pathnames [Re: files, printers, etc. [Re: The feel of a Lisp

Chris Hanson
Thu, 8 May 1997 14:09:24 -0500

BeOS supported something like this in DR8.  Basically, the filesystem and
the database were built on top of the same substrate (I think the
filesystem was built on top of the DB, actually).  You could identify a
file by its pathname, but that was discouraged because (unlike Unix) users
will expect to be able to rename and rearrange files & folders and still
have things work.  Therefore you could identify filesystem objects by a
machine-unique persistent "record_ref."

With DR9 of BeOS, Be has a new filesystem that supports any number of
attributes to a file (i.e. property lists).  Unfortunately, they're doing
away with record_refs due to portability concerns.  Two steps forward...

At 1:11 PM -0500 5/8/97, Alaric B. Williams wrote:
>Right... Pathnames should be purely user-level things; we traverse
>a directory graph to find an object, but as soon as the system gets your
>choice, it asks for that object's "persistent reference handle" (an
>opaque object which might be something like "host IP:inode #" or a
>nice location-independent ID allowing for object migration, depending
>on the networking layer, and invisible to the application), for the
>simple reason that files can be renamed. The file hierachy is a
>user-level structure, full stop.