Functional Reactive Python

foremanyellowDéveloppement de logiciels

7 nov. 2013 (il y a 5 années et 4 mois)

99 vue(s)

Functional Reactive Python

On Introducing CS to High School

John Peterson

Western State College

Gunnison, CO


Get high school students excited about
CS with a 1 week summer camp

Bring them to Gunnison in the

Use video gaming as the bait

Lots of outdoor activity

Integrate Math / Physics into the


Teach basic principles

avoid putting
too much “in a can” or just doing drag &
drop / menu pulling (Not Alice)

Avoid big chunks of code

no time to
teach software engineering

Lots of small projects rather than one
big one

Don’t hide from basic math / physics

Use freely available software

Declarative Language Research

High school students are the best
audience I’ve ever had for exploring new
programming styles

No pre
conceived notions about how to

Eager to learn

Will tell you what they think

The Game Engine

After considering a few possibilities, we
settled on the Panda3D game engine.

Good support from CMU

Python is a good instructional language

All the basic features we needed were


We also chose to cover 3
D modeling a bit. We
used Blender but this had some problems
interacting with Panda:

Texture loss (we never figured this out)

Lack of support for joints / animations in the
files exported to Panda

Poorly adapted to the Windows environment

In spite of this, many students spent a lot of
time building Blender models.

Student Work

While building these 3
D models isn’t
programming, it did help us get
accustomed to working in 3 space and
allowed students to personalize their 3

Software Issues

Our goal was not to teach any specific
language or software system

We wanted every program to be as
declarative as possible: elegant and

FRP (Functional Reactive Programming)
was used to hide time flow from the

Plan A

Write a custom front end (in Haskell of
course) to allow domain
semantics and notations, provide
time type safety, and give
appropriate error messages.

Probably the right thing to do

Not enough time!

Doesn’t solve the debugging problem

Need to build custom IDE

Plan B

Attempt to recreate FRP using Python

Able to use existing IDE (Pydev in
Eclipse), good debugger

Python has lambda (just uglified lisp
after all)

Students are likely to run into Python

Much less work!!

Programming Style

Create factory functions for all game objects
(models, lights, gui controls, camera)

return an object “handle”

Use python keywords and good defaulting to
avoid complexity in the factory

“Moving parts” are defined by reactive signals

Signals can be defined separately to allow
circular reference

Use signals like “time” to get motion

Functional Reactive Programming

This is work I did back in the Yale (yuck!)

The big idea is to hide time flow

event handles, explicit updating.

A “signal” (reactive value) is something
that seems to mutate on its own as time
progresses. Time flow becomes implicit.

Example: Moving Bear

from Panda import *


p = pandaBear()

p.position = P3(sin(time), 3, cos(time))

world.cameraPos = P3(0,


Teaching points: sin / cos, coordinate system, trajectories

More Positioning

p1 = pandaBear()

p2 = pandaBear()

def f(t):

return P3(sin(t), 3, cos(t))

p1.position = f(time)

p2.position = f(time + pi)

p1.HPR = P3(time*2,0,0)

p2.scale = 0.2 * (5 + sin(2*time))

world.cameraPos = P3(0,

Adding Controls

Panda3d has lots of user interface devices. We
wrapped these up in FRP style to make them
easy to use.

h = hangglider()

Text("Use slider to adjust heading")

s = Slider(max = 2*pi)


h.position = P3(sin(time), 3, cos(time))

h.HPR = P3(s.value, 0, 0)

world.cameraPos = P3(0,

Teaching Math

A great thing about 3
D gaming is that
you can’t hide from math! Students
needed to pick this up fast.

We covered linear interpolation, basic
trig, paths, periodic functions, and
vector arithmetic (Thanks to Andy Keck
for being a great math teacher!)

We didn’t shy away from calculus

an example using derivatives.

Point Where You Go

Text("Plane circles while pointing forward")

p = boeing707()

b = P3C(3, time, sin(time*3))

p.position = b

db = deriv(P3(0,0,0), b)

hpr = P3toHPR(db) # Explained in lecture

p.HPR = P3(getX(hpr)

world.cameraPos = P3(0,

Teaching Programming

We also wanted to teach basic ideas in

Function definition was easy

We used for loops which arranged objects
in various patterns to talk about
conditional behavior, nested loops,
arrays, and random numbers

Exploding Balls

mycolors = [red, blue, green, yellow, purple, brown, white, black]

bagOfBalls = []

endPos = []

world.cameraPos = P3(0,


explodeSpeed = 3 #.1 is very slow, 10 is very fast

for p in range(0,getRand(50,88)):

theColor = mycolors[getRand(0,7)]

if getRand(0,1) == 1:

bagOfBalls.append(volleyBall(color = theColor))


bagOfBalls.append(soccerBall(color = theColor))

