home *** CD-ROM | disk | FTP | other *** search
- Submitted-by: bvs@bitblks.BitBlocks.COM
-
- Dan Bernstein writes:
- > bdiff, cmp, comm, diff, diff3, join, look, more, nohup, pr, sdiff, tee,
- > and touch can be written to work with file descriptors under v7. Under
- > BSD, the syscalls for working with descriptors are more complete, so you
- > can do chgrp, chmod, chown, etc. without explicit filenames.
-
- I don't see a fundamental conflict between making these programs
- work with file descriptors (actually, `handles') and wanting a
- naming scheme where every object of interest can be accessed by
- a name. Not every file operation makes sense on every object but
- that is okay. Just as write() to a directory does not make sense,
- chmod() of a network connection does not either. Naming is
- orthogonal to what operations make sense on an object or even,
- whether the object is long-lived or temporary.
-
- What many people (including me) are saying is that we want to
- access more kinds of objects than those provided by a traditional
- unix filesystem and we want a unified approach to accessing these
- objects.
-
- This does not mean implementing all of these new mechanisms in the
- unix kernel. Far from it. Any one should be able to attach a
- name-space object (a directory, but that has taken on too specific
- a meaning) into the existing naming hierarchy. Not all objects
- may be directly named in a given namespace (e.g. the namespace of
- internet services) but so what?
-
- I agree with Dan that the existing file protection system is not
- appropriate for network connections and that a more general scheme
- needs to be worked out; but it does not invalidate the usefulness
- of a uniform naming hierarchy.
-
- > The pure solution is to have every program keep a file descriptor around
- > for talking to the shell. Then requests for opening files can be
- > shuttled down that descriptor.
-
- Replace the `shell' with the root of the naming tree and we are in
- agreement! Given that any one can use any `shell' we can't depend
- on having a specific shell around. Besides, the job of a shell is
- to interpret user commands; it should not be in the business of
- implementing super-open -- UNIX philosophy(tm), you know :-)
-
- -- Bakul Shah
- bvs@BitBlocks.COM or ..!uunet!amdcad!light!bvs
-
-
- Volume-Number: Volume 26, Number 76
-
-