Cryptographic hash golf

Question

This contest is over.

Due to the nature of cops-and-robbers challenges, the cops challenge becomes a lot easier when the interest in the associated robbers challenge has diminished. Therefore, while you can still post hash functions, your answer will not get accepted or form part of the leaderboard.

This challenge is a search for the shortest implementation of a hash function that is collision resistant, i.e., it should be infeasible to find two different messages with the same hash.

As a cop, you try to invent and implement a hash function finding the best compromise between code size and collision resistance. Use too many bytes and another cop will outgolf you!

As a robber, you try to foil the cops' attempts by cracking their functions, proving that they are unsuitable. This will force them to use more bytes to strengthen their algorithms!

Cops challenge

Task

Implement a cryptographic hash function H : I -> O of your choice, where I is the set of all non-negative integers below 2²³⁰ and O is the set of all non-negative integers below 2¹²⁸.

You can either implement H as an actual function that accepts and returns a single integer, a string representation of an integer or an array of integers or a full program that reads from STDIN and prints to STDOUT in base 10 or 16.

Scoring

• H that it has to resist the robbers challenge defined below.

  If a robber defeats your submission in the first 168 hours after posting it, it is considered cracked.

• The implementation of H should be as short as possible. The shortest uncracked submission will be the winner of the cops challenge.

Additional rules

• If you implement H as a function, please provide a wrapper to execute the function from within a program that behaves as explained above.

• Please provide at least three test vectors for your program or wrapper (example inputs and their corresponding outputs).

• H can be your novel design (preferred) or a well-known algorithm, as long as you implement it yourself. It is forbidden to use any kind in-built hash function, compression function, cipher, PRNG, etc.

  Any built-in commonly used to implement hashing functions (e.g., base conversion) is fair game.

• The output of your program or function must be deterministic.

• There should be a free (as in beer) compiler/interpreter that can be run on a x86 or x64 platform or from within a web browser.

• Your program or function should be reasonably efficient and has to hash any message in I below 2²¹⁹ in less than a second.

  For edge cases, the (wall) time taken on my machine (Intel Core i7-3770, 16 GiB of RAM) will be decisive.

• Given the nature of this challenge, it is forbidden to change the code of your answer in any way, whether it alters the output or not.

  If your submission has been cracked (or even if it hasn't), you can post an additional answer.

  If your answer is invalid (e.g., it doesn't comply with the I/O specification), please delete it.

Example

Python 2.7, 22 bytes

def H(M):
 return M%17

Wrapper

print H(int(input()))

Robbers challenge

Task

Crack any of the cops' submissions by posting the following in the robbers' thread: two messages M and N in I such that H(M) = H(N) and M ≠ N.

Scoring

• Cracking each cop submissions gains you one point. The robber with the most points wins.

  In the case of a tie, the tied robber that cracked the longest submission wins.

Additional rules

• Every cop submission can only be cracked once.

• If a cop submission relies on implementation-defined or undefined behavior, you only have to find a crack that works (verifiably) on your machine.

• Each crack belongs to a separate answer in the robbers' thread.

• Posting an invalid cracking attempt bans you from cracking that particular submission for 30 minutes.

• You may not crack your own submission.

Example

Python 2.7, 22 bytes by user8675309

1

and

18

Leaderboard

Safe submissions

1. CJam, 21 bytes by eBusiness
2. C++, 148 bytes by tucuxi
3. C++, 233(?) bytes by Vi.

Uncracked submissions

You can use this Stack Snippet to get a list of not yet cracked answers.

function g(p){$.getJSON('//api.stackexchange.com/2.2/questions/51068/answers?page='+p+'&pagesize=100&order=desc&sort=creation&site=codegolf&filter=!.Fjs-H6J36w0DtV5A_ZMzR7bRqt1e',function(s){s.items.map(function(a){var h=$('<div/>').html(a.body).children().first().text();if(!/cracked/i.test(h)&&(typeof a.comments=='undefined'||a.comments.filter(function(b){var c=$('<div/>').html(b.body);return /^cracked/i.test(c.text())||c.find('a').filter(function(){return /cracked/i.test($(this).text())}).length>0}).length==0)){var m=/^\s*((?:[^,(\s]|\s+[^-,(\s])+)\s*(?:[,(]|\s-).*?([0-9]+)/.exec(h);$('<tr/>').append($('<td/>').append($('<a/>').text(m?m[1]:h).attr('href',a.link)),$('<td class="score"/>').text(m?m[2]:'?'),$('<td/>').append($('<a/>').text(a.owner.display_name).attr('href',a.owner.link))).appendTo('#listcontent');}});if(s.length==100)g(p+1);});}g(1);

table th, table td {padding: 5px} th {text-align: left} .score {text-align: right} table a {display:block}

