Re: [Gnu-arch-users] extended attributes

[Tom Lord]

    > From: Andrew Suffield <address@hidden>

    >> Are you sure?  Extended attributes have no particular
    >> semantics.  There's no rule to say when they should be copied
    >> and when not, when they should modified and when not.  They are
    >> an example of the "property lists solve _everything_!" class of
    >> design errors.

    > It is fairly obvious that you can build that on top of posix extended
    > attributes. 

You can, sure -- but with no greater difficulty or loss of generality
than you could build isomorphic functionality on top of ordinary
directories (sans extended attributes).

In fact, building the same thing on directories would lead to greater
generality because, for example, tools like `grep' and `find' would,
without modification, implicitly recognize your new "attributes".

    > Property lists *do* solve "everything", for the same reasons
    > that hard drives solve "everything", even though the data that
    > is stored is not specified in their documentation.

Abstractly, all that attributes give you in the filesystem is a way to
bundle a bunch of names under a single common name such that a single
inode can be used to address them all.  Directories already do that.


    > Extended attributes are a generic interface for attaching
    > metadata to files. You're *supposed* to implement the rest in
    > userspace, and that's what will probably happen. There's no
    > reason why the filesystem should implement any of the things you
    > describe.

Doing this stuff in userspace is just fine with me.  No filesystem
modifications are needed for that.   Filesystem modifications will
only make generic tools less useful.

-t