Author

Topic: [ANN] Nxt :: descendant of Bitcoin - page 105. (Read 383989 times)

legendary
Activity: 2142
Merit: 1010
Newbie
November 24, 2013, 01:57:00 AM
Thanks. So, the account numbers are derived from sha256 digests, but they don't contain any embedded checksums themselves? In other words, any number between 1 and Long.MAX_VALUE (i.e., 9223372036854775807) could potentially be an account number? This is not very safe, because when entering a transaction a single typo in the account number will result in the money being permanently lost (or, in the best case, going to some random person's account).

Also, isn't the space of possible account numbers too small, thus increasing the chance of collisions? When compared for example to the number of possible bitcoin addresses, which are at least 27 characters long (and also contain a checksum)?

https://bitcointalksearch.org/topic/m.3393070
legendary
Activity: 2142
Merit: 1010
Newbie
newbie
Activity: 48
Merit: 0
November 24, 2013, 12:54:20 AM
Hello to all,

I had an idea for unclaimed Nxt and/or fauceting.

Distribute to many international volunteer based open-source/free culture/etc groups of small size. This would create interest in a dedicated community that understands the long-term goals of Nxt and not speculators. Each group is embedded in many different projects and countries and so will better be able to promote adoption and novel uses outside of narrow internet space.

I can contribute to organizing this if you like by providing a long list of international organizations.

I have been following this project from its beginning but due to my socio-economic status have not been able to invest. I am happy it is looking good for those who could afford to buy in though Smiley.

Regards,

nYun

thank you nYun for the idea. If you want to take part in promoting NXT we're going to open a fund to reimburce productive people like you. You'll be able to get the NXT you didn't originally buy.
newbie
Activity: 2
Merit: 0
November 23, 2013, 10:00:37 PM
Hello to all,

I had an idea for unclaimed Nxt and/or fauceting.

Distribute to many international volunteer based open-source/free culture/etc groups of small size. This would create interest in a dedicated community that understands the long-term goals of Nxt and not speculators. Each group is embedded in many different projects and countries and so will better be able to promote adoption and novel uses outside of narrow internet space.

I can contribute to organizing this if you like by providing a long list of international organizations.

I have been following this project from its beginning but due to my socio-economic status have not been able to invest. I am happy it is looking good for those who could afford to buy in though Smiley.

Regards,

nYun
sr. member
Activity: 392
Merit: 250
November 23, 2013, 08:07:45 PM
It is not exactly what i wanted. I need something like btc vanitygen - to get account id number with some pattern e.g begining with 55555....

Here is a simple brute-force modification of vanity.java to do what you wanted. Sorry, I have no place to upload a compiled binary, you would have to compile the java file yourself. I only modified the main method.

Code:
import java.math.BigInteger;
import java.security.MessageDigest;
import java.util.concurrent.ThreadLocalRandom;
import java.util.concurrent.atomic.AtomicLong;

public class vanity {

    /* key size */
    public static final int KEY_SIZE = 32;

    /* 0 */
    public static final byte[] ZERO = {
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
    };

    /* the prime 2^255-19 */
    public static final byte[] PRIME = {
        (byte)237, (byte)255, (byte)255, (byte)255,
        (byte)255, (byte)255, (byte)255, (byte)255,
        (byte)255, (byte)255, (byte)255, (byte)255,
        (byte)255, (byte)255, (byte)255, (byte)255,
        (byte)255, (byte)255, (byte)255, (byte)255,
        (byte)255, (byte)255, (byte)255, (byte)255,
        (byte)255, (byte)255, (byte)255, (byte)255,
        (byte)255, (byte)255, (byte)255, (byte)127
    };

    /* group order (a prime near 2^252+2^124) */
    public static final byte[] ORDER = {
        (byte)237, (byte)211, (byte)245, (byte)92,
        (byte)26,  (byte)99,  (byte)18,  (byte)88,
        (byte)214, (byte)156, (byte)247, (byte)162,
        (byte)222, (byte)249, (byte)222, (byte)20,
        (byte)0,   (byte)0,   (byte)0,   (byte)0,
        (byte)0,   (byte)0,   (byte)0,   (byte)0,
        (byte)0,   (byte)0,   (byte)0,   (byte)0,
        (byte)0,   (byte)0,   (byte)0,   (byte)16
    };

    /********* KEY AGREEMENT *********/

    /* Private key clamping
     *   k [out] your private key for key agreement
     *   k  [in]  32 random bytes
     */
    public static final void clamp(byte[] k) {
        k[31] &= 0x7F;
        k[31] |= 0x40;
        k[ 0] &= 0xF8;
    }

    /* Key-pair generation
     *   P  [out] your public key
     *   s  [out] your private key for signing
     *   k  [out] your private key for key agreement
     *   k  [in]  32 random bytes
     * s may be NULL if you don't care
     *
     * WARNING: if s is not NULL, this function has data-dependent timing */
    public static final void keygen(byte[] P, byte[] s, byte[] k) {
        clamp(k);
        core(P, s, k, null);
    }

    /* Key agreement
     *   Z  [out] shared secret (needs hashing before use)
     *   k  [in]  your private key for key agreement
     *   P  [in]  peer's public key
     */
    public static final void curve(byte[] Z, byte[] k, byte[] P) {
        core(Z, null, k, P);
    }

    /********* DIGITAL SIGNATURES *********/

    /* deterministic EC-KCDSA
     *
     *    s is the private key for signing
     *    P is the corresponding public key
     *    Z is the context data (signer public key or certificate, etc)
     *
     * signing:
     *
     *    m = hash(Z, message)
     *    x = hash(m, s)
     *    keygen25519(Y, NULL, x);
     *    r = hash(Y);
     *    h = m XOR r
     *    sign25519(v, h, x, s);
     *
     *    output (v,r) as the signature
     *
     * verification:
     *
     *    m = hash(Z, message);
     *    h = m XOR r
     *    verify25519(Y, v, h, P)
     *
     *    confirm  r == hash(Y)
     *
     * It would seem to me that it would be simpler to have the signer directly do
     * h = hash(m, Y) and send that to the recipient instead of r, who can verify
     * the signature by checking h == hash(m, Y).  If there are any problems with
     * such a scheme, please let me know.
     *
     * Also, EC-KCDSA (like most DS algorithms) picks x random, which is a waste of
     * perfectly good entropy, but does allow Y to be calculated in advance of (or
     * parallel to) hashing the message.
     */