<script src="https://ajax.googleapis.com/ajax/libs/jquery/2.1.1/jquery.min.js"></script><link rel="stylesheet" type="text/css" href="//cdn.sstatic.net/codegolf/all.css?v=83c949450c8b"><table><tr><th>Language</th><th class="score">Length</th><th>User</th></tr><tbody id="listcontent"></tbody></table>

Answer 1

Python, 109 bytes [cracked, and again]

def f(n,h=42,m=2**128):
 while n:h+=n&~-m;n>>=128;h+=h<<10;h^=h>>6;h%=m
 h+=h<<3;h^=h>>11;h+=h<<15;return h%m

I tried implementing Jenkins' one-at-a-time function as-is, with the only difference being the seed and the number of bits.

Fun fact: Apparently Perl used the Jenkins' hash at some point.

Wrapper

print(f(int(input())))

Examples

>>> f(0)
12386682
>>> f(1)
13184902071
>>> f(2**128-1)
132946164914354994014709093274101144634
>>> f(2**128)
13002544814292
>>> f(2**128+1)
13337372262951
>>> f(2**(2**20))
290510273231835581372700072767153076167

Answer 2

CJam, 21 bytes

1q3*{i+_E_#*^26_#)%}/

Takes a string of bytes as input.

In pseudocode:

hash = 1
3 times:
    for i in input:
        hash = hash + i
        hash = hash xor hash * 14^14
        hash = hash mod (26^26 + 1)
output hash

Example hashes:

"" (empty string) -> 1
"Test" -> 2607833638733409808360080023081587841
"test" -> 363640467424586895504738713637444713

It may be a bit on the simple side, the output range is only a little more than 122 bits, the triple iteration strengthening is already a bit broken as it does exactly the same thing every time, so input that hash to 1 in the first iteration will be a full break. But it is short, and there is no fun in being too safe.

Answer 3

C++, 148 bytes

typedef __uint128_t U;U h(char*b,U n,U&o){U a=0x243f6a8885a308d,p=0x100000001b3;for(o=a;n--;)for(U i=27;--i;){o=(o<<i)|(o>>(128-i));o*=p;o^=b[n];}}

__uint128_t is a GCC extension, and works as expected. The hash is based on iterating FNV hash (I've borrowed their prime, although a is the first digits of Pi in hex) with a sha1-like rotation at the start of each iteration. Compiling with -O3, hashing a 10MB file takes under 2 seconds, so there is still margin for upping the iterations in the inner loop - but I'm feeling generous today.

De-uglified (changed variable names, added comments, whitespace and a pair of braces) for your cracking pleasure:

typedef __uint128_t U;
U h(char* input, U inputLength, U &output){
    U a=0x243f6a8885a308d,p=0x100000001b3;    
    for(output=a;inputLength--;) {   // initialize output, consume input
        for(U i=27;--i;) {                          // evil inner loop
            output = (output<<i)|(output>>(128-i)); // variable roll 
            output *= p;                            // FNV hash steps
            output ^= input[inputLength];        
        }
    }
    // computed hash now available in output
}

Golfing suggestions are welcome (even if I don't get to improve the code based on them).

edit: fixed typos in de-uglified code (golfed version remains unchanged).

Answer 4

CJam, 44 bytes [cracked]

lW%600/_z]{JfbDbGK#%GC#[md\]}%z~Bb4G#%\+GC#b

Input is in base 10.

CJam is slow. I hope it runs in 1 second in some computer...

Explanations

lW%600/            e# Reverse, and split into chunks with size 600.
_z                 e# Duplicate and swap the two dimensions.
]{                 e# For both versions or the array:
    JfbDb          e# Sum of S[i][j]*13^i*19^j, where S is the character values,
                   e# and the indices are from right to left, starting at 0.
    GK#%GC#[md\]   e# Get the last 32+48 bits.
}%
z~                 e# Say the results are A, B, C, D, where A and C are 32 bits.
Bb4G#%             e# E = the last 32 bits of A * 11 + C.
\+GC#b             e# Output E, B, D concatenated in binary.

