Re: ELF dynamic loader name [was: sbrk(2) broken]

On Mon, 07 Jan 2008 09:49:20 +0200
Danny Braniss <danny_at_cs.huji.ac.il> wrote:

> I'm concerned in trying to solve a problem we are facing here, were 
> students/researchers
> write code, and soon will be hit by incompatible platforms.

Encourage them to write their code in something portable, like
Java, scheme, python, matlab/octave?  If they have to use
C/C++/Fortran/ etc, they could get used to distribution in
source?

The binary compatibility wheel-of-reincarnation is an interesting
one to watch.  When I was a student and post-grad at Uni, our
applications, when shared with colleagues, could very well have
needed to run on any of Vax, 68k, MIPS (32 or 64 bit), SPARC
(32-bit), ia32, x86 (16-bit: complicated pointers), with a few
PowerPC and Alpha systems coming in at the end.  So we used
matlab or handed around source code.  Before that it was
all-the-world's-a-vax (unless you were in an IBM shop.)  We've all
been in a peculiar bubble for a few years where "almost everyone"
has been using ia32, and it has been easy to think that that's all
there is (except for weirdos), and that therefore binary
distribution is OK.  I reckon that we're just coming out of that
mode, and transiting through something less even, probably until
amd64 completes it's clean-sweep and becomes the "one and only"
architecture again (to howls of protest from the ARM/embedded
crowd...)  That'll be a little way off, though...

[I'm doing a lot of my own new coding in PLT scheme at the
moment, and having a ball with it.  (lang/drscheme in ports)
Fast enough for what I'm doing, byte-code, static or JIT compiled,
and runs everywhere (including Windows and OSX).]

What would be *really* cool would be the ability to have a JVM or
LLVM back-end in the kernel, as a first-class peer of the ELF
loader.  Anyone know if anyone has tried such a thing on *BSD (or
even Linux, I guess)?

Cheers,

-- 
Andrew