TODO

TODO before release:
-----
rename directories: h -> include, o -> .libs
h/String.h -> STL string
h/List.h -> libstdc++
install should ignore CVS/ directories.
Rewrite terminal handling, use TERMIOS
    -> Split screen scrollback
    -> /window, /exec interface
    -> Handle SIGWINCH
    -> Use multiple input lines/status bar (multiple line status bar)
    -> Replace all TTY stuff with ncurses stuff (man ncurses)
Repeat -> C++, with inverse last command a la zmud.
Commands -> static autocomplete
Perl eval: %
Evaluating commands ('/') in send
finish removing one_argument, convert to OptionParser.
Make all keys in InputLine.cc be KEYPRESS hooks.
#ifndef BLAH_H protect all header files.
Uniform way of handling errors.  report() and error() should be used in the code,
    and configurable so that they are printed to the screen (w/attribs?) or displayed
    in the status bar.
/config command to handle what used to be /disable and /enable.
Error when 'sudo su - bob' and then run dirt (tty permissions)
Config.cc needs to be rewritten
Keys in Hotkey.cc, should be rewritten as individual hooks.
Remove restart via alt-t?  Seems such a kludge.
Chat subcommands in Chat.cc: handleUserCommand should all be hooks.
Where did this message come from: @@ Dumping %Config to file
    when running dirt as 'bob'?
Where does this message come from: read -1 bytes from socket 8, errno: (0)...
    when running more than one copy of dirt on the same machine.
    Seems that the second copy read()'s in a loop, spamming the CPU.
Commands to easily open a connection to a new mud, add a new mud, etc.
Editing hooks/aliases/triggers could simply copy the /trig (etc) line to the command line.
    for instance, when you hit 'e' from the window listing all triggers.
Change Session.cc so there are less string copies (triggerCheck, in particular)
    merge triggerCheck() with inputReady?
Retromud prompts are not grabbed.  Does retromud use Telnet GA?
    prompt seems to work for Abandoned Reality, but prompt is not removed from mud output.
*^H style prompts.
Titles, top/bottom messages for windows.
after a /clear, text appears at *bottom* of a window.
Window is not painted in background color -- tab, end of lines are unpainted. (when scrolling)
Use hash_map to store Windows by name for fast access (keypress hook)
Option to 'croak()' on perl errors?
When output is redirected to a file, screen (even more) badly corrupted.  Could be an indicator
    of which drawing routines are depending on tty, and which ones are using "print".
Write Python exception handler a la SIG{__WARN__} in perl.
Rearrange StackedInterpreter, so everything doesn't fail...what to do if both interpreters
    define a given function?  What to do if only one does?
/eval for python doesn't return values:     /eval -rLpython 1+2
Make sure in all circumstances, only ONE embedded interpreter attempts to run a command
    (re: /hook -f tries both...)
/hook -t"^(.*)$" -TOUTPUT -F printuncolored = /eval print "\$1\n";
    generates an error, as more output is received, dirt starts disabling *other* hooks,
    but not this one!
Perl warnings from hooks should print the hook name.
    /hook -t"^(.*)$" -TOUTPUT -F printuncolored = /eval print "$asdfasdf\n";
/eval -- print $Trigger::Triggers{'zsays'}->{'t'}
    doesn't work...
Key: alt-n, alt-p for next, previous match (alt-/)
Is ProxyWindow used at all?  (remove it)
Lines that wrap are missing the first character on the wrapped line.
Auto-hook disabler seems to disable the next hook run after an error (which is not always 
    the same hook!)  i.e. make a bad /trig and then type something -- it will disable tab_complete.
Consider a "gag" attribute: will still allow all hooks to match (and modify!) the line,
    but when it gets to the end, it will not be written to the screen.
Error parsing init script (i.e. trigger.pl) causes obscure error with no backtrace.
