Re: Announce: initrd-tftp 0.1

[Erik Andersen]

On Sat Jan 08, 2000 at 07:46:15PM +0000, Alan Cox wrote:
> > you are proposing that the embedded distro now need to have have a copy
> > ld, plus kernel object files? Hmm. ok, add in one copy of binutils,
> > (about 2.2 megs), plus an _uncompressed_ copy of the kernel object files
> > (about 1.2 meg), plus the initrd image. My embedded system just grew by
> > almost 3.5 megs.
> 
> Your distribution grew by 2.5 floppy disks. I assume you are using glibc,
> if so you clearly dont care about the odd 2Mbytes here and there. Your 
> embedded system hasnt changed size at all. Since your embedded customers
> almost certainly want to compile their own applications they are kind of
> likely to want binutils and gcc anyway. You have to ship the kernel
> source so whats the problem.

Yes I ship the kernel source, but come on, Nobody is ever going to
install the kernel source on the embedded system itself. That would be
silly (and it wouldn't fit).

Just adding binutils would double the size of my distribution, which
is not acceptable thank you very much. Admins can cross compile apps
and such for the embedded system on a separate host system (not on the
embedded system itself -- it is too underpowered for that anyway). Full
source is on the CD for doing just that. The reason it is called an
"embedded system" and not "workstation" is that it has a very specific
purpose, which doesn't involve being large or having binutils installed.

> > It seems to me that having this patch in the kernel, thereby de-coupling
> > the kernel from the initrd would be much better.
> 
> So you think its more important to ship 3.5Mb less code to a small number
> of embedded systems people than to slap another chunk of code into the kernel
> that needs maintaining, updating and downloading by zillions of mainstream
> users.
> 
> I don't.

We seem to be at an impass then. You have a good point. But if you carry
that point to its conclusion you are effectivly saying that things in
the kernel that are not commonly used by everyone should be deleted.
How many peple have 64 Gigs of ram installed? Delete that code. There
goes 90% of drivers/net/, 99% of drivers/cdrom/. Most folks use x86, and
the kernel would be easier to maintain and would download faster if we
delete the source for all those other obnoxious architectures...

I'm being ridiculous of course, but my point is that the kernel already
supports lots of things that are not intrusive upon the lives and
thoughts of folks that don't use them, but are extremely useful for the
few folks that do. Furthermore, it is going to be a while before Linux
takes over the desktop. It is taking over embedded systems _now_, just
like it is taking over the big iron machines. Adding 64Gig ram support
to x86 is code that is not used by many folks but is "another chunk of
code ... that needs maintaining, updating and downloading". I really
don't think obscure features for big iron systems are any different then
obscure features for embedded systems.

Do you have any technical problems with the initrd-tftp patch?

 -Erik

--
Erik B. Andersen   Web:    http://www.xmission.com/~andersen/ 
                   email:  [e-mail]
--This message was written using 73% post-consumer electrons--