A85EncodeBytes(interp, bytesPtr, numBytes, ctxPtr)
    Tcl_Interp *interp;		/* the current interpreter */
    unsigned char *bytesPtr;	/* the 4-byte sequence to encode */
    int numBytes;		/* how many bytes to encode. If < 4, this
                                 * is the last set in the run. */
    A85EncodeContext *ctxPtr;	/* the encoding context */
{
    long int word = 0;
    int i;
    char toEmit[5];

    for (i=numBytes ; i < 4 ; i++) {
        bytesPtr[i] = 0;
    }

    for (i=3 ; i >= 0 ; i--) {
        word <<= 8;
        word |= bytesPtr[i];
    }

    if (word == 0) {
        A85EmitChar(interp, 'z', ctxPtr);
    }  else {
        int tmp = 0;

        if (word < 0) {
            /* Because some don't support unsigned long */
            tmp = 32;
            word = word - (long)(85L * 85 * 85 * 85 * 32);
        }
        if (word < 0) {
            tmp = 64;
            word = word - (long)(85L * 85 * 85 * 85 * 32);
        }

        /*
         * we emit from least significant to most significant char, so that
         * the 0 chars from an incomplete 4-tuple are the last ones in the
         * sequence and can be omitted (for the last 4-tuple in the array)
         */

        toEmit[4] = EN((word / (long)(85L * 85 * 85 * 85)) + tmp);
        word %= (long)(85L * 85 * 85 * 85);
        toEmit[3] = EN(word / (85L * 85 * 85));
        word %= (85L * 85 * 85);
        toEmit[2] = EN(word / (85L * 85));
        word %= (85L * 85);
        toEmit[1] = EN(word / 85);
        word %= 85;
        toEmit[0] = EN(word);

        /*
         * Emit only 'numBytes+1' chars, since the extra ones are all '!'
         * and can therefore be reconstructed by the decoder (if we know the
         * number of bytes that were encoded).
         */