Re: [fuse-devel] Implementing readdir() offsets

[Sven U. <sve...@gm...>]

Hello Nikolaus,

> > There is a little problem with that. fuse_add_direntry seems to
> > require an offset for every returned entry, not just the last
> > one. Since my offsets are filenames, I would have to keep an
> > offset-to-filename mapping for every single returned entry which
> > is basically just as bad as copying the entire directory contents
> > into a temporary table.
> 
> I just realized that in SQLite I already have that mapping with the
> rowid, I just need to make it autoincrement so that it becomes
> unique.  

What do you use as your primary key?  For me, this is the rowid
anyway...

> That should do nicely without any overhead. The only drawback is
> that I have to require off_t to be 64bit for that to work -- but
> hopefully that is the case on most systems nowaday.

But are offsets allowed to be non-contiguous?  What are they used for,
anyway?  Anything outside of readdir?  Would all those extra offsets
ever be used?

That said, if you had to keep a table of mappings, you could easily
have a timer which will clean up that table after so many seconds.  I
do this for a number of things which are moderately expensive to get
from SQL (just the query overhead, really), but constantly needed when
exporting via NFS...

Sven
-- 
    _  ___  ___  ___ 
 __| |/ __|| __|/ __|                             The dCache File System
/ _` | (__ | _| \__ \              An archive file-system for PB of data
\__,_|\___||_|  |___/                    http://www.desy.de/~utcke/Data/