			BUGS:

The rn emulation, vrn, probably needs a lot of work.

The ndbm() code should be fixed to use network byte order in order to
enable data sharing between unlike machines.

The checkgroups message prepends all local groups to the newsgroups file
every time it's invoked. It shouldn't do that...the whole issue of newsgroup
list handling needs to be re-examined.

Lint chokes on the uucast() call in sendbatch.c -- why is this?

The pre-fetch strategy for subject lists forces too much crunching and
generates monster temp files. It needs to be changed to be more conservative;
possibly grab next two pages.

Expire -r in copying mode doesn't work.  Complicated abstraction problem
to fix. Rdhistory/wrhistory/edbm need to be rewritten to assume that the
upper layer does the freeing.

			TO DO:

Full filter language in rnkill.c, with cacheing support. The driver loop in
filter.c code should be fixed to allow the filter to cache information on new
articles while it's waiting for reader requests. Also, it ought to be possible
to make it log filtering info.

Change handling of external groups in authorization file? Mel has a point...

RFC1049-conformant Content-Type interpretation in D.news/msgopen.c. The
transport layer should never see this. Does it?

Mailbox cracking so we can read mail through news. Deletion is the sticky
part, effectively we might have to write or rip off somebody's generic
mailbox handler.

Notes-style display of the top messages of threads only, with count of
followups.

Do we have to assume that case in the unique parts of IDs will get smashed?
If so, we must change the timestamp encoding scheme.
Add SNEFRU hash function to ID generation code.

Fold remaining nrecnews features into inews.

The NEWNEWS command cannot be easily implemented for the nntpd server.
This is because there is no easy way to tell what all the new news
after a specified date is.  Assuming that history.dat is in
chronological order is wrong, since this will not be the case after an
expire.

Lint barfs on the readdir() calls. Why is this?

Now that we have response_codes.h, can we simplify the make procedure by
always compiling the network support? Is the space tradeoff worth it?
