x. Fix builder flag setting so only wizards can do if no RESTRICTED_BUILDING.
x. Fix toading players to change macro ownership.
3. Pennies #defines.  (#52)
x. Fix headers for port number and other strange defines.
x. Set rwho to use the comment from the version #, not a seperate string.
x. Fix @dig parenting.
7. Add @q to quit mud editor or fix documentation.
x. Choose timeout for select more intelligently.
x. Fix check_prop_perms in db_prims.c to check for ':' in prop name.
xx. Fix restart script to be esaier to modify, remove 'return 0'.
xx. Default set of flags for new players
12. Start collecting up ''interesting'' additions from other people
    which might go into an unsupported examples directory tree.  Perhaps
    humor suggestions on certain currently unsupported things which might
    actually be useful.
xx. Stick stuff for TM_STRUCT_HAS_ZONE in config.h.
xx. Include example lockout.sites file.
15. Get basic admin documention finished.
16. Rewrite the sanity checker.
xx. Have the lockout file accept * in the dbref field.
18. Fix strftime docs to work like function (%x, maybe others).
xx. Change string routines to take char * and then internally change to
    use unsigned char *.
xx. There is a call to unparse_object in interface.c if !HOSTNAMES.  Fix this.
21. Edit db.h to let people to know to use Wizard(p) not HasFlag(p, WIZARD).
22. More security checking for the sanity checker.  Perhaps look for say/pose
    actions on objects which are in rooms/players other than their owner.
    Check for unlinked exits in #0, check for unlinked dark exits anywhere.
    Check for exits in #0 not owned by wizards.
    Any other ideas???  (email me)
23. Check that terminal detaching works (losing controlling tty).  rc/es
    problem?  Appears to be rc problem, maybe call setsid anyways?
24. Fix up flags.h and do up a mkflags.sh for easier addition/subtraction
    of flags?
25. Get the sanity checker to do something with macros.  Perhaps macros
    should be inside of the database anyways.  It's a pain being in two
    files.
26. Add the scripts that tt used to call from cron to make backups and
    to mail wizards appropriate files.  (Was this script lost?  If so
    it should be rewritten.)
27. Think about having the string interning do splits on -/ and other
    common seperators.  This could save a fair amount of memory, depending.
    I thought about this before and then didn't do it for some reason.
    New:  I thought about it a tiny bit, unlikely to save much memory at
    all, since prop names aren't where we waste memory??
    New2: actually we can waste memory in prop names if you have
    centralized db's like mail.  Splitting on -/ might pay off pretty well
    on those db's.  I don't know how many like that are still around, tt
    would have saved memory tho.
28. Document global variables and where they live.
29. Pull the various strings for the interface code into a common header for
    ease of change.  (currently they are split over interface.c and config.h).
xx. Allow homing of objects to people.
31. examine to show description
xx. MFLAGS doesn't work under newer gnu makes
xx. ^C during generation of config.h causes problems.
xx. global @stat for non-wizards.
xx. Check whispering to nonplayer objects.
36. Check @tel to match whole object names before partial players or
    whatever the heck it's doing.  Probably a problem in match.c.
37. Add runtime config file.
38. Muf manual entries for 'put' and fix 'rotate' to mention negative args.
xx. Make GOD into a variable head_player read in from the dumped db header 
    like player_start and global_environment.  Have sanity issue a warning
    if head_player isn't a wizard.
40. Clean up string compression code and maybe dump the bigraphs into
    the db header.  Can gain extra bigraphs by using unused spots in 
    isprint (like control characters).  Putting bigraphs in db header
    makes decompress need that information, which is a bit of a lose.
xx. Better support for god bits.  (#39)
xx. Stick code & frames into seperate tables managed by the interp code
    rather than keeping them on program and player objects.  Have the
    frame code have entries for an interactive program and a list
    of background programs for the player.  Have interp_quit_external
    just kill the interactive one.
43. Redo the db layer again to make it use less space and still be
    growth efficient.  Use a two tiered approach.  This takes no more
    accesses than current method, but will save mucho space.  In optimal
    best case the idea of using seperate types for a garbage object than
    a normal object can only win if (#Garbage * 10) > #NonGarbage.  This
    is very unlikely and also assumes you have an optimally smart malloc.
44. Write a tune program which the makefile compiles and runs to create
    a tune.h header file.  Use this to optimize allocation hunk sizes
    for various hunking routines.
xx. Have sanity checker issue fatal warnings if player_start and
    global_environment aren't rooms.  Also have it display the current
    values for these (and head_player) when in security mode.
xx. Put in code to dump hashtable statistics at a certain time interval.