    error in parsing gag.pl generated NO message AT ALL.
Killing help window causes help to misbehave.  Check for window and recreate it if necessary?
/delay, /repeat, or /run command that will let me schedule something to be sent.
Should Window::focused be static?
/input, /inputline like command to set input line.
xterm doesn't send alt- key sequences properly...
/gag -Ft'^\\s*$' blankline
    doesn't work...because the string to be matched is "", and when it is gagged it is
    replaced with "" -- and dirt detects that they are the same.  Need another way to do
    gagging since "" is a valid mud output line.  (after crlf stripped -- maybe leave in
    crlf for triggers?)
    See Session.cc:693 (Session::triggerCheck)
Try to initiate a ytalk with the dirt terminal...hang? no...just fucks up the screen.
    Turn mesg n on dirt startup...
Make a perl class to do a scrollable text box (/help) and one to do a selection 
    list box (/alias, /trig, /gag, etc)
Add options to /group to do most /hook operations, on the WHOLE GROUP.
    Do the same with hook?  If no =, then try to change the attributes of a hook?
Focus mechanism:
    Keypresses which go with a "focused" window should grab higher priority when active.
    Should not be able to pull up command history while in scrollback
    Keypresses for modal windows (/help) should have higher priority than 
        background keys (scrollback)
    What about /help, history combo?  Who should get up/dn pgup/pgdn?
        help should trap up/dn to prevent history from popping up.
Option to use favorite language for /echo?  (i.e. /echo -> /eval print
    "<args>\n" so variable expansions work)
Saved.pl should not be clobbered if it has been modified since last loading.
/alias crap = crap $2 $1
    hangs dirt.
Hooks without a matcher will not have RHS expanded as a quoted string.  Is this bad?
Numpad doesn't work in xterm.
up/down doesn't work in eterm.
/paste /nopaste commands (when /paste is active, all user-typed commands have the 
    argument to /paste prepended)
/help doesn't behave so well if window is too small.
SEND, COMMAND hooks: rather than using regexes, grab first argument and then
    look in hash_map to see if it is a command (/hook -C)
    problem with multiple hooks with same name?
Convert usage of "Sprintf"...or not?
    http://www.bero.org/clap/classString.html
    child of string that adds useful routines (like sprintf)
/aliases should be added to static autocomplete list.
/aliases should be mud-specific.  Mud name is in $mud.  Modify config.pl
Define keypad-walking hooks in C++, with group "numpad"
Write script to convert existing mclrc files
/set command to set config variables.  (stuff in dirtrc)
    /set debug command to turn on debugging.  For instance, print which hook catches
    a line of output.  See Hook.cc:199 (Hook::run).
    Also, print ${color_code} rather than colors, so people can write regexes that match.
/enable blah;/hook -l blah
    -- semicolon parsing should come before command parsing.
Recursion counter like tf has.
How about this:
    Allow perl regex like syntax on the command line:
        /hook -t /this is a regex/ -F blah = blah
        /hook -t /this: (\w+) will match any word/ -F word = /echo $1
        /hook -t /this: won't give a warning about unmatched quotes/ noquotes = /echo woo!
    Quoted regexes are subject to normal parsing rules.
    Denote some regex delmiters and find the end the same way as in FilterProxy.
    From perl: &run should take arguments:
        &run("/hook myhook = /echo shit", '-t' => qr/this is a regex/);
        any option given in the hash that follows will be transposed onto the command line
        with the proper escaping.
        Is this possible:
            &run("/hook -t'" . /my regex/ . "' -F shit = /echo shit");
prompts terminated by an empty packet? (Some Diku's, apparently...try 