    /* Signature generation primitive, calculates (x-h)s mod q
     *   v  [out] signature value
     *   h  [in]  signature hash (of message, signature pub key, and context data)
     *   x  [in]  signature private key
     *   s  [in]  private key for signing
     * returns true on success, false on failure (use different x or h)
     */
    public static final boolean sign(byte[] v, byte[] h, byte[] x, byte[] s) {
        /* v = (x - h) s  mod q  */
        byte[] tmp1=new byte[65];
        byte[] tmp2=new byte[33];
        int w;
        int i;
        for (i = 0; i < 32; i++)
            v[i] = 0;
        i = mula_small(v, x, 0, h, 32, -1);
        mula_small(v, v, 0, ORDER, 32, (15-v[31])/16);
        mula32(tmp1, v, s, 32, 1);
        divmod(tmp2, tmp1, 64, ORDER, 32);
        for (w = 0, i = 0; i < 32; i++)
            w |= v[i] = tmp1[i];
        return w != 0;
    }

    /* Signature verification primitive, calculates Y = vP + hG
     *   Y  [out] signature public key
     *   v  [in]  signature value
     *   h  [in]  signature hash
     *   P  [in]  public key
     */
    public static final void verify(byte[] Y, byte[] v, byte[] h, byte[] P) {
        /* Y = v abs(P) + h G  */
        byte[] d=new byte[32];
        long10[]
            p=new long10[]{new long10(),new long10()},
            s=new long10[]{new long10(),new long10()},
            yx=new long10[]{new long10(),new long10(),new long10()},
            yz=new long10[]{new long10(),new long10(),new long10()},
            t1=new long10[]{new long10(),new long10(),new long10()},
            t2=new long10[]{new long10(),new long10(),new long10()};

        int vi = 0, hi = 0, di = 0, nvh=0, i, j, k;

        /* set p[0] to G and p[1] to P  */

        set(p[0], 9);
        unpack(p[1], P);

        /* set s[0] to P+G and s[1] to P-G  */

        /* s[0] = (Py^2 + Gy^2 - 2 Py Gy)/(Px - Gx)^2 - Px - Gx - 486662  */
        /* s[1] = (Py^2 + Gy^2 + 2 Py Gy)/(Px - Gx)^2 - Px - Gx - 486662  */

        x_to_y2(t1[0], t2[0], p[1]); /* t2[0] = Py^2  */
        sqrt(t1[0], t2[0]); /* t1[0] = Py or -Py  */
        j = is_negative(t1[0]); /*      ... check which  */
        t2[0]._0 += 39420360; /* t2[0] = Py^2 + Gy^2  */
        mul(t2[1], BASE_2Y, t1[0]);/* t2[1] = 2 Py Gy or -2 Py Gy  */
        sub(t1[j], t2[0], t2[1]); /* t1[0] = Py^2 + Gy^2 - 2 Py Gy  */
        add(t1[1-j], t2[0], t2[1]);/* t1[1] = Py^2 + Gy^2 + 2 Py Gy  */
        cpy(t2[0], p[1]); /* t2[0] = Px  */
        t2[0]._0 -= 9; /* t2[0] = Px - Gx  */
        sqr(t2[1], t2[0]); /* t2[1] = (Px - Gx)^2  */
        recip(t2[0], t2[1], 0); /* t2[0] = 1/(Px - Gx)^2  */
        mul(s[0], t1[0], t2[0]); /* s[0] = t1[0]/(Px - Gx)^2  */
        sub(s[0], s[0], p[1]); /* s[0] = t1[0]/(Px - Gx)^2 - Px  */
        s[0]._0 -= 9 + 486662; /* s[0] = X(P+G)  */
        mul(s[1], t1[1], t2[0]); /* s[1] = t1[1]/(Px - Gx)^2  */
        sub(s[1], s[1], p[1]); /* s[1] = t1[1]/(Px - Gx)^2 - Px  */
        s[1]._0 -= 9 + 486662; /* s[1] = X(P-G)  */
        mul_small(s[0], s[0], 1); /* reduce s[0] */
        mul_small(s[1], s[1], 1); /* reduce s[1] */


        /* prepare the chain  */
        for (i = 0; i < 32; i++) {
            vi = (vi >> 8) ^ (v[i] & 0xFF) ^ ((v[i] & 0xFF) << 1);
            hi = (hi >> 8) ^ (h[i] & 0xFF) ^ ((h[i] & 0xFF) << 1);
            nvh = ~(vi ^ hi);
            di = (nvh & (di & 0x80) >> 7) ^ vi;
            di ^= nvh & (di & 0x01) << 1;
            di ^= nvh & (di & 0x02) << 1;
            di ^= nvh & (di & 0x04) << 1;
            di ^= nvh & (di & 0x08) << 1;
            di ^= nvh & (di & 0x10) << 1;
            di ^= nvh & (di & 0x20) << 1;
            di ^= nvh & (di & 0x40) << 1;
            d[i] = (byte)di;
        }

        di = ((nvh & (di & 0x80) << 1) ^ vi) >> 8;

        /* initialize state */
        set(yx[0], 1);
        cpy(yx[1], p[di]);
        cpy(yx[2], s[0]);
        set(yz[0], 0);
        set(yz[1], 1);
        set(yz[2], 1);

        /* y[0] is (even)P + (even)G
         * y[1] is (even)P + (odd)G  if current d-bit is 0
         * y[1] is (odd)P + (even)G  if current d-bit is 1
         * y[2] is (odd)P + (odd)G
         */

        vi = 0;
        hi = 0;

        /* and go for it! */
        for (i = 32; i--!=0; ) {
            vi = (vi << 8) | (v[i] & 0xFF);
            hi = (hi << 8) | (h[i] & 0xFF);
            di = (di << 8) | (d[i] & 0xFF);

            for (j = 8; j--!=0; ) {
                mont_prep(t1[0], t2[0], yx[0], yz[0]);
                mont_prep(t1[1], t2[1], yx[1], yz[1]);
                mont_prep(t1[2], t2[2], yx[2], yz[2]);

                k = ((vi ^ vi >> 1) >> j & 1)
                  + ((hi ^ hi >> 1) >> j & 1);
                mont_dbl(yx[2], yz[2], t1[k], t2[k], yx[0], yz[0]);

                k = (di >> j & 2) ^ ((di >> j & 1) << 1);
                mont_add(t1[1], t2[1], t1[k], t2[k], yx[1], yz[1],
                        p[di >> j & 1]);

                mont_add(t1[2], t2[2], t1[0], t2[0], yx[2], yz[2],
                        s[((vi ^ hi) >> j & 2) >> 1]);
            }
        }

        k = (vi & 1) + (hi & 1);
        recip(t1[0], yz[k], 0);
        mul(t1[1], yx[k], t1[0]);

        pack(t1[1], Y);
    }

    ///////////////////////////////////////////////////////////////////////////

    /* sahn0:
     * Using this class instead of long[10] to avoid bounds checks. */
    private static final class long10 {
        public long10() {}
        public long10(
            long _0, long _1, long _2, long _3, long _4,
            long _5, long _6, long _7, long _8, long _9)
        {
            this._0=_0; this._1=_1; this._2=_2;
            this._3=_3; this._4=_4; this._5=_5;
            this._6=_6; this._7=_7; this._8=_8;
            this._9=_9;
        }
        public long _0,_1,_2,_3,_4,_5,_6,_7,_8,_9;
    }