47. Have code to do walks through property lists killing off junked
    properties whenever there is free time.  Have a junk_prop counter
    which keeps the # of currently junked properties on each object.  This
    way you can tell when you're done.  Maybe walk down 1 object everytime
    there is X seconds with no activity.  Note that inline dumps do
    this compaction on the whole db at dump time.
48. Add more usage warnings to the sanity checker.  This should probably
    get an option instead of being default like now.  Check for # of
    objects in contents lists, # of actions on an object.  Check for
    duplicate exit names on rooms and objects.  Check total number of
    properties and string bytes for a player.  Maybe some sort of
    exit ratio test.  Check for undesced/unsucc things, etc.
49. Have the sanity checker remove exteneous files in the muf directory.
xx. Stick the prop_list structure directly on the general_object
    structure.  The current method of having it as a seperate object
    was designed to save memory for objects with no properties.  But
    this can only save memory if (#NoProp * 3) > #Prop.
    And if a junked counter goes in this structure then it's down
    to (#NoProp * 4) > #Prop, which isn't going to happen in a real db.
51. Stick back in the server versions of map-list and sort-list.
    Think about a sort-keyed-list.  Think about having sort-keyed-list
    cheat for doing an integer sort (do them all in one pass w/ radix)
    and semi cheat for string sort doing maybe X swaps on each drop in
    instead of just 1.
52. Quota system hooks instead of pennies.  Provide an emulation of
    pennies in the system, and also provide a simple object # version
    which uses a db object referenced off of global_environment for
    the normal max count and extra deltas for specific players.  (#3)
53. If there was a quota system, does it make sense for the sanity
    checker to make sure that no one is over their quota?  That probably
    wouldn't work in the case that the quota system called to a muf
    program.
53. Timestamps.  How do the other servers decide on what exactly
    constitutes an access, a write, or a use.  Check fuzzball, daemon,
    and dragonmud.
54. Make server page do echo?
55. Standard macro library, obviously.
56. Server versions of toupper/tolower.  Other backings of certain
    standard library functions?
57. Call a compression program to handle dumped db's.  On forked dumps
    just call it.  On inline dumps use vfork, except that vfork is broken
    on the same machines fork is broken.  If you do inline dumps, no
    compression?
58. Recompress property names in dump again?  Problem is that compressed
    strings aren't directly comparable wrt case insensitivity.  So this
    only saves a little space in the disk image.
xx. Check if program names still have that stupid default scroll description
    set on them, and remove it if so.
60. Add a typecheck primitive, this lets you write a compiler which
    does compile time type checking be able to call and be called
    by other programs.
61. Change the current type checking to first use a lookup table
    which gets indexed by the type of the top two stack types and
    the bytecode.  (Note will want to have two dummy entries off
    the bottom of the stack.)  Should improve speed.
xx. Run programs on #0 and player on connect/disconnect?
63. Read over the boolexp code carefully.  There might be a bug in it.
    Also recheck the new recycle code again.
64. All of match.c should be rewritten.  It's still a mess, and a lot
    of time is spent in the exit matching code.  The exit matching
    code does some pretty inefficient things, too.  Checks locks
    multiple times, bad bad.
65. Have the compiler/interpreter use an actual bytecode as opposed
    to the current stream of 8 byte structures.  Even a trivial splitting
    of the struct inst's in the func_addr structures into 2 arrays
    one of the type and the other of the data would save a bit.  The type
    would fit in 1 byte, so this would save 3/8 of the code space.  Of
    course an actual bytecode method would save more (but unclear how
    much more).
66. Use some sort of last used timestamp on programs.  And periodically
    throw out the bytecode to programs not recently used.  This will
    force them to recompile if used later, but maybe free up some
    memory.
67. Use a similar (65.) splitting of the type from the data for
    the address and datastacks of the frame.  This will reduce size
    of frames from just over 8k to just over 5k.
68. Maybe have the makefile determine if tolower or toupper are broken
    on this machine, and in that case provide our own version?
