incursion rewrite experiment

[Richard T]

Hello stalwarts,

It's great to see some recent list activity.

The Incursion source code as you know is dated from the 90's. It is quite an accomplishment by Julian in bringing together what are now dated things like a compiler compiler and a C++ preprocessor to build the modules. And the code is dated too, it likely predates the std namespaces for programming fundamentals like strings, lists and so on. And then it is 32-bit and things are easier these days if you're 64-bit. All this adds up to mean that it's become a legacy project which has kind of fallen into a natural cycle of decay. You can still run the old builds but you have to deal with the unfixed problems. 

Some people have made the odd bug fix in their own forks of the code, there's even someone who merged some stuff into my old long neglected bitbucket repository that moved to github. I did engage in an effort to try and modernise it a year or two ago but it was encumbered by all the details in the first paragraph. Someone else did offer to chip in and help out, and followed through but they had strong personal programming preferences that I found incompatible (not skill issue but modern directional related).

So what's up with this rewrite? Well, things have changed. We have "AI", but since it only approximates intelligence through probability and loops and my mother calls it A1, I am going to use her term for it. Anyway, a week ago in order to improve my skills I paid for a claude code subscription and have been exploring how to use it best. One of the projects was the Incursion one, because it occupies space in my head. 

Well, I pointed A1 at my working Incursion install and said convert it while I do other things. And well, it's making progress. It has written a script parser to load in all the Goblin Cave module content, it has a dungeon generator and viewer. It has written itself tools to stress test and play the game itself. I even had it research how to run a debugger and diagnose memory leaks. It's still a long way from playable but it is looking promising. Again, this isn't real intelligence it's mostly like having a pretty good team of developers who lack a bit of context and need to be kept on track.

So what we have is probably the only possible path forward for Incursion that just moves us away from the dated code base. The goal for the project is to get the existing module content playable. Once it gets to that point we have a lot of other options.

Some of you may have objections to AI and maybe you don't want a modern Incursion out of principle, well, that is your choice. But I want to never look at the old Incursion code ever again. And I want Incursion atrophying to stop occupying space in my mind.

Anyway, thoughts?  Bueller  ..  Bueller  .. Bueller?

https://github.com/rmtew/incursion-roguelike-jai

Cheers,

Richard.

[Popo McTroll]

I know that some people are very much against the use of AI, especially when it comes to video game development, but honestly I don't mind it as long as it gets the job done well. I don't know anything about coding and I'm probably the worst roguelike player in the world, hehe, but I look forward to trying out new versions and seeing how they play. 

[Richard T]

There are three aspects to the project.

1. It needs to actually produce something worth sharing yet. There is a running dungeon viewer but what it shows needs to be more "there".

2. It needs to faithfully reproduce the dungeon generation, and the game play.

3. Then we're into nightmare mode. It is a lot easier for us to have ideas on how to add the things that were never done, than to replicate faithfully how another code-base works. I'm not going to add ideas here because that's putting the cart before the horse.

Also non-programmers like yourself can get AI to make slop changes and try things out. A lot of my development is now chat-driven development, I tell it what I want done and it does it. Now, I have a lot of programming experience and I read what it says and suggest technical observations and preferences, but that's for changes I want to keep. It's not quite that simple, because I am using the jai programming language.

Proof is in the pudding. When it has gotten dungeon generation working I'll publish the files.

--

---
You received this message because you are subscribed to the Google Groups "Incursion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to incursion+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/incursion/6188a395-099b-4eb0-9010-a8b71045a9c4n%40googlegroups.com.

[Frumple]

Well... good luck? I'm on the fence-ish about AI and generally avoidant of AI-gen stuff (the pile of ethical and infrastructure issues surrounding it plus trouble with the whole "fuzzy jpeg of the internet" thing, mostly, makes it where it's generally easier to just not for anything I'd be likely to use it for, ha), but if it works for you it works for you. Coding specifically is one of the use cases for it I'm least bothered by, too. 

