Re: Projects in the Hurd

[Jeff Bailey]

On Wed, Jan 29, 2003 at 10:00:18AM -0600, Tom Hart wrote:

> W.r.t. programs with translators, of course, this (superior) style of 
> programming couldn't come about at present, since people would still 
> want APT to work on systems based on Linux and *BSD kernels. Something 
> I've been thinking is, I don't see how translators are inseperably tied 
> to multi-servers. Couldn't a Linux or *BSD kernel be hacked to support 
> translators?

It certainly could be, of course.  Traditions Unix design doesn't allow
the end user to easily change kernel tables.  After you overcome that,
your kernel would just have to refer to the table to see if there was an
active translator associated with the inode and pass the calls to it.

But that only really helps for filesystem translators.

The other magic in translators is that most traditional syscalls are
turned into RPCs and sent to various translators.  With a bit of black
magic, the user can replace those translators with his or her own for
doing system development work, or other specialised hacking and do so
without compromising the integrity of the system - Nothing is running
with any greater priviledges that the user naturally has.

Even this could probably be done with a bunch of work in the kernel and
glibc, though.  The Hurd was designed from the ground up with these
ideas, though.  Grafting them onto a traditional Unix design would be
non-trivial.

Tks,
Jeff Bailey