Re: Efficient output for integer types

[Andrew Gierth]

>>>>> "David" == David Fetter <da...@fetter.org> writes:

 David> + return pg_ltostr_zeropad(str, (uint32)0 - (uint32)value, minwidth - 
1);

No, this is just reintroducing the undefined behavior again. Once the
value has been converted to unsigned you can't cast it back to signed or
pass it to a function expecting a signed value, since it will overflow
in the INT_MIN case. (and in this example would probably output '-'
signs until it ran off the end of memory).

Here's how I would do it:

char *
pg_ltostr_zeropad(char *str, int32 value, int32 minwidth)
{
        int32           len;
        uint32          uvalue = value;

        Assert(minwidth > 0);

        if (value >= 0)
        {
                if (value < 100 && minwidth == 2) /* Short cut for common case 
*/
                {
                        memcpy(str, DIGIT_TABLE + value*2, 2);
                        return str + 2;
                }
        }
        else
        {
                *str++ = '-';
                minwidth -= 1;
                uvalue = (uint32)0 - uvalue;
        }
                        
        len = pg_ultoa_n(uvalue, str);
        if (len >= minwidth)
                return str + len;

        memmove(str + minwidth - len, str, len);
        memset(str, '0', minwidth - len);
        return str + minwidth;
}

 David>  pg_ltostr(char *str, int32 value)
 David> +       int32   len = pg_ultoa_n(value, str);
 David> +       return str + len;

This seems to have lost its handling of negative numbers entirely (which
doesn't say much for the regression test coverage)

-- 
Andrew (irc:RhodiumToad)