Matt Zucker
mzucker1@swarthmore.edu
http://www.swarthmore.edu/NatSci/mzucker1/
make zucker
./zucker > image.ppm
./zucker "Hello!!!" > hello.ppm
./zucker faster\! -preview > image2.ppm
# Display the PPM image with your favorite image viewer
The rendered image is centered and is output as a 600 by 220 pixel PPM. The vector font is proportional.
This entry uses no local variables. None! At! All!
This program makes pretty pictures. To see the default output, try running
./zucker > image.ppm
On the author’s 2.5 GHz MacBook Pro, this command takes approximately 1 minute and 18 seconds to run – your runtime may vary. You will need an image viewer capable of displaying Netpbm images in order to display the output. On the Mac, Xee works well; otherwise ImageMagick and/or GraphicsMagick are both capable of displaying Netpbm files, and are available on multiple platforms. (Using Xee is particularly nice because it periodically reloads the image as it is rendered.)
Aside from the default image, you may create your own by running the program as follows:
./zucker "your text" > image.ppm
To mitigate long runtimes, the program can be invoked in a “preview”
mode by appending -preview
to the command line arguments, like so:
./zucker faster\! -preview > image.ppm
If your platform distinguishes between binary and text streams, you
may need to insert some code to reopen standard output in binary mode
at the start of main
. An ASCII platform is assumed.
This program is quite suitable for the inexperienced programmer, due to a number of considerations:
It is fairly short: just 4 preprocessor directives followed by 29 lines of source.
It compiles cleanly as ANSI C under gcc 4.2.1 with very strict compilation flags.
It uses a very small subset of the keywords available in C, namely
char
, float
, int
, typedef
, struct
, for
, and return
.
More difficult concepts such as void
, if
, switch
, and
while
are eschewed for simplicity’s sake.
It also uses a very small number of library functions: just putchar
,
cos
, sin
, atan2
, sqrt
, and pow
.
New programmers often prefer to use global variables when possible. Hence, all variables in this program (aside from function parameters) have global scope.
An important aspect of the computation is graphically illustrated by the source code itself: the geometric construction of a reflection ray at an object’s surface, given the incoming direction and surface normal.
The program is not quite a traditional ray tracer, but a “sphere tracer” which uses distance fields to encode proximity to objects in the scene. Using sphere tracing to compute visibility instead of analytic ray tracing provides two advantages: first, it greatly eases the computation of ray-torus intersection (which would otherwise require solving a quartic polynomial), and second, it makes it possible to compute ambient occlusion (which is used to generate soft shadows). Unlike many traditional ray tracers and sphere tracers, this one uses iteration instead of recursion to compute the effect of reflections. This, unfortunately, resulted in the programmer going insane; however, program size was reduced and readability hindered. Neither the plaintext for the program’s default output nor for the Netpbm header appears in the program source, which hinders readability as well. Also, unlike some previous IOCCC entries which render ASCII text, this one uses a vector (as opposed to raster) font. Although the program does not generate uppercase letters, it handles them in a reasonable manner when they appear in the input.
Even when the program is pretty-printed and preprocessor macros are
expanded, it should be fairly resistant to analysis due to reckless
reuse of global variables, to the inherently opaque nature of the
math, and to its overall density. Global variables are used to store
the results of intermediate computation. As a result, seemingly simple
functions have important side effects. Although some of the global
variables are simply constants, their use is overloaded. For instance,
the variable X
is used both as the number 40 and as the ASCII
character '('
. Maintainability is hindered due to hardcoding
several aspects of the program in multiple locations. As an example,
it is impossible to modify either the image width or height without
changing the source in at least three different places, including
inside the large string literal.
Some of the functions are simple vector algebra operations such as vector construction, dot product, and normalization. However, it was found to be much more compact to define a single operation for multiply-and-add rather than to have separate functions for addition, subtraction, and scaling. Again, compactness led to obfuscation.
This program heavily abuses for
loops. The comma operator is used to
get as much milage as possible out of statements inside of loop
declarations. Between for
loops, the comma operator, and the ternary
operator, it was possible to avoid if
and while
altogether.
Bitwise and arithmetic operators are used instead of logical operators
whenever gcc doesn’t warn about it. Conversely, integer
multiplication, division, and modulus are sometimes used in place of
traditional bitwise operators just to keep the reader on his or her
toes. Automatic casting between characters, integers, and floats is
performed just about everywhere. Fascinatingly, this doesn’t seem to
affect performance too badly. Sometimes when an explicit cast
(i.e. from float to int) is needed, the program instead simply assigns
to a free variable of the desired type instead. Both the array[index]
and index[array]
notations are used interchangably throughout the
program.
Two other programs were written to support development of this program. The first used a randomized search algorithm to deduce the most compact encoding of the vector font instructions into ASCII; the second used a modified form of the Knuth-Plass algorithm to layout the final text on top of a hand-generated ASCII “stencil”. The programs were not submitted as entries for this year’s IOCCC, not because they lack obfuscation, but because neither is written in C (they are in C++ and Python, respectively).
© Copyright 1984-2012,
Leo Broukhis, Simon Cooper, Landon Curt Noll
- All rights reserved |