I would be interested in avoiding the i-num lookup in fuse because I have to do my own lookup to get extra information for my filesystem.  In fuse, you do a i-num -> path lookup and in my filesystem I then do path -> internal data lookup.

Do both options you list have the same complexity?  Would either of them need a hash table lookup, or is it just a matter of passing an inode struct vs an i-num as the key?


On Fri, 2004-02-13 at 00:16, Miklos Szeredi wrote:
> Yes it would.  It seems from what you're saying, though, that this
> may not be in line with the goals of fuse.

It is, only there was no demand for this up to now.  From the start I
thought about using either a path based or an inode-reference based
interface (this would be very similar to the kernel VFS interface) but
the path based won because of simplicity.

Now I'm hesitating between an i-num based and an inode-reference based
interface.  The i-num based is the most generic, and the filesystem
can implement any inode lookup algorithm it likes.  The
inode-reference based is easier to use, because the filesystem doesn't
need to implement the i-num -> inode mapping, but this makes the
interface more complex and less flexible.

What do you think?