On 2004-11-30 darkstar@city-net.com wrote:

> * I think I would prefer a single cannonical namespace from the kernel
> perspective and make it possible to use locators to lookup cannonical
> device names.  It should be possible (and IMO is desireable) to implement
> this lookup method generally to support "locator" identification in
> userland.  This mechanism must ensure that the target of a sccessful
> "locator" lookup will not become something else until all entities that
> referenced it by means of locators are no longer using that reference.
> This could possibly be done with procfs, separate from devfs.

  Er, I may be seriously misunderstanding the idea behind locators with
this idea and "complex userland device identification" (which I guess is
part of the idea of wedges and needs to be done at a different point in
the process?).  But realizing this also made me realize that there is a
big difference between listing and lookup and that I have been confusing
the two in my conception of how devfs would work.  For instance, being
able to list the names of the partitions (or just number of partition) on
a disk means you already read the partition table.  I don't like the idea
of reading the partition table just because I inserted the disk.  On the
other hand, it would be handy to be able to say "please mount partition
FOOBAR from wd1 on /mnt" and fail if partition FOOBAR doesn't exist on
wd1, just as mounting wd1b would fail if that partition doesn't exist.
This is a much different model of device existance than I had in mind and
might be a major agrument for a special devfs.  There could also
potentially be a separate "read the partiton map now" command that would
then set up the partiton device entries, which would get back to the "it
appears when it is actually there" mode, however perhaps there are other
cases where "it is actually there" isn't so easy to determine in advance
or where you would like to test for presence and use if possible all in
one shot.

Matthew Orgass
darkstar@city-net.com