EBCDIC bug(s?) in our apr-util//crypto/crypt_blowfish.c implementation

[William A Rowe Jr]

This doesn't look good.  Does an EBCDIC architecture fan want to whip
up the fix?


static char *BF_crypt(const char *key, const char *setting,
        char *output, int size,
        BF_word min)
{
...     static const unsigned char flags_by_subtype[26] =
                {2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
                0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 4, 0};
...     if (setting[0] != '$' ||
            setting[1] != '2' ||
            setting[2] < 'a' || setting[2] > 'z' ||
            !flags_by_subtype[(unsigned int)(unsigned char)setting[2] - 'a'] ||