69. Own version of strncasecmp (not everywhere, not ansi). 
70. Would splitting on spaces in interned strings save space?  Have
    one type of interned string be an array of words, you'd pass over
    the string once counting spaces, then allocate the array, then
    pass over it again filling in the consituent parts.  Each item in that
    array is an interned item.  This would essentially be giving a word
    compression on the database.  It'd only save space if there were
    a lot of common words (true) and if the average length of these
    words was greater than the 4+ bytes needed to hold this extra pointer
    (probably not true).  This also makes the procedure for getting the
    text of an interned string harder, but any sort of splitting scheme
    has that drawback.  You could perhaps make a decision about the
    cost incurred for strings which don't have lots of the longer common
    words in them and just store them unexpanded (not sure how to do this).
71. For wizard docs besure to mention useful things like that if you are
    getting rid of a lot of people, that it's smarter to toad them all to
    a special player whose only purpose is to temporarily hold their
    objects, then @purge that special player.
72. Wiz-only primitive to get next property.  A primitive to get the
    next object owned by someone.  (nextown, ownnext, ??)
73. Have sanity checks inside the main program, as opposed to just in the
    seperate sanity checker.  Majority of time spent in 'sanity' is 
    load time.  This is a waste, and encourages people to not sanity check
    their db's.  Have the fatal error sanity checking in the main program,
    perhaps with an option to do the various warning checking then as
    well.  Also provide a smaller standalone version (like the old one)
    which also does this.  The repair operations could either go in both
    or just the smaller standalone checker.
74. Port concentrator, set up to allow the concentrators to run on other
    hosts.  This is useful for forwarding, as well as concentrating.
xx. Make sure that text of programs is removed when you get a new program
    via @prog, in case the old one didn't get erased.
xx. Change from using wait3 to just waitpid.
xx. How about #ifdef HAVE_UNISTD_H, and let people set it in config.h.
xx. For systems with unistd.h, maybe use sconf to get OPEN_MAX, instead
    of using getdtablesize.
79. Credits file.
80. Have a way of substituting runtime variables into the help/man/etc
    documentation when it is displayed.  This is particularly useful/easy
    to do if a runtime config method shows up.
81. Make it so that ansify.h doesn't need to be editted.  Have
    entries in config.h called 'BROKEN_TOLOWER' or similar.  Perhaps
    rename ansify.h then. 
xx. Rewrite the laston script to be far more intelligent.  Have it
    get dbrefs & names from the db (indexed by dbref in an awk array)
    also be sure to count up the number of objects owned by dbref.
    Then go through the status log keeping track of the latest
    connect time (both as integer and string) for each dbref.  Spit
    all of this output out sorted by connect time.  This way you can
    easily see who the old players are and also how much they owned.
83. Think about putting the normal docs into mud/docs/* instead of
    mud/data/*.  That makes it easier to share among multiple
    servers on same machine or such, could just be a link to the
    docs directory for the source.  Also, should think about
    renaming forth.ref to man.txt or something.
84. Check out the h_errno stuff for gethostbyaddr and see how portable
    it is.
85. db muf predicates work on !ok? things
86. need to make bufsprint understand widths for integers and strings
87. fix editor to print line numbers correctly again.
88. fix locks so you can call 'match' w/o nuking args
89. @attach by number, check all @commands for more uniform and
    liberal matching.
90. Extract and merge programs.
91. Fix lockout.c to include the needed files for netinet/in.h.
92. Store the uptime somewhere.
xx. Possibly allow tzset(), tzname[] to get timezone info.
93. Write a type checker optimizer for muf which makes use of the
    already existing ability to mark a word as compile time
    type checked.  This could drastically help, since a lot of interp
    time is spent type checking.  Of course, only helps if a reasonable
    portion of words are able to be compile time checked.
94. Pull out the WHO stuff (and maybe OUTPUT*) from interface.c.  With
    the con* routines you can do this in other ways.  For efficiency
    there should probably be an array indexed by fd# to the descriptor
    struct.
95. "help commandname" doesn't work if the commandname is in the file
    as "name1,name2" where name2 is the name you're using, and you
    didn't put a space in the file before it.
96. Do block diagrams of data structures and show what the fields are
    and the connections.  For example the struct code pieces and hash
    table they are indexed from, and when they are unconnected, etc.
97. Restart script doesn't like it when there is no db.new file.
98. Object creation primitives.  Do you have a 'set_type' command?  Or
    a creation command?  What about getnlink, setnlink?  What other
    creation like primitives are missing?  recycle from muf?
xx. Have @force notify the victim of who forced them and what they were
    forced to do.
xxx. Notify both local and remote rooms when a remote kill is performed.


TODO,v 2.38 1997/07/03 00:15:44 dmoore Exp