deadly_bad_idea_to_do_ please tell me. It is, after 
> all, just an RFC, means I request you to comment on this. I could even ask 
> for kernel protection for the mice.

It's an old idea that has been implemented at least once... I suggest
you read the archives

john

-- 
"When your name is Winner, that's it. You don't need a nickname."
	- Loser Lane

Re: [RFC] make localconfig

[Zwane Mwaikambo]

On Sat, 24 Aug 2002, Thunder from the hill wrote:

> Oh, please. This was not my intention. I was just suggesting something 
> like this could be done in the real world, also don't I have these ideas. 
> I just think it might save you some time when you've accidently rm'd your 
> .config, and all you want to have is a .config for the local box. Also 
> wouldn't I load all the modules one by one, possibly.
> 
> If you think it's a _deadly_bad_idea_to_do_ please tell me. It is, after 
> all, just an RFC, means I request you to comment on this. I could even ask 
> for kernel protection for the mice.

For this kind of thing, code talks. Otherwise no one will take heed.

	Zwane
-- 
function.linuxpower.ca

Re: [RFC] make localconfig

[Volker Kuhlmann]

> > I just think it might save you some time when you've accidently rm'd your 
> > .config, and all you want to have is a .config for the local box. Also 
> > wouldn't I load all the modules one by one, possibly.

> For this kind of thing, code talks. Otherwise no one will take heed.

Even with code noone seems to have taken heed. The code has been
available for ages. I think it's very useful. gunzip </proc/config.gz
gives you the config of the running kernel, make cloneconfig sets up
the source tree. Free and convenient with every SuSE kernel.

Volker

-- 
Volker Kuhlmann			is possibly list0570 with the domain in header
http://volker.orcon.net.nz/		Please do not CC list postings to me.

Re: [RFC] make localconfig

[Thunder from the hill]

Hi,

On Sun, 25 Aug 2002, Volker Kuhlmann wrote:
> Even with code noone seems to have taken heed. The code has been
> available for ages. I think it's very useful.

I think you're talking about different code.

> gunzip </proc/config.gz gives you the config of the running kernel,

What if the current kernel is some allmodconfig, and you don't want that, 
for some reason? And by the way, that's not related to the transition I 
mentioned. If you want that, just go and do make allmodconfig.

> make cloneconfig

Why not make oldconfig?

			Thunder
-- 
--./../...-/. -.--/---/..-/.-./..././.-../..-. .---/..-/.../- .-
--/../-./..-/-/./--..-- ../.----./.-../.-.. --./../...-/. -.--/---/..-
.- -/---/--/---/.-./.-./---/.--/.-.-.-
--./.-/-.../.-./.././.-../.-.-.-

Re: [RFC] make localconfig

[Mike Fedyk]

On Sun, Aug 25, 2002 at 05:17:24AM -0600, Thunder from the hill wrote:
> Hi,
> 
> On Sun, 25 Aug 2002, Volker Kuhlmann wrote:
> > Even with code noone seems to have taken heed. The code has been
> > available for ages. I think it's very useful.
> 
> I think you're talking about different code.
> 
> > gunzip </proc/config.gz gives you the config of the running kernel,
> 
> What if the current kernel is some allmodconfig, and you don't want that, 
> for some reason? And by the way, that's not related to the transition I 
> mentioned. If you want that, just go and do make allmodconfig.
> 
> > make cloneconfig
> 
> Why not make oldconfig?

Because then you'd just be comparing the current to the default config, not
much gained there...

Re: [RFC] make localconfig

[Thunder from the hill]

Hi,

On Sun, 25 Aug 2002, Zwane Mwaikambo wrote:
> For this kind of thing, code talks. Otherwise no one will take heed.

I can't currently supply code, but the thing I'm doing is:

 - parse the dmesg for the vital stuff (the things that were loaded up to 
   the moment may be used very often...)
 - Find out what code it's belonging to
 - Configure that code in
 - The rest is CONFIG_MODULE

So what?

			Thunder
-- 
--./../...-/. -.--/---/..-/.-./..././.-../..-. .---/..-/.../- .-
--/../-./..-/-/./--..-- ../.----./.-../.-.. --./../...-/. -.--/---/..-
.- -/---/--/---/.-./.-./---/.--/.-.-.-
--./.-/-.../.-./.././.-../.-.-.-

Re: [RFC] make localconfig

[Toon van der Pas]

On Sat, Aug 24, 2002 at 03:25:58PM -0600, Thunder from the hill wrote:
> Hi,
> 
> On Sat, 24 Aug 2002, Toon van der Pas wrote:
> > > localconfig(9)	    Generating local configuration	    localconfig(9)
> > 
> > Yow!  The return of Aunt Tillie compiling a kernel!   :-)
> 
> Oh, please. This was not my intention.