    /********************* radix 2^8 math *********************/

    private static final void cpy32(byte[] d, byte[] s) {
        int i;
        for (i = 0; i < 32; i++)
            d[i] = s[i];
    }

    /* p[m..n+m-1] = q[m..n+m-1] + z * x */
    /* n is the size of x */
    /* n+m is the size of p and q */
    private static final int mula_small(byte[] p,byte[] q,int m,byte[] x,int n,int z) {
        int v=0;
        for (int i=0;i            v+=(q[i+m] & 0xFF)+z*(x[i] & 0xFF);
            p[i+m]=(byte)v;
            v>>=8;
        }
        return v;
    }

    /* p += x * y * z  where z is a small integer
     * x is size 32, y is size t, p is size 32+t
     * y is allowed to overlap with p+32 if you don't care about the upper half  */
    private static final int mula32(byte[] p, byte[] x, byte[] y, int t, int z) {
        final int n = 31;
        int w = 0;
        int i = 0;
        for (; i < t; i++) {
            int zy = z * (y[i] & 0xFF);
            w += mula_small(p, p, i, x, n, zy) +
                (p[i+n] & 0xFF) + zy * (x[n] & 0xFF);
            p[i+n] = (byte)w;
            w >>= 8;
        }
        p[i+n] = (byte)(w + (p[i+n] & 0xFF));
        return w >> 8;
    }

    /* divide r (size n) by d (size t), returning quotient q and remainder r
     * quotient is size n-t+1, remainder is size t
     * requires t > 0 && d[t-1] != 0
     * requires that r[-1] and d[-1] are valid memory locations
     * q may overlap with r+t */
    private static final void divmod(byte[] q, byte[] r, int n, byte[] d, int t) {
        int rn = 0;
        int dt = ((d[t-1] & 0xFF) << 8);
        if (t>1) {
            dt |= (d[t-2] & 0xFF);
        }
        while (n-- >= t) {
            int z = (rn << 16) | ((r[n] & 0xFF) << 8);
            if (n>0) {
                z |= (r[n-1] & 0xFF);
            }
            z/=dt;
            rn += mula_small(r,r, n-t+1, d, t, -z);
            q[n-t+1] = (byte)((z + rn) & 0xFF); /* rn is 0 or -1 (underflow) */
            mula_small(r,r, n-t+1, d, t, -rn);
            rn = (r[n] & 0xFF);
            r[n] = 0;
        }
        r[t-1] = (byte)rn;
    }

    private static final int numsize(byte[] x,int n) {
        while (n--!=0 && x[n]==0)
            ;
        return n+1;
    }

    /* Returns x if a contains the gcd, y if b.
     * Also, the returned buffer contains the inverse of a mod b,
     * as 32-byte signed.
     * x and y must have 64 bytes space for temporary use.
     * requires that a[-1] and b[-1] are valid memory locations  */
    private static final byte[] egcd32(byte[] x,byte[] y,byte[] a,byte[] b) {
        int an, bn = 32, qn, i;
        for (i = 0; i < 32; i++)
            x[i] = y[i] = 0;
        x[0] = 1;
        an = numsize(a, 32);
        if (an==0)
            return y; /* division by zero */
        byte[] temp=new byte[32];
        while (true) {
            qn = bn - an + 1;
            divmod(temp, b, bn, a, an);
            bn = numsize(b, bn);
            if (bn==0)
                return x;
            mula32(y, x, temp, qn, -1);

            qn = an - bn + 1;
            divmod(temp, a, an, b, bn);
            an = numsize(a, an);
            if (an==0)
                return y;
            mula32(x, y, temp, qn, -1);
        }
    }

    /********************* radix 2^25.5 GF(2^255-19) math *********************/

    private static final int P25=33554431; /* (1 << 25) - 1 */
    private static final int P26=67108863; /* (1 << 26) - 1 */

    /* Convert to internal format from little-endian byte format */
    private static final void unpack(long10 x,byte[] m) {
        x._0 = ((m[0] & 0xFF))         | ((m[1] & 0xFF))<<8 |
                (m[2] & 0xFF)<<16      | ((m[3] & 0xFF)& 3)<<24;
        x._1 = ((m[3] & 0xFF)&~ 3)>>2  | (m[4] & 0xFF)<<6 |
                (m[5] & 0xFF)<<14 | ((m[6] & 0xFF)& 7)<<22;
        x._2 = ((m[6] & 0xFF)&~ 7)>>3  | (m[7] & 0xFF)<<5 |
                (m[8] & 0xFF)<<13 | ((m[9] & 0xFF)&31)<<21;
        x._3 = ((m[9] & 0xFF)&~31)>>5  | (m[10] & 0xFF)<<3 |
                (m[11] & 0xFF)<<11 | ((m[12] & 0xFF)&63)<<19;
        x._4 = ((m[12] & 0xFF)&~63)>>6 | (m[13] & 0xFF)<<2 |
                (m[14] & 0xFF)<<10 |  (m[15] & 0xFF)    <<18;
        x._5 =  (m[16] & 0xFF)         | (m[17] & 0xFF)<<8 |
                (m[18] & 0xFF)<<16 | ((m[19] & 0xFF)& 1)<<24;
        x._6 = ((m[19] & 0xFF)&~ 1)>>1 | (m[20] & 0xFF)<<7 |
                (m[21] & 0xFF)<<15 | ((m[22] & 0xFF)& 7)<<23;
        x._7 = ((m[22] & 0xFF)&~ 7)>>3 | (m[23] & 0xFF)<<5 |
                (m[24] & 0xFF)<<13 | ((m[25] & 0xFF)&15)<<21;
        x._8 = ((m[25] & 0xFF)&~15)>>4 | (m[26] & 0xFF)<<4 |
                (m[27] & 0xFF)<<12 | ((m[28] & 0xFF)&63)<<20;
        x._9 = ((m[28] & 0xFF)&~63)>>6 | (m[29] & 0xFF)<<2 |
                (m[30] & 0xFF)<<10 |  (m[31] & 0xFF)    <<18;
    }

    /* Check if reduced-form input >= 2^255-19 */
    private static final boolean is_overflow(long10 x) {
        return (
            ((x._0 > P26-19)) &&
            ((x._1 & x._3 & x._5 & x._7 & x._9) == P25) &&
            ((x._2 & x._4 & x._6 & x._8) == P26)
            ) || (x._9 > P25);
    }

