Does the scheduler die?  It sometimes trips spuriously when its parent
   is being debugged, but I haven't got it to die when running "au naturel".
   Still, the location when it does coredump is suspicious --- a sprintf
   into interp->result, which may not always point to allocated storage
   in tcl 8.x...

Check ns_conn url and ns_url2file for behavior wrt PATH_INFO; fix if needed.

Remaining db stuff:

  *) Compile and test oracle driver.

  *) With my Ns_Sets, you need an Ns_SetTrunc(.., 0) to actually reclaim
     storage for anything that's been put in the set in the meantime (due
     to use of the Apache pools).  The result is that right now, a select
     consumes the storage required for all rows retrieved before it is
     finally reclaimed.  It may be easier to fix this in the set code than
     the driver (I'd need something that wipes the values without affecting
     the keys; I could then just call that from Ns_GetRow), but it needs
     some attention.  See also the ns_set related item below.

  *) run CheckPools after each request.  (Obviously doesn't work
     to run it as a scheduled proc, as that would only check the
     pools in the scheduler process...).

  *) Write the support code for Apache modules that load db modules

Interaction with virtual servers needs testing.

ns_server active --- requires extended status; and won't give full URLs
   even then (you only get the first few dozen bytes of the request, so
   long request lines will be truncated).  Currently stubbed out.

Redo ns_sets so that the ns_set structure itself is not in the OWN_POOL,
  if we have one; then we can clear the pool containing the real_table
  without flushing and potentially invalidating the ns_set structure
  itself.  (Right now, we make the correct but dicey assumption that ---
  *without* ALLOC_USE_MALLOC --- we will reallocate the ns_set structure
  right on top of its original location, so that pointers to it will
  remain valid).

ns_write header reparsing

run registered procs in an ordinary response handler, not (like
   filters) in fixups

Look at diffs between aolserver 3b61 and 3rc1; anything new we want
   to drag in?  (Don't want to keep the sprintf(interp->result, ...)
   crap if they haven't...)

ns_returnfile needs if-modified-since support.

properly seed the random number generator

Figure out "invalid method" errors when a POST to an ADP script
   encounters a tcl syntax error...

???Switch ns_share to namespace implementation