Well, the two dimensional things seemed to be a weakness... It was intended to make some slow calculations faster at the beginning. But it cannot run in a second no matter what I do, so I removed the slow code finally.

It should be also better if I have used binary bits and higher bases.

C version

__uint128_t hash(unsigned char* s){
    __uint128_t a=0,b=0;
    __uint128_t ar=0;
    __uint128_t v[600];
    int l=0,j=strlen(s);
    memset(v,0,sizeof v);
    for(int i=0;i<j;i++){
        if(i%600)
            ar*=19;
        else{
            a=(a+ar)*13;
            ar=0;
        }
        if(i%600>l)
            l=i%600;
        v[i%600]=v[i%600]*19+s[j-i-1];
        ar+=s[j-i-1];
    }
    for(int i=0;i<=l;i++)
        b=b*13+v[i];
    a+=ar;
    return (((a>>48)*11+(b>>48))<<96)
        +((a&0xffffffffffffull)<<48)
        +(b&0xffffffffffffull);
}

Answer 5

C++, 182 characters (+ about 51 characters of boilerplate)

h=0xC0CC3051F486B191;j=0x9A318B5A176B8125;char q=0;for(int i=0;i<l;++i){char w=buf[i];h+=((w<<27)*257);j^=(h+0x5233);h+=0xAA02129953CC12C3*(j>>32);j^=(w+0x134)*(q-0x16C552F34);q=w;}

Boilerplate:

void hash(const unsigned char* buf, size_t len, unsigned long long *hash1, unsigned long long *hash2)
{
    unsigned long long &h=*hash1;
    unsigned long long &j=*hash2;
    size_t l = len;
    const unsigned char* b = buf;

    // code here
}

Runnable program with a golfed function

#include <stdio.h>

// The next line is 227 characters long
int hash(char*b,int l,long long&h,long long&j){h=0xC0CC3051F486B191;j=0x9A318B5A176B8125;char q=0;for(int i=0;i<l;++i){char w=b[i];h+=((w<<27)*257);j^=(h+0x5233);h+=0xAA02129953CC12C3*(j>>32);j^=(w+0x134)*(q-0x16C552F34);q=w;}}

int main() {
    char buf[1024];
    int l  = fread(buf, 1, 1024, stdin);
    long long q, w;
    hash(buf, l, q, w);
    printf("%016llX%016llX\n", q, w);
}

Answer 6

Pyth, 8 Cracked

sv_`.lhQ

Try it online

A bit of a silly answer, I'll explain how it works because most people can't read Pyth. This takes the natural log of one plus the input, and then converts that to a string. That string is reversed, and then evaluated and then converted to an integer.

A python translation would look like:

import math
n = eval(input()) + 1
rev = str(math.log(n))[::-1]
print(int(eval(rev)))

Answer 7

Python 3, 216 bytes [cracked]

def f(m):
 h=1;p=[2]+[n for n in range(2,102)if 2**n%n==2];l=len(bin(m))-2;*b,=map(int,bin((l<<(l+25)//26*26)+m)[2:])
 while b:
  h*=h
  for P in p:
   if b:h=h*P**b.pop()%0xb6ee45a9012d1718f626305a971e6a21
 return h

Due to an incompatibility with the spec I can think of at least one slight vulnerability, but other than that I think this is at least brute-force proof. I've checked the first 10 million hashes, among other things.

In terms of golf this would be shorter in Python 2, but I've sacrificed some bytes for efficiency (since it's probably not going to win anyway).

Edit: This was my attempt at implementing the Very Smooth Hash, but unfortunately 128-bits was far too small.

Wrapper

print(f(int(input())))

Examples

>>> f(0)
2
>>> f(123456789)
228513724611896947508835241717884330242
>>> f(2**(2**19)-1)
186113086034861070379984115740337348649
>>> f(2**(2**19))
1336078

Code explanation

def f(m):
 h=1                                             # Start hash at 1
 p=[2]+[n for n in range(2,102)if 2**n%n==2]     # p = primes from 2 to 101
 l=len(bin(m))-2                                 # l = bit-length of m (input)
 *b,=map(int,bin((l<<(l+25)//26*26)+m)[2:])      # Convert bits to list, padding to
                                                 # a multiple of 26 then adding the
                                                 # bit-length at the front

 while b:                                        # For each round
  h*=h                                           # Square the hash
  for P in p:                                    # For each prime in 2 ... 101
   if b:h=(h*P**b.pop()                          # Multiply by prime^bit, popping
                                                 # the bit from the back of the list
           %0xb6ee45a9012d1718f626305a971e6a21)  # Take mod large number

 return h                                        # Return hash

An example of the padding for f(6):