    /* Convert from internal format to little-endian byte format.  The
     * number must be in a reduced form which is output by the following ops:
     *     unpack, mul, sqr
     *     set --  if input in range 0 .. P25
     * If you're unsure if the number is reduced, first multiply it by 1.  */
    private static final void pack(long10 x,byte[] m) {
        int ld = 0, ud = 0;
        long t;
        ld = (is_overflow(x)?1:0) - ((x._9 < 0)?1:0);
        ud = ld * -(P25+1);
        ld *= 19;
        t = ld + x._0 + (x._1 << 26);
        m[ 0] = (byte)t;
        m[ 1] = (byte)(t >> 8);
        m[ 2] = (byte)(t >> 16);
        m[ 3] = (byte)(t >> 24);
        t = (t >> 32) + (x._2 << 19);
        m[ 4] = (byte)t;
        m[ 5] = (byte)(t >> 8);
        m[ 6] = (byte)(t >> 16);
        m[ 7] = (byte)(t >> 24);
        t = (t >> 32) + (x._3 << 13);
        m[ 8] = (byte)t;
        m[ 9] = (byte)(t >> 8);
        m[10] = (byte)(t >> 16);
        m[11] = (byte)(t >> 24);
        t = (t >> 32) + (x._4 <<  6);
        m[12] = (byte)t;
        m[13] = (byte)(t >> 8);
        m[14] = (byte)(t >> 16);
        m[15] = (byte)(t >> 24);
        t = (t >> 32) + x._5 + (x._6 << 25);
        m[16] = (byte)t;
        m[17] = (byte)(t >> 8);
        m[18] = (byte)(t >> 16);
        m[19] = (byte)(t >> 24);
        t = (t >> 32) + (x._7 << 19);
        m[20] = (byte)t;
        m[21] = (byte)(t >> 8);
        m[22] = (byte)(t >> 16);
        m[23] = (byte)(t >> 24);
        t = (t >> 32) + (x._8 << 12);
        m[24] = (byte)t;
        m[25] = (byte)(t >> 8);
        m[26] = (byte)(t >> 16);
        m[27] = (byte)(t >> 24);
        t = (t >> 32) + ((x._9 + ud) << 6);
        m[28] = (byte)t;
        m[29] = (byte)(t >> 8);
        m[30] = (byte)(t >> 16);
        m[31] = (byte)(t >> 24);
    }

    /* Copy a number */
    private static final void cpy(long10 out, long10 in) {
        out._0=in._0; out._1=in._1;
        out._2=in._2; out._3=in._3;
        out._4=in._4; out._5=in._5;
        out._6=in._6; out._7=in._7;
        out._8=in._8; out._9=in._9;
    }

    /* Set a number to value, which must be in range -185861411 .. 185861411 */
    private static final void set(long10 out, int in) {
        out._0=in; out._1=0;
        out._2=0; out._3=0;
        out._4=0; out._5=0;
        out._6=0; out._7=0;
        out._8=0; out._9=0;
    }

    /* Add/subtract two numbers.  The inputs must be in reduced form, and the
     * output isn't, so to do another addition or subtraction on the output,
     * first multiply it by one to reduce it. */
    private static final void add(long10 xy, long10 x, long10 y) {
        xy._0 = x._0 + y._0; xy._1 = x._1 + y._1;
        xy._2 = x._2 + y._2; xy._3 = x._3 + y._3;
        xy._4 = x._4 + y._4; xy._5 = x._5 + y._5;
        xy._6 = x._6 + y._6; xy._7 = x._7 + y._7;
        xy._8 = x._8 + y._8; xy._9 = x._9 + y._9;
    }
    private static final void sub(long10 xy, long10 x, long10 y) {
        xy._0 = x._0 - y._0; xy._1 = x._1 - y._1;
        xy._2 = x._2 - y._2; xy._3 = x._3 - y._3;
        xy._4 = x._4 - y._4; xy._5 = x._5 - y._5;
        xy._6 = x._6 - y._6; xy._7 = x._7 - y._7;
        xy._8 = x._8 - y._8; xy._9 = x._9 - y._9;
    }

    /* Multiply a number by a small integer in range -185861411 .. 185861411.
     * The output is in reduced form, the input x need not be.  x and xy may point
     * to the same buffer. */
    private static final long10 mul_small(long10 xy, long10 x, long y) {
        long t;
        t = (x._8*y);
        xy._8 = (t & ((1 << 26) - 1));
        t = (t >> 26) + (x._9*y);
        xy._9 = (t & ((1 << 25) - 1));
        t = 19 * (t >> 25) + (x._0*y);
        xy._0 = (t & ((1 << 26) - 1));
        t = (t >> 26) + (x._1*y);
        xy._1 = (t & ((1 << 25) - 1));
        t = (t >> 25) + (x._2*y);
        xy._2 = (t & ((1 << 26) - 1));
        t = (t >> 26) + (x._3*y);
        xy._3 = (t & ((1 << 25) - 1));
        t = (t >> 25) + (x._4*y);
        xy._4 = (t & ((1 << 26) - 1));
        t = (t >> 26) + (x._5*y);
        xy._5 = (t & ((1 << 25) - 1));
        t = (t >> 25) + (x._6*y);
        xy._6 = (t & ((1 << 26) - 1));
        t = (t >> 26) + (x._7*y);
        xy._7 = (t & ((1 << 25) - 1));
        t = (t >> 25) + xy._8;
        xy._8 = (t & ((1 << 26) - 1));
        xy._9 += (t >> 26);
        return xy;
    }

