pl/sql - can collection loop through column names?

问题:

The output from the below code is:

         |LAT|MISC|SID|NO
MIN_LENGTH|1|2|1|1
MAX_LENGTH|6|6|4|2

The output is as I expect, but is there anyway to loop through the columns using an index (ie. j) instead of doing RESULTS(I).MAX_LENGTH , RESULTS(I).MAX_LENGTH etc ? The concern is that when adding extra columns to the 'R_RESULT_REC' record, another loop is required.

set serveroutput on
DECLARE     
  TYPE R_RESULT_REC IS RECORD
    (COL_NAME        VARCHAR2(100),
      MIN_LENGTH         NUMBER,
      MAX_LENGTH         NUMBER
      );       
  TYPE tr_RESULT IS TABLE OF R_RESULT_REC;
  RESULTS   TR_RESULT := TR_RESULT();
  v_counter NUMBER := 1;
BEGIN
  FOR J IN (SELECT DISTINCT COLUMN_NAME FROM ALL_TAB_COLUMNS 
            WHERE OWNER = 'SYSTEM'
            and TABLE_NAME = 'SPECCHAR')
  LOOP
      RESULTS.EXTEND;
      RESULTS(V_COUNTER).COL_NAME := J.COLUMN_NAME;
      EXECUTE IMMEDIATE 'SELECT MIN(LENGTH('||J.COLUMN_NAME||')),
      MAX(LENGTH('||J.COLUMN_NAME||'))
      FROM '||'SYSTEM'||'.'||'SPECCHAR' INTO 
      RESULTS(V_COUNTER).MIN_LENGTH,
      RESULTS(V_COUNTER).MAX_LENGTH; 
      V_COUNTER := V_COUNTER + 1;
  END LOOP;
     FOR I IN RESULTS.FIRST .. RESULTS.LAST LOOP
       IF I = RESULTS.LAST THEN
         DBMS_OUTPUT.PUT_LINE(RESULTS(I).COL_NAME);
       ELSIF I = RESULTS.FIRST THEN
         DBMS_OUTPUT.PUT('         |'||RESULTS(I).COL_NAME||'|');
        ELSE
         DBMS_OUTPUT.PUT(RESULTS(I).COL_NAME||'|');
       END IF ;
    END LOOP;
    FOR I IN RESULTS.FIRST .. RESULTS.LAST LOOP
       IF I = RESULTS.LAST THEN
         DBMS_OUTPUT.PUT_LINE(RESULTS(I).MIN_LENGTH);
        ELSIF I = RESULTS.FIRST THEN
         DBMS_OUTPUT.PUT('MIN_LENGTH|'||RESULTS(I).MIN_LENGTH||'|');
        ELSE
         DBMS_OUTPUT.PUT(RESULTS(I).MIN_LENGTH||'|');
       END IF ;
    END LOOP;
     FOR I IN RESULTS.FIRST .. RESULTS.LAST LOOP
       IF I = RESULTS.LAST THEN
         DBMS_OUTPUT.PUT_LINE(RESULTS(I).MAX_LENGTH);
        ELSIF I = RESULTS.FIRST THEN
         DBMS_OUTPUT.PUT('MAX_LENGTH|'||RESULTS(I).MAX_LENGTH||'|');
        ELSE
         DBMS_OUTPUT.PUT(RESULTS(I).MAX_LENGTH||'|');
       END IF ;
    END LOOP;
end;

回答1:

This uses DBMS_SQL, so it's pretty snarly to read. The main reason I saw to use it was that I could get columnar descriptions of a SQL statement and to a buffer-based, not object-based fetch.

Rather than making calls to DBMS_OUTPUT during the processing, it builds a table of records for output, using associative arrays for simplicity.

It could further be refined to have an array or parsable list of functions to apply to each function, but that seems excess to current requirements. The nature of the code would require editing if new aggregation functions are being added.

Call overview (2c + a + s):

• 3 loops;
  • 2 loops over column list (c),
  • 1 loop over number of analytic functions (a).
• 1 SQL statement against table data (s).

OP's call overview (c*s + a + 1):

• 1 loop, executing a sql statement against table data per column (c*s)
• a+1 loops, where a is the number of analytic functions

Test data:

  1  select  min(length(GP_ID)),  max(length(GP_ID)),
  2          min(length(GGP_ID)),  max(length(GGP_ID)),
  3          min(length(OBJECT_NAME)),  max(length(OBJECT_NAME))
  4*   from AMUSCH.GP
SQL> /

MIN(LENGTH(GP_ID)) MAX(LENGTH(GP_ID)) MIN(LENGTH(GGP_ID)) 
MAX(LENGTH(GGP_ID)) MIN(LENGTH(OBJECT_NAME)) MAX(LENGTH(OBJECT_NAME))
                  1                  7                   1                              
                  4                        9                       41

Code:

declare
  p_owner         varchar2(30);
  p_table_name    varchar2(30);

  TYPE OUTPUT_TAB_TYPE IS TABLE OF VARCHAR2(32767) index by binary_integer;
  OUTPUT_TAB OUTPUT_TAB_TYPE;

  l_columns_tab   dbms_sql.desc_tab;
  l_columns_cur   integer;
  l_columns_sql   varchar2(32767);
  l_columns_cnt   number;

  l_minmax_sql    varchar2(32767);
  l_minmax_cur    integer;
  l_minmax_tab    dbms_sql.desc_tab;
  l_minmax_cnt    number;

  l_fetch_ok      number;
  l_fetch_value   number;
begin

  p_owner := 'AMUSCH';
  p_table_name := 'GP';

  output_tab(1) := lpad(' ', 20, ' ');
  output_tab(2) := lpad('MIN_LENGTH', 20, ' ');
  output_tab(3) := lpad('MAX_LENGTH', 20, ' ');

  l_columns_sql := 'select * from ' || p_owner || '.' || p_table_name || 
     ' where 1 = 0';
  l_columns_cur := dbms_sql.open_cursor;
  dbms_sql.parse (l_columns_cur, l_columns_sql, dbms_sql.native);
  dbms_sql.describe_columns (l_columns_cur, l_columns_cnt, l_columns_tab);

  -- build the min/max sql statement
  l_minmax_sql := 'select ' ;
  for i in 1..l_columns_cnt
  loop
    l_minmax_sql := l_minmax_sql || 
          ' min(length(' || l_columns_tab(i).col_name || ')), ';
    l_minmax_sql := l_minmax_sql || 
          ' max(length(' || l_columns_tab(i).col_name || ')), ';
  end loop;
  l_minmax_sql := substr(l_minmax_sql, 1, 
                         length(l_minmax_sql) - 2); -- trim trailing comma
  l_minmax_sql := l_minmax_sql || ' from ' || p_owner || '.' || p_table_name;

  l_minmax_cur := dbms_sql.open_cursor;
  dbms_sql.parse (l_minmax_cur, l_minmax_sql, dbms_sql.native);
  dbms_sql.describe_columns (l_minmax_cur, l_minmax_cnt, l_minmax_tab);

  for i in 1..l_minmax_cnt
  loop
    dbms_sql.define_column(l_minmax_cur, i, l_fetch_value);
  end loop;

  l_fetch_ok := dbms_sql.execute(l_minmax_cur);

  loop
    l_fetch_ok := dbms_sql.fetch_rows(l_minmax_cur);
    exit when l_fetch_ok = 0;

    -- loop over the columns selected over
    for i in 1..l_columns_cnt
    loop
      output_tab(1) := output_tab(1) || '|' || l_columns_tab(i).col_name;

      dbms_sql.column_value(l_minmax_cur, (2*i-1), l_fetch_value);
      output_tab(2) := output_tab(2) || '|' || 
        lpad(l_fetch_value, length(l_columns_tab(i).col_name), ' ');
      dbms_sql.column_value(l_minmax_cur, (2*i), l_fetch_value);
      output_tab(3) := output_tab(3) || '|' || 
        lpad(l_fetch_value, length(l_columns_tab(i).col_name), ' ');

    end loop;
  end loop;

  if dbms_sql.is_open(l_minmax_cur) then
    dbms_sql.close_cursor (l_minmax_cur);
  end if;

  if dbms_sql.is_open (l_columns_cur) then
    dbms_sql.close_cursor (l_columns_cur);
  end if;

  for i in output_tab.first..output_tab.last
  loop
    dbms_output.put_line(output_tab(i));
  end loop;
end;
/

Results:

                |GP_ID|GGP_ID|OBJECT_NAME
      MIN_LENGTH|    1|     1|          9
      MAX_LENGTH|    7|     4|         41

回答2:

If you want to use the DBMS_SQL package (which is sometimes very complex), then there is a DBMS_SQL.COLUMN_VALUE function that may work for you.

update:
Or even better: DBMS_SQL.DESC_REC you can refer to: http://docs.oracle.com/cd/B19306_01/appdev.102/b14258/d_sql.htm#i996963
notice example 8

I haven't tested it

update:
Perhaps what you really want is to loop on an Object type attributes and not a table column, so maybe you should try a different approach:
Make your type R_RESULT_REC an Object type in the DB and then you can loop on the query results:

SELECT attr_name
FROM user_type_attrs
WHERE type_name = 'R_RESULT_REC'

It's not like working with indexes but you still don't need to hard code the column names / type attributes

here is the code (based on yours):

CREATE OR REPLACE TYPE R_RESULT_REC AS OBJECT
(
  COL_NAME        VARCHAR2(100),
  MIN_LENGTH         NUMBER,
  MAX_LENGTH         NUMBER
);
/

and then:

DECLARE

  TYPE tr_RESULT IS TABLE OF R_RESULT_REC;
  RESULTS   TR_RESULT := TR_RESULT();
  v_counter NUMBER := 1;
  v_max     number;
  v_min     number;

BEGIN
  FOR J IN (SELECT DISTINCT COLUMN_NAME
              FROM ALL_TAB_COLUMNS
             WHERE OWNER = 'SYSTEM'
               and TABLE_NAME = 'SPECCHAR') LOOP

    EXECUTE IMMEDIATE 'SELECT MIN(LENGTH(' || J.COLUMN_NAME || ')),
      MAX(LENGTH(' || J.COLUMN_NAME || '))      FROM ' ||
                      'SPECCHAR'
      INTO v_min, v_max;

    RESULTS.EXTEND;
    RESULTS(V_COUNTER) := new R_RESULT_REC(J.COLUMN_NAME, v_min, v_max);
    V_COUNTER := V_COUNTER + 1;
  END LOOP;

  for r in (select attr_name
              from all_type_attrs t
             where t.owner = 'SYSTEM'
               and t.type_name = 'R_RESULT_REC') loop

    FOR I IN RESULTS.FIRST .. RESULTS.LAST LOOP
      IF I = RESULTS.LAST THEN
        execute immediate 'declare rec R_RESULT_REC := :0; begin' ||
                          ' DBMS_OUTPUT.PUT_LINE(rec.' || r.attr_name || ');' ||
                          'end;'
          using RESULTS(I);
      ELSIF I = RESULTS.FIRST THEN
        execute immediate 'declare rec R_RESULT_REC := :0; begin' ||
                          ' DBMS_OUTPUT.PUT(''' || r.attr_name ||
                          '|'' || rec.' || r.attr_name || ' || ''|'');' ||
                          'end;'
          using RESULTS(I);
      ELSE
        execute immediate 'declare rec R_RESULT_REC := :0; begin' ||
                          ' DBMS_OUTPUT.PUT(rec.' || r.attr_name ||
                          ' || ''|''); ' || 'end;'
          using RESULTS(I);
      END IF;
    END LOOP;

  end loop;

end;

If you'll add another attribute to the Record (and initiate it with values) , it will automatic display it.

回答3:

Take advantage of Oracle's stats for this.

First, fully build stats on table using dbms_stats.gather_table_stats

Then, create the following function to help translate the raw low/high values that Oracle stores in all_tab_columns

create or replace function show_raw(i_raw raw, i_type varchar2)
return varchar2 is
  l_varchar2 varchar2(32);
  l_number number;
  l_date date;
  l_nvarchar2 nvarchar2(32);
  l_rowid rowid;
  l_char char;
begin

  if (i_type = 'VARCHAR2') then
    DBMS_STATS.CONVERT_RAW_VALUE(i_raw, l_varchar2);
    return to_char(l_varchar2);
  elsif(i_type = 'NUMBER') then
    DBMS_STATS.CONVERT_RAW_VALUE(i_raw, l_number);
    return to_char(l_number);
  elsif(i_type = 'DATE') then
    DBMS_STATS.CONVERT_RAW_VALUE(i_raw, l_date);
    return to_char(l_date);
  elsif(i_type = 'NVARCHAR2') then
    DBMS_STATS.CONVERT_RAW_VALUE(i_raw, l_nvarchar2);
    return to_char(l_nvarchar2);
  elsif(i_type = 'ROWID') then
    DBMS_STATS.CONVERT_RAW_VALUE(i_raw, l_rowid);
    return to_char(l_rowid);
  elsif(i_type = 'CHAR') then
    DBMS_STATS.CONVERT_RAW_VALUE(i_raw, l_char);
    return l_char;
  else return 'Unknown type value';
  end if;
end;

Then, just select the low/high values for each column:

select column_id, 
  column_name, 
  data_type, 
  show_raw(low_value, data_type) as min_val, 
  show_raw(high_value, data_type) as max_val
from all_tab_columns
where table_name = 'SOME_TABLE'
and owner = 'SOME_OWNER'
;