[1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0]

(len 3)(------------------ 23 zeroes for padding -------------------------)(input 6)
       (---------------------------- length 26 total ------------------------------)

Answer 8

C, 87 bytes [cracked]

This is the complete program; no wrapper required. Accepts binary input via stdin, and outputs a hexadecimal hash to stdout.

c;p;q;main(){while((c=getchar())+1)p=p*'foo+'+q+c,q=q*'bar/'+p;printf("%08x%08x",p,q);}

This only calculates a 64-bit hash, so I'm taking a bit of a gamble here.

In case anyone's wondering, the two constants 'foo+' and 'bar/' are the prime numbers 1718578987 and 1650553391.

---

Examples:

Ignores leading zeroes:

echo -ne '\x00\x00\x00\x00' |./hash
0000000000000000

Single-byte inputs:

echo -ne '\x01' |./hash
0000000100000001
echo -ne '\xff' |./hash
000000ff000000ff

Multi-byte inputs:

echo -ne '\x01\x01' |./hash
666f6f2dc8d0e15c
echo -ne 'Hello, World' |./hash
04f1a7412b17b86c

Answer 9

J - 39 bytes - cracked

Function taking a string as input and returning an integer < 2¹²⁸. I am assuming we have to name our function to be valid, so drop another 3 chars from the count if we can submit anonymous functions.

H=:_8(".p:@+5,9:)a\(a=.(2^128x)&|@^/@)]

For those of you that don't read hieroglyphics, here's a rundown of what I'm doing.

• a=.(2^128x)&|@^/@ This is a subroutine* which takes an array of numbers, and then treats it as a power tower, where exponentiation is taken mod 2¹²⁸. By "power tower", I mean if you gave it the input 3 4 5 6, it would calculate 3 ^ (4 ^ (5 ^ 6)).
• (".p:@+5,9:)a This function takes a string, converts it to the number N, and then calculates the (n+5)-th and (n+9)-th prime numbers, and then throws the a from before on it. That is, we find p(n+5) ^ p(n+9) mod 2¹²⁸ where p(k) is the k-th prime.
• H=:_8...\(a...)] Perform the above function on 8-character subblocks of the input, and then a all the results together and call the resulting hash function H. I use 8 characters because J's "k-th prime" function fails when p(k) > 2³¹, i.e. k=105097564 is the largest safe k.

Have some sample outputs. You can try this yourself online at tryj.tk, but I really recommend doing this at home by downloading the interpreter from Jsoftware.

   H=:_8(".p:@+5,9:)a\(a=.(2^128x)&|@^/@)]
   H '88'
278718804776827770823441490977679256075
   H '0'
201538126434611150798503956371773
   H '1'
139288917338851014461418017489467720433
   H '2'
286827977638262502014244740270529967555
   H '3'
295470173585320512295453937212042446551
   30$'0123456789'  NB. a 30 character string
012345678901234567890123456789
   H 30$'0123456789'
75387099856019963684383893584499026337
   H 80$'0123456789'
268423413606061336240992836334135810465

* Technically, it's not a function in and of itself, it attaches to other functions and acts on their output. But this is a semantic issue of J, not a conceptual difference: the program flow is as I described it above.

Answer 10

Python 3, 118 bytes [cracked]

def H(I):
    o=0;n=3;M=1<<128
    for c in I:i=ord(c);o=(o<<i^o^i^n^0x9bb90058bcf52d3276a7bf07bcb279b7)%M;n=n*n%M
    return o

Indentation is a single tab. Simple hash, haven't really tested it thoroughly yet.