5,5), getRand(
5,5), getRand(

startPos = P3(0,0,0)

for index in range(0, len(endPos)):

d = endPos[index]

bagOfBalls[index].position = startPos + time*explodeSpeed*d

Teaching Physics

Finally, we were able to get to some simple physics.
Here’s a bouncing ball:

def launch(p0, v0):

a = P3(0,0,

v = v0 + integral(a)

p = p0 + integral(v)

return react(p, when(getZ(p) < 0, bounce, p, v))

def bounce(p0, v0):

return launch(P3(getX(p0), getY(p0),

0.9*P3(getX(v0), getY(v0),


And the Roller Coaster

def f((t0, p0, v0), t1):

deltaT = t1


pe = scaleP3(deltaT, v0) + p0

p1 = P3(getX(pe), getY(pe), coasterPath(getX(pe)))

dir = p1


oldv = absP3(v0)

deltaz = getZ(p1)


s1 = oldv*oldv

2 * g * deltaz

if s1 < 0:

newv =



newv = sqrt(s1) * 0.999

v1 = scaleP3(newv, normP3(dir))

return ((t1, p1, v1), p1)

Roller Coaster Notes

We covered the basic physics to make
this understandable

Students can explore changes to the
physical laws

Needed to “crack open” the underlying
signal extrapolation function. That was
tough since you see the signal state and
delta t. But you can also use this to
explain how integral and derivitave work

Panda Effects

Effects are fun for the students even
though they don’t contribute to the
core subject material.

We did lighting, particle effects
(explosions, sparkles, flames, ..), and

Check out our disco

Character Animation

Panda has a number of built
in jointed
characters. These are easy to control

they have additional signals to set joint

What we didn’t do was introduce a way to
create animations easily. No easy way
to save a pose or sequence poses.



I haven’t showed you any games. No
paddleball or asteroids or shooting
gallery or space invaders or doom.

Why? Because we didn’t get them
working yet. There was plenty to do
without gaming but next time this needs
to be fixed.

There were FRP problems, notational
problems, and time problems.

Game Engine Issues

Any game engine has a huge API

it’s a lot of
work to build on such a big base

Need to “normalize” existing models. We
wanted to use a common size / orientation so
students could swap models easily.

No 3
D design environment for positioning
models or creating animations.

Many high level abstractions are hard to
encapsulate. We avoided many of these
(collisions, tasks, animations).

Things the Worked

Gaming as a way to unify math, cs, and physics

Wide appeal

gamers and non gamers both
had fun

Lots of assistants

kept students from
getting frustrated (thanks to Kate, Darek,
and Stephan)

FRP made programs small and easily

Great source of projects for CS students

Summer employment for CS students

Things That Didn’t Work Well

Blender (too complex, importing

FRP “in the large”

not sure how to do
this yet


not really a great language for
programmers. Notational issues
kept coming up (no user
definable infix
operators, couldn’t get field selection to
lift, overloading problems, …)

For Next Year

More gaming!

Better publicity

it’s hard to get students in the

Better software

too much time in the debugger,
poor notations, no type checking

Better preparation

more examples, better lesson
planning (especially for math / physics integration
into the CS stuff)

Music videos

this is something that we definitely
need to get to that leads nicely up to games. We
used Haskore to teach music but didn’t get the sound
to synch correctly with the action

Programming Language Issues

Making time flow implicit is a big win.

Laziness is crucial. We spent way to much time
having to re
invent lazy evaluation. Forward
reference was a big problem

Python’s O
O system kept getting in the way

Python lacks the syntactic flexibility that we
really needed

Good type systems make things a LOT easier.
We did some “load time” type checking but
didn’t get full H
M. Students didn’t
understand the debugger.

Disciplinary Education

CS instruction suffers from tunnel vision

we don’t do a
good job with students that are not going all the way
into CS.

Integrating the math and physics into our curriculum
broadens the appeal and sends a message to future
CS students

The creative side of CS is often hidden in CS1

game engine gives us a chance to work in a more
creative environment

We were able to cover a wide range of math

not just
one specific topic. The visual nature of the system
helps a lot.

We hope to sell this beyond just CS instruction


Game engines are an appealing way to get students in
the door to CS departments

The broad range of topics helped keep all of the
students on board

D modeling was a nice contrast to programming

it let
students blow off steam when programming got hard

CS needs to better serve students in other disciplines

We need to get students interested in CS earlier

The 2 girls at the camp enjoyed it as much as the hard
core gamers.

Scenes from Camp

The Last Conclusion

A good recreational program will make
students forget the frustration of
software development.