The following need doing and will not happen fast, unless someone does
them for me. word2x has no need for these features.

- caching: cache "sectors" to improve performance. I leave this to the
operating system and library. I have severe doubts about this being
reasonable on all operating systems. YMMV.

- documentation: write some accurate documentation including all the
limitations and stuff not implemented.

- POSIX interface: implement all the missing POSIX support.

- ..: This probably requires writing a recursive loader that fills in
the parent id for each file entry representing a directory. OLE .DOC
format is badly broken if you want a "real" filesystem.

- writing (with proper concurrency control): a complete implementation
probably requires a daemon to keep track of things across multiple
processes. Most purposes should be satisfied by keeping a central
record of who has what open how, including understanding of
multithreading issues. An aray of pointers allocated in
ole::ole(istream *) or some method called by that method looks like
the way to do this.

N.B. It is MANDATORY, due to limitations of the .DOC format, that you
sense when a file <1K grows to more than 1K you move its data from
small blocks into big blocks. via-versa is also MANDATORY when a file
shrinks. 1K files live in big blocks. Yuck.

- Implement the Micro$ucks interface for M$ users: The M$ interface
sucks but M$ programmers will want this. I implement a subset of the
POSIX interface and try hard to make sure it looks just like using a
real filesystem modulo the limitations. (.. being by far the biggest
issue).

- Support class id's: should be easy enough. I can think of no earthly
use for these myself. Maybe you have one.

- Implement RPC and that sort of gunk: this should be easy given the
right information. Be aware of the live data this creates and make
*very* sure it is easy to disable. Please use support Sun RPC as
software ag's stuff crashes too easily to be trusted as root. The
source is secret just to make sure we can not fix security problems.
[Finally the package is liable to be $$$$, so useless for truly free
software. Need any more bad news?]

We all trust security fixed versions of the portmapper and the source
code is avaiable so we can check the security ourselves. Secure RPC
has good quality security and is being implemented on Linux at
least. The implementation will hopefully inspire porters to get it
running on other systems too.


OK, that should keep the OLE implementors happy for a long while.
Patches to address all the above accepted.