Call as follows:

print(H("123456789"))

result: 73117705077050518159191803746489514685

Answer 11

C++, 239 bytes

My very first code golf! [Please be gentle]

#define r(a,b) ((a<<b)|(a>>(64-b)))
typedef uint64_t I;I f(I*q, I n, I&h){h=0;for(I i=n;--i;)h=r(h^(r(q[i]*0x87c37b91114253d5,31)*0x4cf5ad432745937f),31)*5+0x52dce729;h^=(h>>33)*0xff51afd7ed558ccd;h^=(h>>33)*0xc4ceb9fe1a85ec53;h^=(h>>33);}

Ungolfed version:

I f(I* q, I n, I& h) // input, length and output
{
    h = 0; // initialize hashes
    for (I i=n;--i;)
    {
        q[i] *= 0x87c37b91114253d5;
        q[i]  = rotl(q[i], 31);
        q[i] *= 0x4cf5ad432745937f;

        h ^= q[i]; // merge the block with hash

        h *= rotl(h, 31);
        h = h * 5 + 0x52dce729;
    }
    h ^= h>>33;
    h *= 0xff51afd7ed558ccd;
    h ^= h>>33;
    h *= 0xc4ceb9fe1a85ec53; // avalanche!
    h ^= h>>33;
}

Not the best hash, and definitely not the shortest code in existence. Accepting golfing tips and hoping to improve!

Wrapper

Probably not the best in the world, but a wrapper nonetheless

I input[500];

int main()
{
    string s;
    getline(cin, s);
    memcpy(input, s.c_str(), s.length());
    I output;
    f(input, 500, output);
    cout << hex << output << endl;
}

Answer 12

Java, 299 291 282 bytes, cracked.

import java.math.*;class H{public static void main(String[]a){BigInteger i=new java.util.Scanner(System.in).nextBigInteger();System.out.print(BigInteger.valueOf(i.bitCount()*i.bitLength()+1).add(i.mod(BigInteger.valueOf(Long.MAX_VALUE))).modPow(i,BigInteger.valueOf(2).pow(128)));}}

Does some operations on BigIntegers, then takes the result modulo 2¹²⁸.

Answer 13

C, 128 bytes [cracked]

p;q;r;s;main(c){while((c=getchar())+1)p=p*'foo+'+s^c,q=q*'bar/'+p,r=r*'qux3'^q,s=s*'zipO'+p;printf("%08x%08x%08x%08x",p,q,r,s);}

This is more or less the same algorithm as my last effort (cracked by Vi.), but now has enough hamster wheels to generate proper 128-bit hashes.

The four prime constants in the code are as follows:

'foo+' = 1718578987
'bar/' = 1650553391
'qux3' = 1903523891
'zipO' = 2053730383

As before, this is a complete program with no need for a wrapper. The integer I is input via stdin as raw binary data (big-endian), and the hash O is printed in hex to stdout. Leading zeroes in I are ignored.

Examples:

echo -ne '\x00' |./hash
00000000000000000000000000000000
echo -ne '\x00\x00' |./hash
00000000000000000000000000000000
echo -ne '\x01' |./hash
00000001000000010000000100000001
echo -ne 'A' |./hash
00000041000000410000004100000041
echo -ne '\x01\x01' |./hash
666f6f2dc8d0e15cb9a5996fe0d8df7c
echo -ne 'Hello, World' |./hash
da0ba2857116440a9bee5bb70d58cd6a

Answer 14

C, 122 bytes [cracked]

long long x,y,p;main(c){for(c=9;c|p%97;c=getchar()+1)for(++p;c--;)x=x*'[3QQ'+p,y^=x^=y^=c*x;printf("%016llx%016llx",x,y);}

Nested loops, half-assed LCGs, and variable swapping. What's not to love?

Here's a ungolf'd version to play around with:

long long x,y,p;

int main(int c){
    // Start with a small number of iterations to
    //   get the state hashes good and mixed because initializing takes space
    // Then, until we reach the end of input (EOF+1 == 0)
    //   and a position that's a multiple of 97
    for (c=9;c|p%97;c=getchar()+1) {

        // For each input c(haracter) ASCII value, iterate down to zero
        for (++p;c--;) {

            // x will act like a LCG with a prime multiple
            //   partially affected by the current input position
            // The string '[3QQ' is the prime number 0x5B335151
            x=x*'[3QQ'+p;

            // Mix the result of x with the decrementing character
            y^=c*x;

            // Swap the x and y buffers
            y^=x^=y;
        }
    }

    // Full 128-bit output
    printf("%016llx%016llx",x,y);
    return 0;
}

This is a fully self-contains program that reads from STDIN and prints to STDOUT.

Example:

> echo -n "Hello world" | ./golfhash
b3faef341f70c5ad6eed4c33e1b55ca7

> echo -n "" | ./golfhash
69c761806803f70154a7f816eb3835fb

> echo -n "a" | ./golfhash
5f0e7e5303cfcc5ecb644cddc90547ed

> echo -n "c" | ./golfhash
e64e173ed4415f7dae81aae0137c47e5

In some simple benchmarks, it hashes around 3MB/s of text data. The hash speed depends on the input data itself, so that should probably be taken into consideration.

