On Tue, 19 Oct 1999, David Summers wrote:
Hi everybody,
> Maybe we could come up with a wrapper for CODA or something that was
> similar to the C++ interface with USERFS (or does CODA already have a C++
> interface)? If so then it might be a good replacement for USERFS????
I see the power of userfs in the kernel <-> user protocol. I know, that
it's not good for transfering it to another OS, but I don't see the point,
why should we do that. We need a simple, fast conversion between VFS calls
and user-fs calls. That's what in CODA is missing and what it makes more
general and available on other platforms.
I can see use of userfs in the mean time, where CODA, podfuk, and others
exist, when it will bring to us the possibility to write a fs with the
same power as the same fs written directly in kernel.
I don't like the C++ library, I would work on a pure C instead. But it
doesn't matter, if someone wants to use the C++ library, he can.
What I see, that we could do on the userfs now:
1. Test it in our environments, if it works well in kernel 2.2.x
and with glibc.
2. Create one site, where we'll put our work (CVS).
3. Assign a version (well behaved in our tests) as first starting
release.
4. Discuss the next development.
In my opinion, we should stabilize the userfs for 2.2.x and then we can
port it to 2.3.x. There's a lot of work more, though: to create a pure C
library, using pthreads, create additional example user-filesystems, etc.
We could perhaps also do some changes in the protocol, so it will be as
much as in VFS <-> kernel-fs. We could also create some benchmarks
for the user-filesystems. These all are just suggestions...
Enjoy,
Jim
.~. Jan Jirmásek
/V\ jim(at)penguin.cz, http://www.penguin.cz/~jim
// \\
/( )\ Linux, the choice of a GNU generation.
^^-^^
