a long rebuttal to the Linux-is-the-engine fallacy

[Marko Vojinovic]

On Monday 28 July 2008 01:21, Alexandre Oliva wrote:
> On Jul 27, 2008, Marko Vojinovic <vvmarko at panet.co.yu> wrote:
> > Memtest runs under the bios operating system.
>
> Nope.  It does rely on probing and some BIOS configuration tables to
> find out what it's running on, but that's about it.  No operating
> system involved.

Oh, really? So you are saying that memtest talks directly to the graphics 
hardware, printing its output pixel-by-pixel on the screen? It doesn't rely 
on bios routines for basically everything except maybe memory-access? My 
knowledge of assembly is a little rusty, but iirc the INT directive is used 
for most of write-to-the-screen tasks, and it actually asks bios to do the 
actual pixel-drawing. And the fonts and 80x25 output and everything is 
defined in the bios, not memtest itself.

The difference between memtest and the kernel is that memtest *relies* on bios 
to run, the kernel generally does not (except for the boot process). Do you 
remember that MS DOS was never actually called "true" operating system, 
precisely because it was actually just a "real mode" app running on top of 
and relying on bios?

> > No program runs without some sort of kernel, except the kernel
> > itself.
>
> That's just not true.  Have a look at all of the *-elf or *-coff
> configurations available in the GNU toolchain.  Those are aimed at
> creating applications for embedded targets without any operating
> system whatsoever.  They run on bare hardware, *sometimes* with a
> loader that enables arbitrary files to be loaded over say a serial
> line or over NFS, *sometimes* configured to just start the program
> stored on (P)ROM.

I wouldn't know about those apps. But that is aside the point, those kind of 
apps are not usually a part of the distro like Fedora. And I don't mean the 
toolchain, but those generated apps that are supposed to run on bare 
hardware. As far as I know, the kernel (xen and similar included) is the only 
binary in the distro that talks *directly* to hardware, circumventing even 
the bios. If there is another, please enlighten me. :-)

> > I am not very familiar with the working of xen, but it looks just like an
> > additional step to booting the kernel.
>
> Most hypervisors are microkernels of their own (the exceptions being
> full-fledged kernels).  Xen is slightly different from most in that,
> even though it is a microkernel, it depends on and redirects to
> another kernel that runs under it a lot of interaction with the
> hardware.  All the kernels running under xen are under constraints
> determined by xen, so they're not as special as they were before, but
> this one kernel (called domain 0) is not as constrained, so it remains
> a bit more special than the others.  But it definitely isn't just a
> loader.

Aha, so xen is something in between a fully-fledged kernel and a loader. This 
makes sense for a virtualized environment.

> What other artifical exceptions and work-arounds are you going to have
> to invent to make it seem like Linux deserves to be more relevant than
> GNU in a distro? 

Well, as many as I find convenient for the job. :-) See below for explanation.

> Isn't that an indication of something about both 
> your intent and about the truth of what you're trying to dispute?

My intent is to demonstrate that there is an alternative to your reasoning, 
because I intutively feel that there is, at least one. So I am just trying to 
construct an argument that can be considered valid enough, while producing a 
different conclusion than your own. My agenda with this here is to 
demonstrate to people on the list (if there are any left reading this 
thread :-) ) that giving names to stuff is largely a matter of taste, and 
that there is no single valid criterion for doing that. I am deeply convinced 
that this is true quite generally, while you seem to advocate otherwise. 
Please take a look at my other (ironic) post, regarding 
"classical/information", to understand what I think of this kind of agenda.

Btw, in mathematical logic, there is a notion of "undebateable propositions", 
and it describes statements that are partially-defined and partially only 
intuitively named, so that there cannot be a resolving answer to the question 
of their truth. That is why they are called "undebateable". One of the famous 
such statements is the Church thesis ("every intutitvely computable function 
is recursive"). It "feels" correct, but one cannot debate its correctness 
successfully.

For me, this whole discussion about GNU and Linux seems undebateable as well, 
and I am constructing alternative arguments in order to demonstrate that 
there cannot be a definitive answer to the question of the name of the os. 
People have to name it based on intuition.

And if your agenda is to influence the intuition of general public regarding 
the name, I claim that it cannot be done by arguments, but by education to 
build a new kind of intuition. And that is not something that can be done in 
a thread of a mailing list --- it needs a completely new approach.

Best, :-)
Marko

P.S. Folks, while I find this thread quite enjoyable, I need to point out that 
I am quite short of free time, and if I become silent from some point on, it 
is mainly because I was forced to stop reading the thread. So if that 
happens, I wish everyone to know that I am really pleased and honored to take 
part in the discussion, as I hope other participants feel as well. :-)