    /* Multiply two numbers.  The output is in reduced form, the inputs need not
     * be. */
    private static final long10 mul(long10 xy, long10 x, long10 y) {
        /* sahn0:
         * Using local variables to avoid class access.
         * This seem to improve performance a bit...
         */
        long
            x_0=x._0,x_1=x._1,x_2=x._2,x_3=x._3,x_4=x._4,
            x_5=x._5,x_6=x._6,x_7=x._7,x_8=x._8,x_9=x._9;
        long
            y_0=y._0,y_1=y._1,y_2=y._2,y_3=y._3,y_4=y._4,
            y_5=y._5,y_6=y._6,y_7=y._7,y_8=y._8,y_9=y._9;
        long t;
        t = (x_0*y_8) + (x_2*y_6) + (x_4*y_4) + (x_6*y_2) +
            (x_8*y_0) + 2 * ((x_1*y_7) + (x_3*y_5) +
                    (x_5*y_3) + (x_7*y_1)) + 38 *
            (x_9*y_9);
        xy._8 = (t & ((1 << 26) - 1));
        t = (t >> 26) + (x_0*y_9) + (x_1*y_8) + (x_2*y_7) +
            (x_3*y_6) + (x_4*y_5) + (x_5*y_4) +
            (x_6*y_3) + (x_7*y_2) + (x_8*y_1) +
            (x_9*y_0);
        xy._9 = (t & ((1 << 25) - 1));
        t = (x_0*y_0) + 19 * ((t >> 25) + (x_2*y_8) + (x_4*y_6)
                + (x_6*y_4) + (x_8*y_2)) + 38 *
            ((x_1*y_9) + (x_3*y_7) + (x_5*y_5) +
             (x_7*y_3) + (x_9*y_1));
        xy._0 = (t & ((1 << 26) - 1));
        t = (t >> 26) + (x_0*y_1) + (x_1*y_0) + 19 * ((x_2*y_9)
                + (x_3*y_8) + (x_4*y_7) + (x_5*y_6) +
                (x_6*y_5) + (x_7*y_4) + (x_8*y_3) +
                (x_9*y_2));
        xy._1 = (t & ((1 << 25) - 1));
        t = (t >> 25) + (x_0*y_2) + (x_2*y_0) + 19 * ((x_4*y_8)
                + (x_6*y_6) + (x_8*y_4)) + 2 * (x_1*y_1)
                + 38 * ((x_3*y_9) + (x_5*y_7) +
                        (x_7*y_5) + (x_9*y_3));
        xy._2 = (t & ((1 << 26) - 1));
        t = (t >> 26) + (x_0*y_3) + (x_1*y_2) + (x_2*y_1) +
            (x_3*y_0) + 19 * ((x_4*y_9) + (x_5*y_8) +
                    (x_6*y_7) + (x_7*y_6) +
                    (x_8*y_5) + (x_9*y_4));
        xy._3 = (t & ((1 << 25) - 1));
        t = (t >> 25) + (x_0*y_4) + (x_2*y_2) + (x_4*y_0) + 19 *
            ((x_6*y_8) + (x_8*y_6)) + 2 * ((x_1*y_3) +
                                 (x_3*y_1)) + 38 *
            ((x_5*y_9) + (x_7*y_7) + (x_9*y_5));
        xy._4 = (t & ((1 << 26) - 1));
        t = (t >> 26) + (x_0*y_5) + (x_1*y_4) + (x_2*y_3) +
            (x_3*y_2) + (x_4*y_1) + (x_5*y_0) + 19 *
            ((x_6*y_9) + (x_7*y_8) + (x_8*y_7) +
             (x_9*y_6));
        xy._5 = (t & ((1 << 25) - 1));
        t = (t >> 25) + (x_0*y_6) + (x_2*y_4) + (x_4*y_2) +
            (x_6*y_0) + 19 * (x_8*y_8) + 2 * ((x_1*y_5) +
                    (x_3*y_3) + (x_5*y_1)) + 38 *
            ((x_7*y_9) + (x_9*y_7));
        xy._6 = (t & ((1 << 26) - 1));
        t = (t >> 26) + (x_0*y_7) + (x_1*y_6) + (x_2*y_5) +
            (x_3*y_4) + (x_4*y_3) + (x_5*y_2) +
            (x_6*y_1) + (x_7*y_0) + 19 * ((x_8*y_9) +
                    (x_9*y_8));
        xy._7 = (t & ((1 << 25) - 1));
        t = (t >> 25) + xy._8;
        xy._8 = (t & ((1 << 26) - 1));
        xy._9 += (t >> 26);
        return xy;
    }

    /* Square a number.  Optimization of  mul25519(x2, x, x)  */
    private static final long10 sqr(long10 x2, long10 x) {
        long
            x_0=x._0,x_1=x._1,x_2=x._2,x_3=x._3,x_4=x._4,
            x_5=x._5,x_6=x._6,x_7=x._7,x_8=x._8,x_9=x._9;
        long t;
        t = (x_4*x_4) + 2 * ((x_0*x_8) + (x_2*x_6)) + 38 *
            (x_9*x_9) + 4 * ((x_1*x_7) + (x_3*x_5));
        x2._8 = (t & ((1 << 26) - 1));
        t = (t >> 26) + 2 * ((x_0*x_9) + (x_1*x_8) + (x_2*x_7) +
                (x_3*x_6) + (x_4*x_5));
        x2._9 = (t & ((1 << 25) - 1));
        t = 19 * (t >> 25) + (x_0*x_0) + 38 * ((x_2*x_8) +
                (x_4*x_6) + (x_5*x_5)) + 76 * ((x_1*x_9)
                + (x_3*x_7));
        x2._0 = (t & ((1 << 26) - 1));
        t = (t >> 26) + 2 * (x_0*x_1) + 38 * ((x_2*x_9) +
                (x_3*x_8) + (x_4*x_7) + (x_5*x_6));
        x2._1 = (t & ((1 << 25) - 1));
        t = (t >> 25) + 19 * (x_6*x_6) + 2 * ((x_0*x_2) +
                (x_1*x_1)) + 38 * (x_4*x_8) + 76 *
                ((x_3*x_9) + (x_5*x_7));
        x2._2 = (t & ((1 << 26) - 1));
        t = (t >> 26) + 2 * ((x_0*x_3) + (x_1*x_2)) + 38 *
            ((x_4*x_9) + (x_5*x_8) + (x_6*x_7));
        x2._3 = (t & ((1 << 25) - 1));
        t = (t >> 25) + (x_2*x_2) + 2 * (x_0*x_4) + 38 *
            ((x_6*x_8) + (x_7*x_7)) + 4 * (x_1*x_3) + 76 *
            (x_5*x_9);
        x2._4 = (t & ((1 << 26) - 1));
        t = (t >> 26) + 2 * ((x_0*x_5) + (x_1*x_4) + (x_2*x_3))
            + 38 * ((x_6*x_9) + (x_7*x_8));
        x2._5 = (t & ((1 << 25) - 1));
        t = (t >> 25) + 19 * (x_8*x_8) + 2 * ((x_0*x_6) +
                (x_2*x_4) + (x_3*x_3)) + 4 * (x_1*x_5) +
                76 * (x_7*x_9);
        x2._6 = (t & ((1 << 26) - 1));
        t = (t >> 26) + 2 * ((x_0*x_7) + (x_1*x_6) + (x_2*x_5) +
                (x_3*x_4)) + 38 * (x_8*x_9);
        x2._7 = (t & ((1 << 25) - 1));
        t = (t >> 25) + x2._8;
        x2._8 = (t & ((1 << 26) - 1));
        x2._9 += (t >> 26);
        return x2;
    }