    duntroon.xidus.net:7000)  How to detect this?  Seems to be a TCP Ack.
    Consider putting things on the prompt line if nothing has been received in
    a few ms.
    So for these kinds of muds use these options:
        showprompt=0
        undelim_prompt=1
Mud-specific options: showprompt, undelim_prompt, echoinput.
    Or do things like TinyFugue with DIKU, LP, Telnet-style muds?
echosend option (mud-specific)
    Print to screen commands as they're sent (as if you had typed them in telnet)
    stuff into input buffer, as undelim_prompt does now.
when creating a /command, give option -A to create an anonymouse perl function.
    the anonymous perl function should call getopt and create $opt_.. variables
    for any args, and split args into @_ so commands can use positional arguments.
command parser seems to be grabbing the wrong equals sign:
    /command -F repeat = /eval -Lperl '@ARG = split(/\\\\w+/,$_); run("/echo found repeat $_")'
    or not...why doesn't this work when the equal sign is in there?
    /command -F repeat = /eval -Lperl \@ARG = split(/\\s+/,"$_"); for(\$i=0;\$i<\$ARG[1];\$i++) { run(\$ARG[2]); }
    Instead of escaping *everything* treat the RHS of /eval as a quoted string
    and generate a function like this:
        sub { $_ = "$_"; eval "$command"; }
Grab Portal mud information:
    http://www.gameaxle.com/PortalMIP8-0.pdf
Note howto get key name:  key_name(key)  (see old Screen:keypress)
/enable -T repg does not tell me that there is no trigger named 'repg'!
Keypress hooks really need to be passed the cursor position.
Other hooks really need to be passed the parenthesized subexpression matches.
...consider adding another parameter to HookStub::operator()...
History::loadHistory does not deal well with long lines in history file (segfault)
autocomplete does not move cursor to end of line (again, perl needs to see cursor_pos)
    How about 2nd arg to operator() being a struct union?
    struct {
        int key;         // KEYPRESS
        int cursor_pos;  // KEYPRESS
        string& mudname; // CONNECT
        void* subex; // pointer to a perl list of ($1, $2, $3...)  what about python?
                     // can we count '(' to determine how many subexpressions?  How about string[] instead?
                     // $#+ returns number of subgroups in last successful match
                     // Need to store ($1...${$#+}, $&, $', $`, @-, @+)
                     // or just store mud line and trigger regex, and re-do matching just before
                     // calling the perl function.
        string& data;
    } hook_args;
    Hook::run(HookType type, string& data, hook_args arg)
    How to pass this to perl?  (as arguments?)
    Can save regex results with @shit = ($line =~ /$regex/) plus %&, $', $`
    The command stack needs another void*. 
        dequeue<pair<string,void*> > commands;  // Interpreter.h:39
    Is one extra piece of data enough?
                            string              void*
        COMMAND   = 0,   // /command            string& re
        USERINPUT = 1,   // what she typed      string& re
        SEND      = 2,   // c-ogres             string& re
        OUTPUT    = 3,   // The small box sh... string& re
        LOSELINK  = 4,   // mud name
        PROMPT    = 5,   // >                   string& re
        INIT      = 6,   // 
        DONE      = 7,   // 
        KEYPRESS  = 8,   // input line          key, cursor_pos (make cursor_pos global?)
        CONNECT   = 9,   // mud name
        IDLE      = 10   // time
    Provide PerlEmbeddedInterpreter::submatch(int n) that returns $n.
    Evaluate string as perl quoted expression in writeMUD.  (end of SEND hooks)
        make a config option for this for people that want to send [$@%&] to their mud.
    Hook argument: -q evaluate RHS as a quoted expression.  :)

