make
./herrmann2
./try.sh
If you use vi(m) you can see a pattern in herrmann2.cup and [herrmann2.ioccc][(herrmann2.ioccc]) by say:
/[pon]\
and
/l
For additional fun in herrmann2.cup try:
/[po]
/q
and in herrmann2.ioccc try:
/n
Be careful when staring at the source code - your eyes might cross, but if you concentrate on your focus you will find illumination!
You probably know those “magic eye” 3D pictures - pictures looking like random noise, but when you look “through” the paper, you either get a three dimensional picture or a headache. With this program, you can create your own 3D pictures in ASCII art. And yes, the program itself is such a 3D picture.
Create a file which describes the desired 3D picture. The file
must be a block of lowercase letters the size of the desired
picture. The letters specify the depth of the corresponding pixel:
a
means close to the paper, z
means far away. The info files
are examples. Be aware that the size of the picture is restricted
to about 9k.
Pipe your picture into the program. The output will be your 3D picture, constructed of random characters (with ASCII code between 32 and 126). For example, type
./herrmann2 < herrmann2.ioccc
./herrmann2 "234 84 045 5 6765 7487 65190 84 656 254 12 43931 818 0 6542 \
341 45 567 76967 7244 606 976567 895 81898 095 68678 1843 4650547 \
565980691 389 04974" < herrmann2.ioccc
Due to the way these 3D pictures work, the left part of the picture won’t look really 3D, so don’t put things too much on the left. The reason is that for each letter in the source file, a character for the right eye is printed. The 3D effect appears when the left eye sees the same character more to the left. This is of course impossible on the left border of the picture.
Don’t use too different depths and don’t use the depths very close to the paper (beginning of the alphabet). This would make it more difficult to see the picture.
When passing an argument, don’t use a too short one; the repeating pattern which arises from the repetition of the argument could interfere with the repeating pattern which creates the 3D effect.
The program is able to handle all common kinds of line breaks: Unix, DOS and Mac. To get an output with a certain kind of line breaks, simply use the same kind of line breaks in the input file.
As the source code is a 3D picture itself, the easiest way to get it (if you lost it or if you don’t have it already) is to use the program itself on the right picture and with the right argument:
./herrmann2 \
'char*d,A[9876];e;b;*ad,a,c; tw,ndr,T; wri; ;*h; _,ar ;on;'\
' ;l ;i(V)man,n {-!har ; =Aadre(0,&e,o||n -- +,o4,=9,l=b=8,'\
'!( te-*Aim)|(0~l),srand (l),,!A,d=,b))&&+((A + te-A(&(*)=+ +95>'\
'e?(*& c_*r=5,r+e-r +_:2-19<-+?|(d==d),!n ?*d| *( (char**)+V+), ('\
' +0,*d-7 ) -r+8)c:7:+++7+! r: and%9- 85! ()-(r+o):(+w,_+ A*(=er'\
'i+(o)+b)),!write,(=_((-b+*h)(1,A+b,!!((((-+, a >T^l,( o-95=+))w?'\
'++ &&r:b<<2+a +w) ((!main(n*n,V) , +-) ),l)),w= +T-->o +o+;{ &!'\
'a;}return _+= ' < herrmann2.ioccc
If I had submitted the program last year, I would have said that it has exactly the maximum allowed length: 1536 characters. Now let’s say it’s a way of protesting against the inflation in program length.
The size of the source 3D picture has been chosen carefully: as big as possible, but small enough to be viewed without problems on normal-sized terminals.
The layout is of a completely new kind. And it describes what the program does more precisely than any comment could.
I don’t think there has ever been an IOCCC program with so severe layout constraints. Once the 3D picture has been fixed, only a sixth of the characters can be chosen freely! (457 of 1840, to be exact.) In contrast, in palindrome programs, for example, half of the characters can be chosen freely.
I did not cheat by using comments or -D
options - something
I’ve seen even some palindrome programs do. In fact, at my first
attempt at writing the program I became convinced that it
isn’t possible without cheating. But then I found out that with
comments or defines, it would become trivial, so I had to try
harder…
If you don’t believe the constraints are hard, try to write just one single line which satisfies them. (To be able to make one function call, for example, you have to fill an entire line with stuff doing mostly nothing.)
Just in case this is not enough: Yes, the code itself is
obfuscated: It’s (almost) only one big expression, with deep
nesting of ()
and ?:
, and full of unnecessary operations which
do not really help making it readable. And it contains a lot of
“magic numbers”. (My favourite is the 7
in line 15 (each
one). Which is the other digit I could have put there instead,
and what other changes to the program would have been necessary?
Without violating the constraints of course.)
The fact that the main loop of the program calls main()
recursively is nothing new. New however is that this program
only needs a logarithmic recursion depth! (Find out how this
works.)
When still in doubt about how to judge this program, type
./herrmann2 < herrmann2.cup
and let the result inspire you. ;-)
I fear that gcc is the only compiler for which this program
works. Due to the severe restrictions, I was forced to use
declarations of int
variables, functions and parameters without
specifying their type. The biggest problem was to put int main(int c, char ** v){
into the program. After some time, I
became convinced that this is impossible. Finally, at least I
managed to squeeze main(c, v){
into the constraints instead.
Also, a lot of warnings are produced - most of them “data definition has no type or storage class”, but also some “parameter names (without types) in function declaration” and “assignment from incompatible pointer type”. I wasn’t able to get rid of them.
At least, it probably works with all sufficiently recent versions of gcc. I tested it with versions 2.7.2.1, 2.8.1, 2.95.2, and ecgs-2.91.66.
Of course I couldn’t put #include
s at the beginning of
the program. Usually, this is not really necessary with
gcc. However, for this program it would have been, because some
functions appear as function pointers, and as such they are not
generated implicitly. So the build script has to pass -include ...
to gcc. A problem with this is that the complete path of
the include files has to be provided. In the build script, I
assume that the path is /usr/include
, but this can easily be
adapted.