Most Erratic Behavior: Jens Schweikhardt DFN Network Operation Center Schlartaeckerweg 3 (Home address) D-71384 Weinstadt Germany http://www.shuttle.de/schweikh/ Judges' comments: To build: make schweikh2 Almost every compiler is supposed to be able to accept the code in schweikh2.c; try compiling schweikh2.orig.c to see if your compiler is fully standard-compliant. To use: ./schweikh2 or ./schweikh2 integer_number Run the program with a small argument, like 5 or 10, several times. Are the results different? Why? Some non-gcc compilers that are not fully ANSI standard do not compile this entry correctly. At the time this hint file was written, some gcc and egcs implementations ran into problem when building the original entry: make schweikh2.orig This program often will produce an error of the form: gcc -ansi schweikh2.orig.c -o schweikh2.orig as: Error: cca00NzV.s, line 58: Missing " at end of string " as: Error: cca00NzV.s, line 58: malformed statement *** Error code 1 (bu21) because the line: ??=line 10 ONE(O(1,1,2,6,0,6)) turns into the line: # 9 "01\012" which causes gcc to give to the assembler the following two lines: .stabs "whey.c ",132,0,0,Ltext1 and the lone " after the #.file line results in an assembler syntax error. In some cases one must compile using 'gcc -g': make schweikh2.orig CFLAGS=-g to trigger this error. And with the proper text beyond the \012, it may be possible to have all kinds of fun adding ``inline assembly'' via #line directives. :-) For extra credit: If your compiler has this bug, transform the line in the C program to inject assembly instructions in such a way that the program will compile. If you are able to do the above, try adding assembly instructions so that it will compile, execute and do something interesting. If you succeed in doing this (especially on FreeBSD, Linux or Sparc), email the author for a free pat on the back. Selected author's comments: What this program does ---------------------- My entry is a yarng, which stands for, you guessed it, yet another random number generator, and is pronounced 'yawn' due to its incredible speed. Seriously, it exploits the randomness of quantum mechanical events that build up the hardware as well as some other non-deterministic factors. Here's how it works: 1) an alarm is scheduled to be delivered in a second 2) an infinite loop is entered that just increments a counter 3) on delivery of the alarm signal, the counter's LSB is printed. 4) while some other counter is not zero, go to 1) 5) exit The yarng therefore produces output at a rate of 1 baud. The number of bits may be specified by an argument and defaults either to UINT_MAX (least likely) or UINT_MAX+1 (most likely). Extra credit: why the different values? How long would you need for an empirical test? For example: $ yarng 32 10111001010101110110010101001111 $ yarng 5 10011 Why I think this is obfuscated ------------------------------ -Ever seen a 'do for ... while' loop? -I avoided 'int' like the plague. Instead I used storage class specifiers (register, auto) to get implicit int. The Standard allows implicit int even in casts as you can see in '(void(*)(register))main'. Where implicit int is impossible, like in the 'int main' declaration I have int split across two lines. All indent programs I use insert space at the beginning of the continuation line and thereby introduce a syntax error. -main (ab)used as a signal handler. Shocking. Yeah, this leads to undefined behaviour. I have yet to see a system where it does not work. If it doesn't work for you, I'd appreciate if you mail me about it. Uh-oh, schweikh did it, uttered "Works for me". Don't quote me on that! -Even if you know how the program works, you just can't predict the output. -Chained use of token pasting operators ## with shuffled arguments -So you think you can't assign to __LINE__ and __FILE__? Not so. The #line preprocessor directive does the trick. But saying '#line 10 "01\n"' is a little lame and I came up with '??=line 10 ONE(O(1,1,2,6,0,6))' -__LINE__ is used as the signal number argument to signal(). If indenting or other editing changes the number of lines the program is likely to be killed after one second... Thanks to POSIX for specifying SIGALRM as 14. You're SOL if your system does not conform. Get a real OS. -All strings (or characters) needed are crammed into __FILE__ using octal notation and then extracted using pointer + offset notation. -Use of a # (null) preprocessor directive (hidden as a %: digraph) Delete this line and prepare for a core dump. -Identifiers reused with different capitalization: one, One, ONE, zero, Zero. ONE is particularly nasty because it appears as a macro taking an arg and as a plain identifier in main's parameters. The rule is that whenever ONE is followed by '(' it must be the macro. -A quote from Kernighan and Plauger, "The Elements of Programming Style", 2nd Ed., page 21: "A useful way to decide if some piece of code is clear or not is the ``telephone test.'' If someone could understand your code when read aloud over the telephone, it's clear enough. If not then it needs rewriting." Now try the telephone test for 'one = zero = atoi (ONE[zero-1])' -Don't you love the aesthetics of symmetry: write (1, 1+__FILE__+1, 1) (telephone test this as well; is it 1, One, ONE, one? Maybe won?). Where does the randomness come from? ------------------------------------ Each bit obviously depends on the number of times the loop body is executed between two calls of the signal handler. But what determines this number? + The accuracy of the clock driving the CPU + The accuracy of the clock driving the timer responsible for alarm signals + On multitasking OS's: The total time the process is scheduled to run between successive invocations of the signal handler. How about the randomness of the yarng? To measure these properties I counted the number of bit sequences and compared them to what I would expect from a generator with uniform distribution (all sequences equally likely to occur). It is interesting to note that yarng's maximum deviations are less than or equal to rand()'s. Randomness however tends to degrade under high load averages on my system (Linux). This could be avoided by allowing the process one second of execution time instead of wall clock time (using the SIGVTALRM sent by the virtual timer instead of SIGALRM). I'm not surprised. Yarng in the end depends on fluctuations of the electrical field and the heat distribution in the oscillators, the implementation and state of the scheduling algorithm, the system load, number and temporal distribution of asynchronous events and last but not least the phase of the moon. God _does_ play dice. Type yarng and watch Him roll.