Execute immediate use in select * from in oracle - oracle

Question

I am trying to get maximum length of column data in oracle by executing a dynamic query as part of Select statement , except seems i don't think we can use execute immediate in an select clause. