Hello Tsutomu,

At 11:24 AM 15/07/2003 +0900, you wrote:

>Hi all
>
>SELECT
> ROUND(-1.4),
> ROUND((-1.4)),
> ROUND(ABS(-1.4))
>FROM
> RDB$DATABASE
>;
>
> ROUND ROUND ROUND
>============ ============ ============
> -1 -1 <null>

This gives you the expected result:

SELECT
ROUND(-1.4),
ROUND((-1.4)),
ABS(ROUND(-1.4))
FROM
RDB$DATABASE;

ROUND ROUND ROUND
============ ============ ============
-1 -1 1

>I think it is caused by ABS() is in ib_udf and ROUND() is in fbudf. Is
>this correct? Is this bug?

Not by any rule about UDFs. You can use UDFs from any number of libraries.

There is probably something "funny" about the arithmetic, though. fbudf
passes parameters by descriptor, whereas ib_udf passes them by
C-datatype. Somehow, it works OK when passing the output from a descriptor
to the ABS() function (which expects a double precision number) but not OK
when passing the output of ABS (double precision) to ROUND.

I seem to recall Claudio stating that these "by descriptor" numeric
functions only work with scaled numbers. You will find that this produces
the expected result:

SELECT
ROUND(-1.4),
ROUND((-1.4)),
ROUND(cast (ABS(-1.4) as numeric (18,0)))
FROM
RDB$DATABASE

ROUND ROUND ROUND
============ ============ ============
-1 -1 1

I guess it's just one of those things one needs to be aware of.

heLen