My apologies.  I really couldn't resist..

> If you think it's a _deadly_bad_idea_to_do_ please tell me. It is,
> after all, just an RFC, means I request you to comment on this.
> I could even ask for kernel protection for the mice.

A serious question about your proposal then:
The .config file doesn't excusively contain information regarding
what hardware to support.  It also carries a lot of information about
the functionality and the optimzations the user wishes to incorporate
in his kernel.  How would localconfig deal with that?

Regards,
Toon.

BTW: Considering that you are not out of your mind (most of the
time), I admire your courage in bringing up this subject.  ;-)
-- 
 /"\                             |
 \ /     ASCII RIBBON CAMPAIGN   |  "Who is this General Failure, and
  X        AGAINST HTML MAIL     |   what is he doing on my harddisk?"
 / \

Re: [RFC] make localconfig

[Thunder from the hill]

Hi,

On Sun, 25 Aug 2002, Toon van der Pas wrote:
> > If you think it's a _deadly_bad_idea_to_do_ please tell me. It is,
> > after all, just an RFC, means I request you to comment on this.
> > I could even ask for kernel protection for the mice.
> 
> A serious question about your proposal then:
> The .config file doesn't excusively contain information regarding
> what hardware to support.  It also carries a lot of information about
> the functionality and the optimzations the user wishes to incorporate
> in his kernel.  How would localconfig deal with that?

Well, the idea is to
 - scan the kernel messages for things that got loaded
 - figure the hardware configuration
 - figure out guessable optimizations
 - use the default values for the rest

I don't want to decide between "y" and "n". That would go wrong. The only 
thing is that there *should* be a decision on whether the support is vital 
for boot or not, if not we compile as module.

This statement is not valid for debug options, though...

> BTW: Considering that you are not out of your mind (most of the
> time), I admire your courage in bringing up this subject.  ;-)

I wonder whether that's a good thing...

			Thunder
-- 
--./../...-/. -.--/---/..-/.-./..././.-../..-. .---/..-/.../- .-
--/../-./..-/-/./--..-- ../.----./.-../.-.. --./../...-/. -.--/---/..-
.- -/---/--/---/.-./.-./---/.--/.-.-.-
--./.-/-.../.-./.././.-../.-.-.-

Re: [RFC] make localconfig

[Thunder from the hill]

Hi,

On Sat, 24 Aug 2002, Thunder from the hill wrote:
> 	This is supposed to be a first step into a new direction where 
> 	we no longer copy vendor kernels from the vendor CD to the system 
> 	in the first position, but rather configure a new kernel for each 
> 	system, hoping that somewhen the boxes will be fast enough to 
> 	handle it in no time.

Let me be clear on this. I'm not for an "Aunt T*ll*e compiling a kernel", 
but I'm rather thinking about following the comments on non-binary 
distribution of programs. (Remember the talks when somebody suggested that 
programs might be distributed more flexible if they're configured and 
compiled on install?)

			Thunder
-- 
--./../...-/. -.--/---/..-/.-./..././.-../..-. .---/..-/.../- .-
--/../-./..-/-/./--..-- ../.----./.-../.-.. --./../...-/. -.--/---/..-
.- -/---/--/---/.-./.-./---/.--/.-.-.-
--./.-/-.../.-./.././.-../.-.-.-

Re: [RFC] make localconfig

[Andrew Rodland]

On Sat, 24 Aug 2002 08:12:46 -0600 (MDT)
Thunder from the hill <thunder@lightweight.ods.org> wrote:

> 	Generate a .config for the local computer, so that the kernel 
> 	could be built right in that moment. Therefor the local computer
> 	is being examined, probed and configured and all the devices
> 	that we find go into your .config.

It turns out that the autoconfigure script included in CML2 is actually
an adaptation of kautoconfigure (Giacomo Catenazzi <cate@debian.org>,
http://sf.net/projects/kautoconfigure), just tweaked to use CML2 and
python... a slightly older version that uses sh (well, bash) is still
available. The ruleset is something like 8 months old by now, but the
features provided are really pretty nifty. I used it once and it worked
very nicely. I don't know if it was the, erm, downfall of CML2 that
killed this project, but I wouldn't mind seeing it come back.