    /* Calculates a reciprocal.  The output is in reduced form, the inputs need not
     * be.  Simply calculates  y = x^(p-2)  so it's not too fast. */
    /* When sqrtassist is true, it instead calculates y = x^((p-5)/8) */
    private static final void recip(long10 y, long10 x, int sqrtassist) {
        long10
            t0=new long10(),
            t1=new long10(),
            t2=new long10(),
            t3=new long10(),
            t4=new long10();
        int i;
        /* the chain for x^(2^255-21) is straight from djb's implementation */
        sqr(t1, x); /*  2 == 2 * 1 */
        sqr(t2, t1); /*  4 == 2 * 2 */
        sqr(t0, t2); /*  8 == 2 * 4 */
        mul(t2, t0, x); /*  9 == 8 + 1 */
        mul(t0, t2, t1); /* 11 == 9 + 2 */
        sqr(t1, t0); /* 22 == 2 * 11 */
        mul(t3, t1, t2); /* 31 == 22 + 9
                    == 2^5   - 2^0 */
        sqr(t1, t3); /* 2^6   - 2^1 */
        sqr(t2, t1); /* 2^7   - 2^2 */
        sqr(t1, t2); /* 2^8   - 2^3 */
        sqr(t2, t1); /* 2^9   - 2^4 */
        sqr(t1, t2); /* 2^10  - 2^5 */
        mul(t2, t1, t3); /* 2^10  - 2^0 */
        sqr(t1, t2); /* 2^11  - 2^1 */
        sqr(t3, t1); /* 2^12  - 2^2 */
        for (i = 1; i < 5; i++) {
            sqr(t1, t3);
            sqr(t3, t1);
        } /* t3 */ /* 2^20  - 2^10 */
        mul(t1, t3, t2); /* 2^20  - 2^0 */
        sqr(t3, t1); /* 2^21  - 2^1 */
        sqr(t4, t3); /* 2^22  - 2^2 */
        for (i = 1; i < 10; i++) {
            sqr(t3, t4);
            sqr(t4, t3);
        } /* t4 */ /* 2^40  - 2^20 */
        mul(t3, t4, t1); /* 2^40  - 2^0 */
        for (i = 0; i < 5; i++) {
            sqr(t1, t3);
            sqr(t3, t1);
        } /* t3 */ /* 2^50  - 2^10 */
        mul(t1, t3, t2); /* 2^50  - 2^0 */
        sqr(t2, t1); /* 2^51  - 2^1 */
        sqr(t3, t2); /* 2^52  - 2^2 */
        for (i = 1; i < 25; i++) {
            sqr(t2, t3);
            sqr(t3, t2);
        } /* t3 */ /* 2^100 - 2^50 */
        mul(t2, t3, t1); /* 2^100 - 2^0 */
        sqr(t3, t2); /* 2^101 - 2^1 */
        sqr(t4, t3); /* 2^102 - 2^2 */
        for (i = 1; i < 50; i++) {
            sqr(t3, t4);
            sqr(t4, t3);
        } /* t4 */ /* 2^200 - 2^100 */
        mul(t3, t4, t2); /* 2^200 - 2^0 */
        for (i = 0; i < 25; i++) {
            sqr(t4, t3);
            sqr(t3, t4);
        } /* t3 */ /* 2^250 - 2^50 */
        mul(t2, t3, t1); /* 2^250 - 2^0 */
        sqr(t1, t2); /* 2^251 - 2^1 */
        sqr(t2, t1); /* 2^252 - 2^2 */
        if (sqrtassist!=0) {
            mul(y, x, t2); /* 2^252 - 3 */
        } else {
            sqr(t1, t2); /* 2^253 - 2^3 */
            sqr(t2, t1); /* 2^254 - 2^4 */
            sqr(t1, t2); /* 2^255 - 2^5 */
            mul(y, t1, t0); /* 2^255 - 21 */
        }
    }

    /* checks if x is "negative", requires reduced input */
    private static final int is_negative(long10 x) {
        return (int)(((is_overflow(x) || (x._9 < 0))?1:0) ^ (x._0 & 1));
    }

    /* a square root */
    private static final void sqrt(long10 x, long10 u) {
        long10 v=new long10(), t1=new long10(), t2=new long10();
        add(t1, u, u); /* t1 = 2u */
        recip(v, t1, 1); /* v = (2u)^((p-5)/8) */
        sqr(x, v); /* x = v^2 */
        mul(t2, t1, x); /* t2 = 2uv^2 */
        t2._0--; /* t2 = 2uv^2-1 */
        mul(t1, v, t2); /* t1 = v(2uv^2-1) */
        mul(x, u, t1); /* x = uv(2uv^2-1) */
    }

    /********************* Elliptic curve *********************/

    /* y^2 = x^3 + 486662 x^2 + x  over GF(2^255-19) */

    /* t1 = ax + az
     * t2 = ax - az  */
    private static final void mont_prep(long10 t1, long10 t2, long10 ax, long10 az) {
        add(t1, ax, az);
        sub(t2, ax, az);
    }

    /* A = P + Q   where
     *  X(A) = ax/az
     *  X(P) = (t1+t2)/(t1-t2)
     *  X(Q) = (t3+t4)/(t3-t4)
     *  X(P-Q) = dx
     * clobbers t1 and t2, preserves t3 and t4  */
    private static final void mont_add(long10 t1, long10 t2, long10 t3, long10 t4,long10 ax, long10 az, long10 dx) {
        mul(ax, t2, t3);
        mul(az, t1, t4);
        add(t1, ax, az);
        sub(t2, ax, az);
        sqr(ax, t1);
        sqr(t1, t2);
        mul(az, t1, dx);
    }

    /* B = 2 * Q   where
     *  X(B) = bx/bz
     *  X(Q) = (t3+t4)/(t3-t4)
     * clobbers t1 and t2, preserves t3 and t4  */
    private static final void mont_dbl(long10 t1, long10 t2, long10 t3, long10 t4,long10 bx, long10 bz) {
        sqr(t1, t3);
        sqr(t2, t4);
        mul(bx, t1, t2);
        sub(t2, t1, t2);
        mul_small(bz, t2, 121665);
        add(t1, t1, bz);
        mul(bz, t1, t2);
    }

    /* Y^2 = X^3 + 486662 X^2 + X
     * t is a temporary  */
    private static final void x_to_y2(long10 t, long10 y2, long10 x) {
        sqr(t, x);
        mul_small(y2, x, 486662);
        add(t, t, y2);
        t._0++;
        mul(y2, t, x);
    }

    /* P = kG   and  s = sign(P)/k  */
    private static final void core(byte[] Px, byte[] s, byte[] k, byte[] Gx) {
        long10
            dx=new long10(),
            t1=new long10(),
            t2=new long10(),
            t3=new long10(),
            t4=new long10();
        long10[]
            x=new long10[]{new long10(),new long10()},
            z=new long10[]{new long10(),new long10()};
        int i, j;

        /* unpack the base */
        if (Gx!=null)
            unpack(dx, Gx);
        else
            set(dx, 9);

        /* 0G = point-at-infinity */
        set(x[0], 1);
        set(z[0], 0);

