The trouble is that you have to do it for _every_ write request in the
hole, unless you do fancy things with the tree and have a sparse map
--- in which case you're losing a lot of the intended efficiency gain
anyway. The direct map table seems to be a better idea all round for
writing into holes.
Actually, as long as the extent table fits in a single block, I don't
think it'll be that bad to be constantly reorganizing the extent table
(after all, it'll just be a single memcpy to make space in the table).
The moment that the extent table exceeds a single block, though, I'd
agree with Stephen that it's more trouble than it's really worth.
- Ted