Answer 15

PHP 4.1, 66 bytes [cracked]

I'm just warming up.

I hope you find this insteresting.

<?for($l=strlen($b.=$a*1);$i<40;$o.=+$b[+$i]^"$a"/$a,$i++);echo$o;

I've tried it numbers as large as 999999999999999999999999999.
The output seemed to be within the 2¹²⁸ range.

---

PHP 4.1 is required because of the register_globals directive.

It works by automatically creating local variables from the session, POST, GET, REQUEST and cookies.

It uses the key a. (E.G.: access over http://localhost/file.php?a=<number>).

If you want to test it with PHP 4.2 and newer, try this:

<?for($l=strlen($b.=$a=$_REQUEST['a']*1);$i<40;$o.=+$b[+$i]^"$a"/$a,$i++);echo$o;

This version only works with POST and GET.

---

Example output:

0 -> 0000000000000000000000000000000000000000
9 -> 8111111111111111111111111111111111111111
9999 -> 8888111111111111111111111111111111111111
1234567890 -> 0325476981111111111111111111111111111111
99999999999999999999999999999999999999999999999999999999999999999999999999999999 -> 0111191111111111111111111111111111111111

(I assure you that there are numbers that produce the same hash).

Answer 16

C, 134 bytes, Cracked

This is complete C program.

long long i=0,a=0,e=1,v,r;main(){for(;i++<323228500;r=(e?(scanf("%c",&v),e=v>'/'&&v<':',v):(a=(a+1)*7)*(7+r)));printf("0x%llx\n", r);}

What it does: The idea is to take input as byte array and append pseudo random (but deterministic) bytes at the end to make the length equal to about 2²³⁰ (a bit more). The implementation reads input byte by byte and starts using pseudo random data when it finds the first character that isn't a digit.

As builtin PRNG isn't allowed I implemented it myself.

There is undefined/implementation defined behavior that makes the code shorter (the final value should be unsigned, and I should use different types for different values). And I couldn't use 128 bit values in C. Less obfuscated version:

long long i = 0, prand = 0, notEndOfInput = 1, in, hash;

main() {
    for (; i++ < 323228500;) {
        if (notEndOfInput) {
            scanf("%c", &in);
            notEndOfInput = in >= '0' && in <= '9';
            hash = in;
        } else {
            prand = (prand + 1)*7;
            hash = prand*(7 + hash);
        }
    }
    printf("0x%llx\n", hash);
}

Answer 17

Python 2.X - 139 bytes [[Cracked]]

This is quite similar to all the other (LOOP,XOR,SHIFT,ADD) hashes out here. Come get your points robbers ;) I'll make a harder one after this one is solved.

M=2**128
def H(I):
 A=[1337,8917,14491,71917];O=M-I%M
 for z in range(73):
  O^=A[z%4]**(9+I%9);O>>=3;O+=9+I**(A[z%4]%A[O%4]);O%=M
 return O

Wrapper (expects one argument in base-16 also known as hexadecimal):

import sys
if __name__ == '__main__':
 print hex(H(long(sys.argv[1], 16)))[2:][:-1].upper()

Answer 18

Python 2.7 - 161 bytes [[Cracked]]

Well since I managed to change my first hash function into an useless version before posting it, I think I will post another version of a similar structure. This time I tested it against trivial collisions and I tested most of the possible input magnitudes for speed.

A=2**128;B=[3,5,7,11,13,17,19]
def H(i):
 o=i/A
 for r in range(9+B[i%7]):
  v=B[i%7];i=(i+o)/2;o=o>>v|o<<128-v;o+=(9+o%6)**B[r%6];o^=i%(B[r%6]*v);o%=A
 return o

Wrapper (not counted in the bytecount)

import sys
if __name__ == '__main__':
 arg = long(sys.argv[1].strip(), 16)
 print hex(H(arg))[2:][:-1].upper()

Run example (input is always a hexadecimal number):

$ python crypt2.py 1
3984F42BC8371703DB8614A78581A167
$ python crypt2.py 10
589F1156882C1EA197597C9BF95B9D78
$ python crypt2.py 100
335920C70837FAF2905657F85CBC6FEA
$ python crypt2.py 1000
B2686CA7CAD9FC323ABF9BD695E8B013
$ python crypt2.py 1000AAAA
8B8959B3DB0906CE440CD44CC62B52DB

Answer 19

Ruby, 90 Bytes

def H(s);i=823542;s.each_byte{|x|i=(i*(x+1)+s.length).to_s.reverse.to_i%(2**128)};i;end

A highly random hash algorithm I made up without looking at any real hashes...no idea if it is good. it takes a string as input.