    The RHS of a hook will be evaluated as a quoted string UNLESS the first 
    char is a '/'.  If it is a '/' then the mud line and matcher will be saved
    for the point at which the RHS is evaluated.
    
    Add note in docs that you can't combine /commands and mud commands with semicolons:
        drool;/echo $foo;fart

I don't think I like Saved.pl.  Make a stream-out and stream-in so that .dirt/Aliases,
    .dirt/Triggers, etc. have same format as people type them (/alias blah = blah).
/bell;/bell doesn't work.  Also need /beep that is same as /bell.
Comment out Option.cc:194?  Not needed?  Is s_nonopt already unquoted?
For COMMAND hooks, -t should ignore leading / (so that switching cmdChar doesn't
    hose all your perl scripts).
Place arguments in @ARGV to discourage people from doing:
    /hook -T COMMAND -t'/mycmd ([a-z]+) ([a-z]+)' twoargs = /run mycmd
    Make -t mutually exclusive with COMMAND
Remove dirtCommand
Tokenizer (use in Option.cc?) http://www.boost.org/libs/tokenizer/index.htm
PerlEmbeddedInterpreter::match determines if the match was successful by grabbing $_
    and seeing if it is empty.  This will FAIL if commands is empty...
/hook -t'shit' myshit = /run myshit
    doesn't work because if the RHS modifies the mud string and is fallthrough,
    hooks of lower priority won't see the modified string.  Why did I decide
    not to evaluate the RHS immediately?
        because of the interaction between /command RHS' and mudstring RHS'.
        if the RHS is '/run crap' and a higher priority hook is just 'foo',
        any mud commands generated by the perl func &crap will be sent to the
        mud before 'foo'.
        run("foo") goes to interpreter.add("foo") via InterpreterPipe::inputReady()
    so running things will get done in the correct order even if the RHS is
    evaluated immediately for /commands.  Printing things will not.  Consider:  
        /hook -Ft 'shit' -p 10 myshit = /eval print "found some shit\n"
        /hook -Ft 'shit' -p 0 myshit2 = /echo found some shit2
    Upon seeing 'shit' in the mud output, the second version will print first,
    despite having a lower priority.
        /echo sends data directly to outputWindow->printf()
        perl's print() sends it to the output pipe which is read with 
        Selectable::select()
    >> Changed /echo to use cout, make sure it works...
    It is desirable to modify the mud string, consider the prompt case.  A naive
    regex like "/hook -Ft'shit'" would match twice if the prompt string sent the
    mud line through twice.
Debug mode for hooks.  Dump everything that matches.
    /debug <HookType>       -- dump the names of hooks that matched the HookType
    /hook -lt 'adsf'        -- list the hooks that will match the string 'adsf'
    /hook -l -T COMMAND -t 'asdf'    -- etc.
Print an error message if a /command isn't caught by any hook.
    /asdf   -> unknown command: '/asdf'
Directly calling C++ (instead of interpreterPipe):    
    http://www-106.ibm.com/developerworks/linux/library/l-inline
    use C-Inline library?
    Swig?
    aaahhh seems to work well.  Now need to deal with removing Interpreter and
    getting return values back from /run.
        For most purposes "caught" (return value from hook.run) is equivalent
        to the /run return value.
    Python: exactly:
        http://www.python.org/doc/current/ext/extending-with-embedding.html