        /* 1G = G */
        cpy(x[1], dx);
        set(z[1], 1);

        for (i = 32; i--!=0; ) {
            if (i==0) {
                i=0;
            }
            for (j = 8; j--!=0; ) {
                /* swap arguments depending on bit */
                int bit1 = (k[i] & 0xFF) >> j & 1;
                int bit0 = ~(k[i] & 0xFF) >> j & 1;
                long10 ax = x[bit0];
                long10 az = z[bit0];
                long10 bx = x[bit1];
                long10 bz = z[bit1];

                /* a' = a + b */
                /* b' = 2 b */
                mont_prep(t1, t2, ax, az);
                mont_prep(t3, t4, bx, bz);
                mont_add(t1, t2, t3, t4, ax, az, dx);
                mont_dbl(t1, t2, t3, t4, bx, bz);
            }
        }

        recip(t1, z[0], 0);
        mul(dx, x[0], t1);
        pack(dx, Px);

        /* calculate s such that s abs(P) = G  .. assumes G is std base point */
        if (s!=null) {
            x_to_y2(t2, t1, dx); /* t1 = Py^2  */
            recip(t3, z[1], 0); /* where Q=P+G ... */
            mul(t2, x[1], t3); /* t2 = Qx  */
            add(t2, t2, dx); /* t2 = Qx + Px  */
            t2._0 += 9 + 486662; /* t2 = Qx + Px + Gx + 486662  */
            dx._0 -= 9; /* dx = Px - Gx  */
            sqr(t3, dx); /* t3 = (Px - Gx)^2  */
            mul(dx, t2, t3); /* dx = t2 (Px - Gx)^2  */
            sub(dx, dx, t1); /* dx = t2 (Px - Gx)^2 - Py^2  */
            dx._0 -= 39420360; /* dx = t2 (Px - Gx)^2 - Py^2 - Gy^2  */
            mul(t1, dx, BASE_R2Y); /* t1 = -Py  */
            if (is_negative(t1)!=0) /* sign is 1, so just copy  */
                cpy32(s, k);
            else /* sign is -1, so negate  */
                mula_small(s, ORDER_TIMES_8, 0, k, 32, -1);

            /* reduce s mod q
             * (is this needed?  do it just in case, it's fast anyway) */
            //divmod((dstptr) t1, s, 32, order25519, 32);

            /* take reciprocal of s mod q */
            byte[] temp1=new byte[32];
            byte[] temp2=new byte[64];
            byte[] temp3=new byte[64];
            cpy32(temp1, ORDER);
            cpy32(s, egcd32(temp2, temp3, s, temp1));
            if ((s[31] & 0x80)!=0)
                mula_small(s, s, 0, ORDER, 32, 1);
        }
    }

    /* smallest multiple of the order that's >= 2^255 */
    private static final byte[] ORDER_TIMES_8 = {
        (byte)104, (byte)159, (byte)174, (byte)231,
        (byte)210, (byte)24,  (byte)147, (byte)192,
        (byte)178, (byte)230, (byte)188, (byte)23,
        (byte)245, (byte)206, (byte)247, (byte)166,
        (byte)0,   (byte)0,   (byte)0,   (byte)0,
        (byte)0,   (byte)0,   (byte)0,   (byte)0,
        (byte)0,   (byte)0,   (byte)0,   (byte)0,
        (byte)0,   (byte)0,   (byte)0,   (byte)128
    };

    /* constants 2Gy and 1/(2Gy) */
    private static final long10 BASE_2Y = new long10(
        39999547, 18689728, 59995525, 1648697, 57546132,
        24010086, 19059592, 5425144, 63499247, 16420658
    );
    private static final long10 BASE_R2Y = new long10(
        5744, 8160848, 4790893, 13779497, 35730846,
        12541209, 49101323, 30047407, 40071253, 6226132
    );

    private static volatile long min = Long.MAX_VALUE;

    public static void main(final String[] args) {

        if (args.length == 0) {
            System.out.println("Usage: java -cp . vanity SECRET_PREFIX [ACCT_PREFIX]");
        } else {
            int threadCount = Runtime.getRuntime().availableProcessors();
            Thread[] threads = new Thread[threadCount];
            for (int i = 0; i < threadCount; i++) {
                Thread thread = new Thread(new Runnable() {
                    @Override
                    public void run() {
                        try {
                            while (true) {
                                int nonce = ThreadLocalRandom.current().nextInt(Integer.MAX_VALUE) + 1;
                                byte[] publicKey = new byte[32];
                                keygen(publicKey, null, MessageDigest.getInstance("SHA-256").digest((args[0] + nonce).getBytes("UTF-8")));
                                byte[] publicKeyHash = MessageDigest.getInstance("SHA-256").digest(publicKey);
                                BigInteger bigInteger = new BigInteger(1, new byte[] {publicKeyHash[7], publicKeyHash[6], publicKeyHash[5], publicKeyHash[4], publicKeyHash[3], publicKeyHash[2], publicKeyHash[1], publicKeyHash[0]});
                                long t = bigInteger.longValue();
                                if (t > 0 && t < min && (args.length == 1 || String.valueOf(t).startsWith(args[1]))) {
                                    System.out.println(args[0] + nonce + " gives " + t + " account");
                                    min = t;
                                }
                            }
                        } catch (final Exception e) {
                            System.out.println(e.toString());
                        }
                    }
                });
                thread.setDaemon(true);
                thread.start();
                threads[i] = thread;
            }
            for (int i = 0; i < threadCount; i++) {
                try {
                    threads[i].join();
                } catch (InterruptedException e) {
                    Thread.currentThread().interrupt();
                }
            }

        }

    }

}

So if you want an account starting with 666666, and having a secret phrase starting with "nxt rules ", running:
Code:
java vanity "nxt rules " 666666
will give you result such as:
Code:
nxt rules 1623669388 gives 6666666733107667231 account
(and keep trying, to get a shorter account number).

Edit: Updated to use all available processors.
full member
Activity: 129
Merit: 100
November 23, 2013, 07:44:29 PM
25 unclaimed accounts left. We can stuck for a long time. Any ideas what to do?

I sent two payments with different secret phrases, so two of those accounts are mine.

The site - http://88.198.210.245:7875 - looks down at the moment, and I can't install the NXT client until I get home from work (any tips on getting java to work through a restrictive firewall? what port is NXT using?).

I'll install in the next 24 hours and claim both my accounts ASAP.

Nxt's P2P port is 7874 and html interface port is 7875.  You need to open up port 7874 to allow Nxt to connect to peers.  You can leave port 7875 closed and simply allow connections from localhost.  If you open up 7875 it allows other people to use your Nxt client's html interface.

Thanks for that!! Smiley
sr. member
Activity: 644
Merit: 250
November 23, 2013, 07:32:33 PM
25 unclaimed accounts left. We can stuck for a long time. Any ideas what to do?