Wrapper:

def buildString(i)
  if(i>255)
    buildString(i/256)+(i%256).chr
  else
    i.chr
  end
end 
puts H buildString gets

Answer 20

Mathematica, 89 bytes, cracked

Last@CellularAutomaton[(a=Mod)[#^2,256],{#~IntegerDigits~2,0},{#~a~99,128}]~FromDigits~2&

Not the shortest.

Answer 21

PHP, 79 Bytes (cracked. With a comment):

echo (('.'.str_replace('.',M_E*$i,$i/pi()))*substr(pi(),2,$i%20))+deg2rad($i);

This does alot of scary things via type-conversions in php, which makes it hard to predict ;) (or at least I hope so). It's not the shortest or most unreadable answer, however.

To run it you can use PHP4 and register globals (with ?i=123) or use the commandline:

php -r "$i = 123.45; echo (('.'.str_replace('.',M_E*$i,$i/pi()))*substr(pi(),2,$i%20))+deg2rad($i);"

Answer 22

C# - 393 bytes cracked

using System;class P{static void Main(string[]a){int l=a[0].Length;l=l%8==0?l/8:l/8+1;var b=new byte[l][];for(int i=0;i<l;i++){b[i]=new byte[8];};int j=l-1,k=7;for(int i=0;i<a[0].Length;i++){b[j][k]=Convert.ToByte(""+a[0][i],16);k--;if((i+1)%8==0){j--;k=7;}}var c=0xcbf29ce484222325;for(int i=0;i<l;i++){for(int o=0;o<8;o++){c^=b[i][o];c*=0x100000001b3;}}Console.WriteLine(c.ToString("X"));}}

Ungolfed:

using System;
class P
{
    static void Main(string[]a)
    {
      int l = a[0].Length;
      l = l % 8 == 0 ? l / 8 : l / 8 + 1;
      var b = new byte[l][];
      for (int i = 0; i < l; i++) { b[i] = new byte[8]; };
      int j = l-1, k = 7;
      for (int i = 0; i < a[0].Length; i++)
      {
        b[j][k] = Convert.ToByte(""+a[0][i], 16);
        k--;
        if((i+1) % 8 == 0)
        {
          j--;
          k = 7;
        }
      }
      var c = 0xcbf29ce484222325;
      for (int i = 0; i < l; i++)
      {
        for (int o = 0; o < 8; o++)
        {
          c ^= b[i][o];
          c *= 0x100000001b3;
        }
      }
      Console.WriteLine(c.ToString("X"));
    }
}

I have never touched cryptography or hashing in my life, so be gentle :)

It's a simple implementation of an FNV-1a hash with some array pivoting on the input. I am sure there is a better way of doing this but this is the best I could do.

It might use a bit of memory on long inputs.

Answer 23

Python 2, 115 bytes [Cracked already!]

OK, here's my last effort. Only 115 bytes because the final newline isn't required.

h,m,s=1,0,raw_input()
for c in[9+int(s[x:x+197])for x in range(0,len(s),197)]:h+=pow(c,257,99**99+52)
print h%4**64

This is a complete program that inputs a decimal integer on stdin and prints a decimal hash value on stdout. Extra leading zeroes will result in different hash values, so I'll just assume that the input doesn't have any.

This works by stuffing 197-digit chunks of the input number through a modular exponentiation. Unlike some languages, the int() function always defaults to base 10, so int('077') is 77, not 63.

Sample outputs:

$ python hash.py <<<"0"
340076608891873865874583117084537586383

$ python hash.py <<<"1"
113151740989667135385395820806955292270

$ python hash.py <<<"2"
306634563913148482696255393435459032089

$ python hash.py <<<"42"
321865481646913829448911631298776772679

$ time python hash.py <<<`python <<<"print 2**(2**19)"`
233526113491758434358093601138224122227

real    0m0.890s   <-- (Close, but fast enough)
user    0m0.860s
sys     0m0.027s