4 ms·

The blog post is from 2005[1]. [1]: http://www.a1k0n.net/ http://www.a1k0n.net/

a1k0n 15y ago
And really not all that interesting. Weird. Hi, HN. Also, we ended up ditching Lisp. Goofing around in Lisp when we had no idea how to deploy it wasn't really a smart move for a four-man startup, in retrospect.
- rcfox 15y ago
  > Goofing around in Lisp when we had no idea how to deploy it wasn't really a smart move for a four-man startup, in retrospect. Would you mind expanding on that? I'd be interested in hearing about why you chose Lisp and why it didn't work out.
  - a1k0n 15y ago
    Well, we wanted to add some "life" to the universe with some automated trade convoys and so forth, and I had the crazy idea to use a rule-based system to implement it, and found an off-the-shelf clone of CLIPS (the expert system type thing), using the Rete algorithm called Lisa (http://lisa.sourceforge.net/ http://lisa.sourceforge.net/), and that was implemented in Common Lisp. It turned out that the impedance mismatch between that and what I actually wanted to do was gigantic, and Lisa was way too slow as it would re-evaluate the whole world anytime anything changed, so I ended up reimplementing the rule interpreter to trigger based on state transitions while keeping the same sort of rules. It was much simpler, and worked ok, but still. Writing it this way led to three levels of indirection: we'd write rules that tried to emulate a state machine that tried to emulate a do-this-then-that-then-that sort of thing. It was ridiculous and impossible to understand. Serializable coroutines were what we were really looking for, and we ended up rewriting the whole thing in Erlang which pretty much does that. We also had memory management issues which we didn't know how to debug in SBCL, so the thing would crash all the time even when it was working.
    - dustingetz 15y ago
      > Goofing around in Erlang FTFY ;) but seriously, why is Erlang any different than Lisp?
      - outworlder 15y ago
        How could someone answer this question? I think you should elaborate on what you are looking for.
- dustingetz 15y ago
  > Goofing around in Lisp when we had no idea how to deploy it wasn't really a smart move for a four-man startup, in retrospect. now THAT is something i want to read about. please blog!
  - pnathan 15y ago
    Agreed, it would be interesting and useful to know about challenges and solutions. It'd also be useful to use that as a springboard & checklist to develop a deployment system for the next four-man startup!
- gnarbarian 15y ago
  A couple of years ago I bought and played your game for a couple months and really liked it. Have you ever considered adding procedural content generation? I have always wanted a space MMO with an infinite number of solar systems to explore. that was the goal for my project a few years ago. I used a few different approaches including genetic algorithms and straight up monte carlo. The GA eventually led me to a sweet spot for constants currently used in the monte carlo generation of planets. It also has an interactive python console. It helps a ton for experimenting and tweaking stuff while everything is running. http://code.google.com/p/stableorbit http://code.google.com/p/stableorbit if anyone is interested in trying it I suggest pulling it from svn. python 2.6 panda3d are the most important dependencies.

a1k0n 15y ago

And really not all that interesting. Weird. Hi, HN. Also, we ended up ditching Lisp. Goofing around in Lisp when we had no idea how to deploy it wasn't really a smart move for a four-man startup, in retrospect.

rcfox 15y ago
> Goofing around in Lisp when we had no idea how to deploy it wasn't really a smart move for a four-man startup, in retrospect. Would you mind expanding on that? I'd be interested in hearing about why you chose Lisp and why it didn't work out.
- a1k0n 15y ago
  Well, we wanted to add some "life" to the universe with some automated trade convoys and so forth, and I had the crazy idea to use a rule-based system to implement it, and found an off-the-shelf clone of CLIPS (the expert system type thing), using the Rete algorithm called Lisa (http://lisa.sourceforge.net/ http://lisa.sourceforge.net/), and that was implemented in Common Lisp. It turned out that the impedance mismatch between that and what I actually wanted to do was gigantic, and Lisa was way too slow as it would re-evaluate the whole world anytime anything changed, so I ended up reimplementing the rule interpreter to trigger based on state transitions while keeping the same sort of rules. It was much simpler, and worked ok, but still. Writing it this way led to three levels of indirection: we'd write rules that tried to emulate a state machine that tried to emulate a do-this-then-that-then-that sort of thing. It was ridiculous and impossible to understand. Serializable coroutines were what we were really looking for, and we ended up rewriting the whole thing in Erlang which pretty much does that. We also had memory management issues which we didn't know how to debug in SBCL, so the thing would crash all the time even when it was working.
  - dustingetz 15y ago
    > Goofing around in Erlang FTFY ;) but seriously, why is Erlang any different than Lisp?
    - outworlder 15y ago
      How could someone answer this question? I think you should elaborate on what you are looking for.
dustingetz 15y ago
> Goofing around in Lisp when we had no idea how to deploy it wasn't really a smart move for a four-man startup, in retrospect. now THAT is something i want to read about. please blog!
- pnathan 15y ago
  Agreed, it would be interesting and useful to know about challenges and solutions. It'd also be useful to use that as a springboard & checklist to develop a deployment system for the next four-man startup!
gnarbarian 15y ago
A couple of years ago I bought and played your game for a couple months and really liked it. Have you ever considered adding procedural content generation? I have always wanted a space MMO with an infinite number of solar systems to explore. that was the goal for my project a few years ago. I used a few different approaches including genetic algorithms and straight up monte carlo. The GA eventually led me to a sweet spot for constants currently used in the monte carlo generation of planets. It also has an interactive python console. It helps a ton for experimenting and tweaking stuff while everything is running. http://code.google.com/p/stableorbit http://code.google.com/p/stableorbit if anyone is interested in trying it I suggest pulling it from svn. python 2.6 panda3d are the most important dependencies.