I sent two payments with different secret phrases, so two of those accounts are mine.

The site - http://88.198.210.245:7875 - looks down at the moment, and I can't install the NXT client until I get home from work (any tips on getting java to work through a restrictive firewall? what port is NXT using?).

I'll install in the next 24 hours and claim both my accounts ASAP.

Nxt's P2P port is 7874 and html interface port is 7875.  You need to open up port 7874 to allow Nxt to connect to peers.  You can leave port 7875 closed and simply allow connections from localhost.  If you open up 7875 it allows other people to use your Nxt client's html interface.
sr. member
Activity: 392
Merit: 250
November 23, 2013, 07:30:52 PM

Thanks. So, the account numbers are derived from sha256 digests, but they don't contain any embedded checksums themselves? In other words, any number between 1 and Long.MAX_VALUE (i.e., 9223372036854775807) could potentially be an account number? This is not very safe, because when entering a transaction a single typo in the account number will result in the money being permanently lost (or, in the best case, going to some random person's account).

Also, isn't the space of possible account numbers too small, thus increasing the chance of collisions? When compared for example to the number of possible bitcoin addresses, which are at least 27 characters long (and also contain a checksum)?
sr. member
Activity: 252
Merit: 250
November 23, 2013, 06:52:26 PM

I noticed that 256 Mb is not enough. Nxt crashes with "Not enough memory" often.

Mine (running on VPS) hangs often; but memory usage never gets anywhere near to 256 Mb. When this happens, webpage does not load.
On other occasions, page loads normally, but the Unlock account function does not work. After entering pass phrase and clicking unlock account, nothing happens.

Everything works again after a reboot.
full member
Activity: 129
Merit: 100
November 23, 2013, 06:13:10 PM
25 unclaimed accounts left. We can stuck for a long time. Any ideas what to do?

I sent two payments with different secret phrases, so two of those accounts are mine.

The site - http://88.198.210.245:7875 - looks down at the moment, and I can't install the NXT client until I get home from work (any tips on getting java to work through a restrictive firewall? what port is NXT using?).

I'll install in the next 24 hours and claim both my accounts ASAP.
legendary
Activity: 1498
Merit: 1000
November 23, 2013, 06:00:03 PM
http://nxtgenesisblock.appspot.com/ ran out of free quotas, we have to wait when a new period begins. GAE resets quotas every 24 hours, so we won't wait longer than that. If u can't claim coins before u go to sleep then PM me with the key and ur account. I'll do it for u later.
Not sure if you are referring to me but I have claimed my coins with no problem...
legendary
Activity: 1498
Merit: 1000
November 23, 2013, 05:58:54 PM
This is the one not compatible with Snow Leopard!

Then u can't run Nxt. U have an option to buy VPS and install Linux.
VPS is an option I will go regardless of setting it up locally Wink

Very busy last days, I am checking now which VPS to choose..

I noticed that 256 Mb is not enough. Nxt crashes with "Not enough memory" often.
Yeah I saw your post upthread! Thanks a lot man..
legendary
Activity: 2142
Merit: 1010
Newbie
November 23, 2013, 05:58:32 PM
http://nxtgenesisblock.appspot.com/ ran out of free quotas, we have to wait when a new period begins. GAE resets quotas every 24 hours, so we won't wait longer than that. If u can't claim coins before u go to sleep then PM me with the key and ur account. I'll do it for u later.
legendary
Activity: 2142
Merit: 1010
Newbie
November 23, 2013, 05:43:17 PM
This is the one not compatible with Snow Leopard!

Then u can't run Nxt. U have an option to buy VPS and install Linux.
VPS is an option I will go regardless of setting it up locally Wink

Very busy last days, I am checking now which VPS to choose..

I noticed that 256 Mb is not enough. Nxt crashes with "Not enough memory" often.
legendary
Activity: 1498
Merit: 1000
November 23, 2013, 05:41:54 PM
This is the one not compatible with Snow Leopard!

Then u can't run Nxt. U have an option to buy VPS and install Linux.
VPS is an option I will go regardless of setting it up locally Wink

Very busy last days, I am checking now which VPS to choose..
legendary
Activity: 2142
Merit: 1010
Newbie
November 23, 2013, 05:38:50 PM
This is the one not compatible with Snow Leopard!

Then u can't run Nxt. U have an option to buy VPS and install Linux.
legendary
Activity: 2142
Merit: 1010
Newbie
November 23, 2013, 05:37:27 PM
The error I get for the latest JRE for 10.6.8

Exception in thread "main" java.lang.NoClassDefFoundError: Nxt
Caused by: java.lang.ClassNotFoundException: Nxt
   at java.net.URLClassLoader$1.run(URLClassLoader.java:202)
   at java.security.AccessController.doPrivileged(Native Method)
   at java.net.URLClassLoader.findClass(URLClassLoader.java:190)
   at java.lang.ClassLoader.loadClass(ClassLoader.java:306)
   at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:301)
   at java.lang.ClassLoader.loadClass(ClassLoader.java:247)


Is there Nxt.zip in the current directory?
legendary
Activity: 1498
Merit: 1000
November 23, 2013, 05:37:06 PM
Will there be a Mac client at launch?  

The client is cross-platformed. It works everywhere, I could launch it on my fridge if it had more memory.
JRE version not compatible with OSX 10.6.8!
You think an older one would work?

Any Java 7 should work.
This is the one not compatible with Snow Leopard!

EDIT: I will try this http://jksha.blogspot.se/2013/09/java-7-and-snow-leopard-osx-106.html
legendary
Activity: 2142
Merit: 1010
Newbie
November 23, 2013, 05:34:40 PM
Will there be a Mac client at launch? 

The client is cross-platformed. It works everywhere, I could launch it on my fridge if it had more memory.
JRE version not compatible with OSX 10.6.8!
You think an older one would work?

Any Java 7 should work.
legendary
Activity: 1498
Merit: 1000
November 23, 2013, 05:18:23 PM
Will there be a Mac client at launch? 

The client is cross-platformed. It works everywhere, I could launch it on my fridge if it had more memory.
JRE version not compatible with OSX 10.6.8!
You think an older one would work?
The error I get for the latest JRE for 10.6.8

Exception in thread "main" java.lang.NoClassDefFoundError: Nxt
Caused by: java.lang.ClassNotFoundException: Nxt
   at java.net.URLClassLoader$1.run(URLClassLoader.java:202)
   at java.security.AccessController.doPrivileged(Native Method)
   at java.net.URLClassLoader.findClass(URLClassLoader.java:190)
   at java.lang.ClassLoader.loadClass(ClassLoader.java:306)
   at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:301)
   at java.lang.ClassLoader.loadClass(ClassLoader.java:247)
Jump to: