How to Fail at Pyweek

About this talk

• how to fail at Pyweek
• and have fun along the way

First try

• game concept: beaver building a dam
• fend off evil birds and fishes
• I was relatively unambitious
• created a game called Dambuilder

Libraries used

• pygame
• pyode
• it worked!

Dambuilder

If first you...

• I succeeded the first try
• but
• if first you succeed...
• you can still fail next time!

Second try

• game concept: world simulation
• a bit broad
• but some actual game would materialize soon enough from this, right?

libraries

• pygame again
• with pgu this time
• cage for cellular automata for the simulation
• managed to get simple GUI going with pgu
• ...didn't do much with it

The simulation

• created a simulation which involved..
• domestication of plants
• tech levels
• population
• growth of nation states
• war between states
• didn't have game yet...

Screenshot

Wait

• my simulation is rather complex
• it's too slow!
• cage is in Python
• rewrite cage to C library (unbound)
• (by this time I knew I'd never finish the game)
• Pyweek ends
• I then let the unfinished unbound linger forever in my SVN repository

Followup

• much later
• find the corepy library
• generate assembly from Python!
• be your own JIT!
• opposite of PyPy!

I know!

• I know!
• I can generate a cellular automaton engine that's really fast with it!
• also half-baked lingering in my SVN repository

Try two and a half

• didn't actually do anything
• just signed up and thought about ideas

Third try

• so second try was perhaps slightly overambitious
• this time, we're going to have space marines
• space marines!
• you can't go wrong with space marines, right?

Libraries

• pyglet
• cocos2d
• kytten
• box2d physics

Got working

• GUI with kytten
• that I ended up not using
• cocos2d agent
• that I ended up not using
• define a simple 2d world (of rectangles) with box2d
• render it with pyglet

Along the way

• cocos2d had a release out
• but there was also a newer release
• a very well hidden release
• I ran into a bug and went to look for it
• some issues with the release procedure...

cocos2d release issues

• the new release was called 0.3.1 beta
• but its setup.py said it was 0.3.0
• CHANGELOG mentions a version 0.4.0 with an unknown release date

More release issues

• some __date__ variable set to 2008-06-27 even though release was in 2010
• code.google.com said "mar 08", while the previous releases were mostly in "mar 2008".
• 'mar 2008' means march 2008, while 'mar 08' means march this year
• not uploaded to pypi, and not even listed on some page so that setuptool's web crawler can find it

So

• joined cocos2d mailing list
• added issues to the issue tracker
• cocos2d is actually really cool
• just needed some minor tiny release process tuneups

cocos2d 0.4.0rc

• the 0.4.0rc0 release earlier this month was managed much better
• an upload to pypi would be nice though... (it's still at 0.3.0)

So back to the game

• those space marines need to fight something
• what do space marines fight?
• aliens!
• don't want dumb aliens that walk into the rectangles all the time
• so they need to be able to find their way between those rectangles
• I need path finding!
• that's not so hard when you have a grid
• but with randomly distributed rectangles everywhere it turns out to be more difficult
• path finding meshes?
• ugh, too complicated, don't want!

Steering behaviors!

• found out about this concept called steering behaviors
• Give your aliens a few simple ways to behave, going to goals, avoiding things
• Found paper and library: opensteer
• But library in C++, so wrote some simple steering behavior code in Python
• (hiding in my subversion forever)
• Still, even writing simple steering behaviors is still complicated
• I'm lazy

Idea

• I know!
• I can use genetic programming to evolve steering behaviors
• Breed code that steers the aliens
• that way I don't need to write the code myself!

GP in Python

• looked into 2 or 3 Python libraries for doing genetic programming
• they're all cool
• and they all sucked in various ways I forget
• cool!
• I get to write my own!

Genetic programming

• GP typically done with lisp-style expression trees (think python AST)
• mutation: change some code in the tree randomly
• crossover: take one branch of one tree and swap it for one from another tree
• crossover algorithms with typed information are horribly complicated

stack language

• I know!
• I'll invent my own forth-style stack language instead!

So

• wrote a simple stack language in Python
• evolved my aliens
• ...
• they sucked even worse than before
• (all went to a corner of the map)

Screenshot?

• sorry
• didn't even bother to make a screenshot this time

By this time

• somewhere by this time I realized...
• ... I'd probably not finish this game during that week
• but I found out about push3

push3

• a stack language dedicated to genetic programming
• read various push3 papers
• idea: implement push3 with corepy and make it really fast
• some other time...

Conclusion

• I didn't succeed at creating a game very often
• but had a lot of fun
• created fun code
• (lingering in my SVN forever)
• and learned a lot
• (applicable at least to future pyweeks!)