memeticai: A Makefile repository from squattingmonk

The MemeticAI Toolkit

March 20, 2004

In case you haven't caught on, this project is very open, flexible
collection of scripts for building responsive, modular NPCs. One
of the challenges and exciting benefits of this framework is that
you can create a single behavior system in many different ways.
Unfortunately this can be a daunting task  -- especially for those
who are not familiar with software frameworks.

Time and time again we see each solution shaped by the specific
circumstance:

* What is the technical skill of the user?
* What are the efficiency requirements of the module?
* Is this for a single-player or multi-player environment?
* How much time does the user want to put into the development?
* Will this be contributed back to the MemeticAI project?
* Can this behavior be combined with other behaviors?

To help develop a roadmap for learning and using the toolkit, several 
types of people have been considered. Let's call them builders,
module developers, memetic developers, and core developers.

1. Builders

We've assumed that builders have little scripting knowledge and will
generally want to make Memetic NPCs by dropping out prefabricated
creatures, adjusting variables through the toolkit Variable Dialog,
or by adjusting sample spawn scripts.

 * Builders should immediately have blueprints of memetic NPCs and be able
   to understand that classes of behaviors can be assembled to influence
   how the NPC functions.

 * They will need to learn about the NPC's public variables. These are
   high level variables for setting common things like schedules, classes
   and text responses. Eventaully, we'll provide a web-based tool to
   browser and document these variable.

 * They will need to learn about the notation used by MeGetConfString().
   This is just a shorthand for mapping an NPC's variable to a variable
   on the module object. It's quite useful if you have a set of NPCs that
   need to share a talk table.
 
2. Module Developers

Module developers generally have programming ability in C or NWScript
and are generally task-oriented. We assume they've read over the
toolkit's APIs and want to take real control of their NPCs. These folks
will probably need to progress through the steep learning curve in stages:

  * They should have a good understanding of the generic NPC classes
    and that they respond to things like combat and doors with response
    tables and generators.

  * They should be pointed to the response tables first to do a
    variety of simple things including idle activities and combat
    strategies. This involves learning how to make a function in a library.

  * They'll want to learn how to author a class. To start, it's probably
    best to model them after an existing class and use pre-written memes,
    or response tables.

  * They should understand memes, priorities, and how memetic objects
    are stored in a library with registered member functions. This gets
    them on the road to implement any memetic object. It really requires
    that they understand MEME_SELF and NPC_SELF before they go much
    further.

3. Memetic Developers

At some point, module developers will be comfortable with building
layered response systems and will start delving deeper into the toolkit.
Many of the APIs will just have to be learned as they go. In particular
the falling topics should be addressed:

  * Debugging via the XML log system is enormously helpful. This cannot
    be stressed enough. A Windows utility can be found in the utils/
    directory in CVS to help format the log file for viewing in IE.
    Once they're developing generators, events, or preempting memes
    they really should be looking at the formatted logs.
  
  * Sure, lists can be picked up by anyone, but temporary variables
    and non-linear time routines are just as critical. These should be
    introduced to influence the development module-independent code.
    What's written for one module may not best for all modules.
  
  * Finally, communication, synchronization and module-wide scheduling
    will require that the developer learn about events, message passing,
    channels and the moment-based scheduler. These are tremendously neat
    things used inconjuction with Points of Interest (PoIs) and quest
    systems.
    
4. Core Toolkit Developers

Of course there's lots of us who have no real desire to make a module. :)
We're stuck in the pattern of reading every line of source, trying to
figure out what makes things tick, before we ever trust that it should be
used. Well, if you're in this camp it's nice to have another set of eyes
looking over the code. There are still bugs to be found, and we're all
open to explain why something's done a particular way.

Nevertheless, there's a progression for core developers as well:

  * The first phase most people go through the "is it efficient" syndrome.
    Experienced developers will note that I've implemented C++ style
    inheritance, with variable binding and mapping, declaration tables,
    list handling and a slew of other things in a btye compiled language
    without arrays. Yeah, it's rough on the CPU, get over it. Everyone's
    asked themselves, "Just how many cycles is "MEME_"+"XYZ"+IntToString(n);?"
    or "How many hash lookups does MeGetStringRef() really do?" and
    "You create a hidden in-game object to represent a data structure!?"

    The truth is, if you spend too long thinking about it, you'll probably
    feel ill and lose all the motivation to do great things. Just let it
    go and be impressed that modern CPUs are so darn fast. We've done our
    best to trade RAM for CPU wherever possible and have done some
    benchmarking to choose the most efficient path, wherever possible.
    (In general, rational thought was put into the design and implementation
    of the project.)
    
  * Continuing past the first phase, most folks want to know how the
    scheduler works. They'll dive into h_ai and find that Bioware has
    made an action queue and that the toolkit knows when a meme is "done"
    because I added a function to the queue that tells me its done. Your
    first thought will likely be "That's clever." until about ten seconds
    later when you realize that if your action queue is ever cleared
    prematurely, your NPC is brain dead until the next time a generator
    fires, creating a higher priority meme. At which point, you'll likely
    think "That sucks." And you'd be right.
    
    Luckily, nothing in Bioware's system should clear that queue. If it
    happens, it's essentially a bug. (In reality there are a few corner
    cases where this happens, but this is far less of an issue than it
    was two years ago.) 

  * Further down the road of tinker-town you might find yourself wondering
    why there are these DelayCommand(0.0,...) calls littered throughout the
    code with the occassional DelayCommand(0.1,...) call. If asked we'd
    be more than happy to explain that the game engine processes all the
    creatures and scheduled scripts in a timeslice. Any given script
    is limited to a fixed number of instructions or maximum amount of
    processing time. If the script does too much, it's terminated. Those
    commands with 0.0 allow us to circumvent this limitation and ask the
    engine to run the command after all other queued scripts are run, in
    the *current* timeslice. The 0.1 asks that it get scheduled in the
    next immediate timeslice.
    
Regardless of your skill, everyone involved in this project hopes the
Memetic Toolkit can provide you with something. Perhaps it's useful,
entertaining or just plain technically challenging -- whatever it is, it's
here to stay.

William Bull (Kilkonie)
bbull@memeticai.org
squattingmonk/memeticai
