> > ...and if you're going to have not-quite-ELF, why ELF at all?
> > Given that it's a fairly flawed format (and not only because
> > of the "which OS?"  issue), being standard is the main thing
> > it has to offer.
>
> No, a slightly modified ELF would still leverage the tremendous
> amount of work that has gone into supporting the hard parts of
> the ELF format.  Indeed, this is precisely why ELF is a good
> choice: it's a standard format that does a good job at most of
> the things we need, and that can be modified to do exactly what
> we need with little effort.

For what it's worth, I agree completely.  The most promising
suggestion I have seen in this thread was the one by Erik M.
Theisen 23 Nov -- create a new "slightly tweaked" ELF (I think he
called it OLF for "Open Linkage Format") where it's possible to
do a transformation back and forth between the new format and
ELF.  Through the use of such conversion tools you can tag
imported binaries with the OS they originate from, and get the
kernel to reliably load it correctly (that is, if the kernel has
compat support for that OS, naturally).  In addition it would
appear that the modifications to the toolchain leverage the ELF
work, and that the modifications are small.  Go back to the
mailing list archive and read his suggestions once more (it's in
the tech-kern.0101 file if you need to go the NetBSD web site to
get it).

I openly admit that I'm not too familiar with the claimed weak
aspects of the ELF format, but reinventing the wheel (even though
the new wheel might be more "perfect") seems like a waste of
time, especially given that there are other and more important
problems to solve.

Regards,

- H=E5vard