Files
Landon Curt Noll 4ee4c731f7 convert markdown copyright TAB chars into SPACE chars
While not strictly necessary, replacing the two ASCII tab characters
in the markdown copyright comment makes it easier to detect problematic
TAB indentation within markdown files.

After this commit, only 90 markdown files have leading ASCII TAB
characters.  And again, while not strictly necessary, we can consider
running expand on those 90 markdown files to see what effect that
might have on the HTML files they produce.

Replace leading ASCII TAB characters with multiples of 8 ASCII SPACE
characters (the default action of the `expand(1)` tool) might have
an negative effect on some HTML formatting where **hack-a-round**
formatting was attempted to fix broken indentation caused by
problematic ASCII TAB characters.

Think of this as a baseline for evaluating leading
TAB characters within markdown files.

TODO: Consider either using `expand(1)` to replace leading ASCII
TAB characters with multiples of 8 ASCII SPACE characters AND/OR
modifying the appropriate `bin` tool to convert leading ASCII TAB
characters into multiples of 8 ASCII SPACE characters.
2024-06-22 22:15:25 -07:00

236 lines
7.7 KiB
Markdown

## To build:
``` <!---sh-->
make
```
## To use:
``` <!---sh-->
./herrmann2
```
## Try:
``` <!---sh-->
./try.sh
```
If you use vi(m) you can see a pattern in [herrmann2.cup](%%REPO_URL%%/2001/herrmann2/herrmann2.cup) and
[herrmann2.ioccc](%%REPO_URL%%/2001/herrmann2/herrmann2.ioccc) by say:
```
/[pon]\
```
and
```
/l
```
For additional fun in [herrmann2.cup](%%REPO_URL%%/2001/herrmann2/herrmann2.cup) try:
```
/[po]
/q
```
and in [herrmann2.ioccc](%%REPO_URL%%/2001/herrmann2/herrmann2.ioccc) try:
```
/n
```
## Judges' remarks:
Be careful when staring at the source code - your eyes might cross, but
if you concentrate on your focus you will find illumination!
## Author's remarks:
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.
### How to use it
1. 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.
2. 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
``` <!---sh-->
./herrmann2 < herrmann2.ioccc
```
3. If you want to control the characters used in the output (for
example if your picture is to consist only of digits), pass them
as the first argument. In this case, each time the program needs a
new random character, it will take the next one from the
argument. When it arrives at the end of the argument, it will
start over at the beginning. Example:
``` <!---sh-->
./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
```
### Some hints to get better results
- 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.
### Extra-Features
- 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:
``` <!---sh-->
./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.
### Why this program should win
- 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
``` <!---sh-->
./herrmann2 < herrmann2.cup
```
and let the result inspire you. ;-)
### Requirements to the compiler / Compatibility
- 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.
<!--
Copyright © 1984-2024 by Landon Curt Noll. All Rights Reserved.
You are free to share and adapt this file under the terms of this license:
Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)
For more information, see:
https://creativecommons.org/licenses/by-sa/4.0/
-->