Graphical scripting toolkit

[fathertime]

Decided to just copy and past this post I made from the NWNX forums. It's talking about a little side project I'm working on. eventually, once it's in a more polished form, I thought it might be interesting to some of you folks here.

Lemme know if there's any interest in something like this.

I'm using the MNXJ plugin to hook into a java application, which allows you to script out behaviors using a visual task decomposition language called TMK (Task-Method-Knowledge).

Currently, it's in the very early prototype stages, but it's coming along nicely. Linked below...
i148.photobucket.com/albums/s35/FatherTimex/PerceptionReaction.jpg
... is a picture of an early model I've hooked up to the OnPerception script for a creature in a module. At this stage, it's fairly simple.. it responds to various perceptions, and does things like walk to the source, say various things, etc. It should give you an idea of what this stuff looks like when it's made. It's worth noting that this isn't just a picture or diagram, but is the actual program that is called by the NWN script.

You can see, it's a bit easier to deal with than the text based scripting language. It's easy to see the branch points, and follow what will happen as the thing runs.

Right now, I'm working on the java bridge code to extend the API availible to the graphical interface. Since this stuff is being done outside NWN, it can't directly call on the script functions, so some bridge code needs to be made to allow strings to trigger actual functions in the script. Currently it can do basic things like move to objects and positions, talk and listen to chat, and a few other things.

One of the nice things is that the models are somewhat separate from the actual module being run.. meaning that you can change the behavior model while the module is running, and the next time the creature it's attached to spawns, it will use the new behavior, without needing to rebuild the module or edit any of the scripts.

The language itself is actually reflective, similar to interpreted languages like LISP. This means that when the models run, they can reflect upon their own structure, and change it. For instance, they could add a new way of interacting with a situation on the fly.

Anyway, I just wanted to thank you guys again for producing the NWNX stuff, and I thought someone here might be interested in this work I'm doing. I'm hoping that the MNXJ plugin will be ported to NWNX4, but I may end up having to do it myself. I do intend to do that eventually though, as most of the stuff I'm working on now should be easilly ported to the new framework (assuming the basic MNXJ commands stay the same).

Hopefully, I'll be able to release this stuff to the community at some point in the future once it's in a more complete state, and it will prove useful to someone.

[Cecil]

Well, the one thing that PW owners will probably be concerned about is whether this will lag down their server.

Also, maybe I don't fully understand it(I am rather sleepy right now so not in a clear mindstate hehe) but can't this be done less messily with just a bunch of switches in the onPerception slots for npcs?

[fathertime]

In terms of lag, it would likely increase the processor load (assuming that you're doing fairly complex stuff), but all of the processing is taking place in a separate VM, so I think that would tend to minimize the "lag" that people on the server experienced. So, for instance, if you had some complex AI going, the module wouldn't be waiting around for it to figure out what to do... and unless you hit the actual processor ceiling, you really shouldn't notice much. Of course, it depends on what kind of hardware you're running on. Later on I'll try to optimize things though, and provide some metrics for what kind of performance hit it'd generate.

In terms of doing it with a bunch of switches in the onPerception slots for npc's, I would suspect that much of it could. However, I suspect that with increased complexity (not the toy example that the image posted shows) you'd quickly have some pretty complex code. I'm curious though, what aspect you're considering "messy" here in the example? I might be able to clear some things up.

[Cecil]

Messiness with connecting external tools to the server. I guess I'm just somewhat scarred from playing on one server, many years ago, that had a databse using javascript, and it lagged the server to hell. (if anyone recalls it, it was menzoberranzan)

[fathertime]

Heh, well that messiness will be behind the scenes.. the end user won't need to deal with all of that stuff. The final experience will be geared towards a tool that will let someone running a module to graphically manipulate the scripts on the fly while the module is running.

In terms of performance, I think it's a given that it'd only be real useful if the performance is good, so don't worry about that. From past experience, I can tell you that I've done similar integration with other more heavyweight simulation environments controlling a lot of entities at once, and performance really shouldn't be an issue.

What you tend to see is that you get an initial jump in resource consumption (a few megs of memory, fairly low processor usage) for the java VM, but a much smaller increase per script being run, so the curve representing the resource usage flattens out considerably.

And, again, I think the fact that a lot of the stuff is running off in a separate VM decreases the kind of lag you're describing. If the game is tied hard to the external stuff, and has to wait while that stuff gets done, then you can see lag.. but if that external stuff is being done in parallel, it tends not to slow the server down from the perspective of the players.

Right now, the only "lag" I'm seeing isn't actual lag, but is rather a slight delay in AI controlled entities reacting... but this isn't related to any kind of resource issue, but is rather due to the fact that the NPC's are getting their action queues modified on their heartbeat, so that's somewhat limiting how fast things go.. although it doesn't seem to be that big an issue yet.. A monster will see a PC, and will wait a second before responding. In a twitch based game, it might be an issue, but in NWN it should be ok.

The big problem though at this point is that the NWNX I'm using is version 2, for NWN1... I'll need to port some stuff to get it to work with NWNX4, for NWN2.

I think all your concerns are totally reasonable though, and they're things I'm keeping in mind. So far, I think it'll be ok, but ya never really know until ya do it.

[Cecil]

Well, I can't wait to see it when it's done then. Maybe I'll actually install nwnx on the server too(I haven't yet for the same reason. I worry about the lag hit from a database, but maybe things are better nowadays)

[fathertime]

Actually, from what I've seen with NWNX, you can actually achieve a significant performance boost if you're making use of a database, as the one with NWN is somewhat sub-par. I can't speak from first hand knowledge of the thing though, as I've not run any big PW's myself.