If you can wrangle the hallucination machine into something workable, that'll be fairly neat, heh.

[Sire]

This is fantastic news! More people should know about this great game.

My concern, having done a TON of AI coding, is you're making it very difficult for yourself by selecting a programming language that is basically *undocumented* from the perspective of Claude Code.

If you would have selected a more well-known programming language, closer to C (C++ seems like the obvious choice, or C#) -- the AI could run almost by itself, converting lots of code autonomously, sometimes running for hours without making mistakes. (Look up the Ralph Wiggum method for really long-running tasks).

I'm sure you've already considered this. And I guess the only reason you're doing the conversion is because you want to learn Jai :) If so, fair enough, and best of luck to you!

Simon

[Richard T]

I appreciate the concern but It has no problem with jai syntax. Getting an AI to convert it to C should be an easy project for anyone interested once it is done.

The real problem is faithful reproduction. It came up with a stress test of analysing 1000 dungeons, then decided that a correct conversion was making the number go up not adhering to the original algorithms. It's a learning experience of how to make best use of this new tool and we're iterating on the approach. 

Anyway, there's not much more to add until there's something to show for it. 

--

---
You received this message because you are subscribed to the Google Groups "Incursion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to incursion+...@googlegroups.com.

To view this discussion visit https://groups.google.com/d/msgid/incursion/cdab44c5-b4c9-4b3e-b2a3-af514775027bn%40googlegroups.com.

[Sire]

> The real problem is faithful reproduction

A tip if you haven't already discovered this: Claude Code gets many times more powerful if it can verify it's own work. You could spit out 100 dungeons using existing code as a reference, then always compare your new outputs, continuing iteratively until it matches the style and look. They don't have to be identical, it can even compare images and find patterns.

https://code.claude.com/docs/en/best-practices#give-claude-a-way-to-verify-its-work

[J J]

As far as I'm concerned about the ethics of AI (or A1) in projects like these:  If this isn't going to be a product people sell, then money isn't involved.  No starving programmers are losing their lunch money because there never was any lunch money to begin with.  

I'm just happy that the old "ugly" webpages are still up.  I adapted the pantheon for use in a d20 game I ran a decade ago because I liked the textures that some of the gods had.  The god of chivalry also being a classist and a supremacist, the goddess of death also being the goddess of romantic love (and capital G Good), the goddess of lust also being the goddess of travel, new experiences, and the two of them being siblings making them basically a perfect complement to each other.

Aside from the occasional random game-crashing bugs, the one thing I remember being a big problem was Intimidation and rooms with tons of monsters.  Crit a goblin, get soft-locked by a pile of messages about other goblins being afraid, fleeing, etc.  It was too much for .y old potato of a laptop to handle.

I guess the only question I have is whether there's a donation page somewhere.  Pet projects run on love, but also on pizza.

--

---
You received this message because you are subscribed to the Google Groups "Incursion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to incursion+...@googlegroups.com.

To view this discussion visit https://groups.google.com/d/msgid/incursion/CAN%3DX-THCn7bMiV4q%3DDZxootRV8m7Q_WjoArG7JbQ5vDKv4be6A%40mail.gmail.com.

[Richard T]

No worries, it's running the dungeon generation in both original source code and new source code and comparing. 2333 steps in sync for random seed 3.

To view this discussion visit https://groups.google.com/d/msgid/incursion/d8d3f8f3-cf54-4ef0-ac1f-2f02988162b1n%40googlegroups.com.

[Richard T]

The website is hosted on github now in the repo for the original source code. It should be pretty much static, and if I put one up for the new version I'll make sure to have an "old website" link. Offer of donations appreciated but at this stage it is premature. If the new version gets released, maybe. Cheers.

To view this discussion visit https://groups.google.com/d/msgid/incursion/CADc07W%3DEvr4RO1HRHK-CWYy2e%2BzU6V0oogvYtC6LWuR74%2BFJ5A%40mail.gmail.com.