Re: [BUGS] [JDBC] Error in DatabaseMetaData.getColumns() with Views

[Dario V. Fassi]

I have found that the problem is worst. In the sample adjunct, you can see that error arise at the time when the view's sql text is parsed and saved in database catalog. Then generic NUMERIC type is forced for every calculated column without regard or precision. And at execute time the f2 column has varying type decimals (in row 2 you can see 4 decimals and in other rows has 3 decimals), this is not a behavior , this is an ERROR. Precision calculation  in Numeric fields it's not so difficult. This problem *must* be corrected in the CREATE VIEW sentence , and I offer my collaboration for that. Regards all, Dario Fassi. Dario V. Fassi wrote: Kris Jurka wrote: On Fri, 2 Jul 2004, Dario Fassi wrote: Hi, I wish to report a erroneous information returned by DatabaseMetaData.getColumns() method. This happens with ResultSet's column (7) COLUMN_SIZE and (9) DECIMAL_DIGITS , when DatabaseMetaData.getColumns() inform about a VIEW with columns formed with : coalesce , case or numeric operations over DECIMAL fields. Suppose CREATE TABLE A ( f1 DEC(6,3), f2 dec(6,3) ); CREATE VIEW B as ( select ( f1 + f2 ) as f from a; Then DatabaseMetaData.getColumns() returns: VIEW B F NUMERIC( 65535 , -65531 ) The problem is that f1+f2 does not retain the numeric(6,3) size restriction, but turns into an unbounded plain "numeric" data type. So when retrieving this data the precision/scale are unavailable and the unreasonable values you see are returned. We could return NULL instead, but I'm not sure that would be much more helpful if the client is expecting real values. Any other ideas? Kris Jurka Yes, a few. In the tool named PgManage (come with the commercial version of Mammoth) ,  the information is accurate values for this MetaData, and I believe, they get that values from pqsql catalog tables and not from DatabaseMetaData interface. More even, the engine resolve the View properly and return data values properly typed  (engine return 999.999  for f1+f2  and  999.999999 for f1*f2  , like is expected). So, the information if know or derived en some way for the engine , and/or is contained some where in catalog's tables. I'm working in CodeGeneration tools for many DB engines (DB2, Oracle, MS-Sql, PostgreSql, etc) , and it's impossible not to use generic DatabaseMetaData interface to obtain metadata information. So this is a very important problem for me. DB2 for example do, data type escalation based on "Error propagation Theory "  , that has rules (I'm not and expert in the field) like : [dec(6,a) + dec(6,b) ] -> [ dec( 6, max(a,b) ) ] [ coalesce( dec(6,a) , dec(6,b) ] -> [ dec[ 6, max(a,b)] ] [ case( dec(6,a) , dec(6,b), dec(6,c), dec(6,d) ] -> [ dec[ 6, max(a,b,c,d)] ] [dec(6,a) * dec(6,b ) -> [ dec( 6,a+b ) ] etc. This rules are taken into account in the engine data formation process, but there are stored some where ??? Thanks for your answer. Dario V. Fassi

CREATE TABLE public.t ( 
 f1 numeric(6,3),
 f2 numeric(6,4),
 f3 numeric(8,2)
) WITHOUT OIDS;


insert into public.t values ( 123.123, 12.1234, 12345.12);
insert into public.t values ( null, 12.1234, 12345.12);
insert into public.t values ( 123.123, null, 12345.12);
insert into public.t values ( 123.123, 12.1234, null);


--
--      Create sql text 
--

create view public v as 
select 
(f1+f2+f3) as fsum,
coalesce(f1,f2,f3) as fcoal,
(f1*f2*f3) as fprod
from public.t;

--
--      Parsed sql view from catalog
--

CREATE OR REPLACE VIEW public.v AS 
SELECT ((t.f1 + t.f2 ) + t.f3 ) AS fsum, 
        CASE WHEN (t.f1 IS NOT NULL ) THEN t.f1 
            WHEN (t.f2 IS NOT NULL ) THEN t.f2 
            WHEN (t.f3 IS NOT NULL ) THEN t.f3 
            ELSE NULL::numeric 
        END  AS fcoal, 
        ((t.f1 * t.f2 ) * t.f3 ) AS fprod    FROM t;


--
--      Returned Data 
--

select * from public.v;

fsum            fcoal           fprod
12480.3664      123.123         18427182.594204384
null            12.1234         null                    << !!! No look good !!!
null            123.123         null
null            123.123         null


--
--
--      WORKAROUND WORKAROUND WORKAROUND
--
--

--
--      Create sql text with forced datatype
--

create view public v2 as 
select 
        cast( (f1+f2+f3)                as dec(9,4) ) as fsum,
        cast( coalesce(f1,f2,f3)        as dec(7,4) ) as fcoal,
        cast( (f1*f2*f3)                as dec(20,9) ) as fprod
from public.t;

--
--      Parsed sql view from catalog
--

CREATE OR REPLACE VIEW public.v2 AS 
SELECT 
        (((t.f1 + t.f2 ) + t.f3))::numeric(9,4 ) AS fsum, 
        ( CASE 
            WHEN (t.f1 IS NOT NULL ) THEN t.f1 
            WHEN (t.f2 IS NOT NULL ) THEN t.f2 
            WHEN (t.f3 IS NOT NULL ) THEN t.f3 
            ELSE NULL::numeric END)::numeric(7,4 ) AS fcoal, 
        (((t.f1 * t.f2 ) * t.f3))::numeric(20,9 ) AS fprod 
FROM t; 

--
--      Returned Data 
--

select * from public.v2;

fsum            fcoal           fprod
12480.3664      123.1230        18427182.594204384
null            12.1234         null                    << It's ok now
null            123.1230        null
null            123.1230        null

---------------------------(end of broadcast)---------------------------
TIP 8: explain analyze is your friend