Investigate removing Interpreter, or making Interpreter not have a command stack.
    Replace with just direct calling.  Can also remove savedmatch in this case (i think).
Investigate removing Interpreter::mark
When in scrollback, put "(N new lines)" in the status bar.
/telnet command, or expand /open to take host/port arguments
Prompt detection: look at the size of the packets we receive.  If it is equal
    to the default packet size (1536?), then do not consider it a prompt until
    after some timeout.
    MTU discovery is a bitch and is variable.
Why do I get two prompts on the prompt line at materiamagica.com?
    Because we never add a \n to the input and neither do they...
    echoinput should add your typing to the input.
    --- currently prompt gets printed twice (why?)
Get rid of showprompt option?
    opposite of snarf_prompt?
Prompt help file: explain various options
Why does "/eval foo" give me an error on line 2?  But /eval foo;boo; give me an
    error on line 1.
gags broken...
Sometimes the first character on a line is missing...
Perl evaluation:
    /hook -Ft'A tall street light' lightgag = /eval $_=""
    /hook -Ft'A tall street light' lightgag = /eval $_="";
    the former generates:
    @ [Perl ERROR] Can't find string terminator '"' anywhere before EOF at (eval 1388) line 1.
/hook -t'follows Zathras into the room(.*)' followers = /eval print("Got after followers: '$1'\n");
    prints: 'ot after followers: '.
/hook -t'follows Zathras into the room(.*)' followers = /eval $foo=$1; print("Got after followers: ,$foo,\n");
    prints ,ot after followers: ,.
/run -Lperl save_configuration prints:
    @ Dumping perl > /run -Lperl save_configuration
    may just be command echoing overwriting the line.  (but it shouldn't, the
    dumping... line has a \n at the end) 
    This appears to be a python interaction:
    @ Dumping perl @ [ERROR] Python could not find function 'save_configuration' anywhere (args: )                   
When using autorepeat and echo command line, blank is echoed.
After typing one character, the cursor bounces back to the beginning of the line?
    Consumes a bunch of CPU too...
    Only when connected.
autorepeat only works if length of command is less than 4.
After /exec the window doesn't close on esc.
Segfault seems to occur in conjunction with autocomplete (and weird
    autocomplete results) -- try to autocomplete in the middle of an
    autocompleted word.
    autocomplete, backspace.  (is repeatable)
    hmm now I can't reproduce
/eval \$ThreeKingdoms::enemy="chinchilla"
    @ [Perl ERROR] Can't find string terminator '"' anywhere before EOF at (eval 2209) line 1, <FILE> line 46.       
    adding a semicolon gets rid of the error message.
Putting things on the screen fails for large xterms (128x90)

Hook:
-----
Why doesn't old grab_prompt work?  (no while loop -- i.e. recursive grabbing)
Inconsistency: perl stuff gets executed *after* the received buffer is printed,
    C++ stuff gets executed *before*.
fallthrough with perl functions doesn't actually work since we don't collect
    the return value -- we don't know if it was caught.  See TriggerHookStub::operator()
    FIXME CHECK THIS!!
KEYPRESS hooks: key is an int.  Convert it to a string (itoa) and pass it to hooks?
    /complete hook needs input line to do its thing (when tab is pressed).
    if -t is specified, checks for key and passes input line.
    if -t NOT specified, just passes key.
-- or --
    -k parameter to /hook: implies -T KEYPRESS.  If specified, key must match.  Input
    line will be passed to hookstub.   (Need KeypressHookStub class)

    Keypresses should be associated with a window.  How to do that?  /hook -W?
    put list of keys on window's status bar?  How about keys for open/alias/trig
    windows?
Should print every line received: /hook -T OUTPUT -F tester = /eval print "got: $_"
    how to do this?

Scripts:
--------
convert:
    auto/gag
        alt-g to bring up gags?
    auto/highlight
        alt-h to bring up highlights?
    auto/mudftp
        Converted, needs to be tested.
    auto/repeat
        bind KEYPRESS to clear line and place new text on command line.  Inverse last command.
    auto/trigger
        '/trig -e' to edit a trigger (copy it to the command line)
        Add '/trig -r' to run a named trigger.
        make alt-t bring up window with triggers in it. Pressing enter on one
        should copy it to the command line for modification.
    sys/idle
    sys/keys
Continue converting commands in Interpreter.cc
Message if command not found...
/help, alt-h should bring up help window.  bind pgup/pgdn to help window.
tell, soul, chatline, etc. windows.

Window:
-------
scroll down a line
scroll up a line
scroll down a page
scroll up a page
write message on top bar
write message on bottom bar
Do all this with /status?  Extend /window?
list bound keys on status line (right justified).

Screen:
-------
Ncurses seems well suited to all this.
Screen.cc, Window.cc use memset on canvas to draw to screen, Screen::refresh redraws.
curses has panel (info panel) to do overlapping windows. (info curses)
    stdscr is below all panels (for mud output)
must use panel(3X) library for overlapping windows.  (cannot use stdscr)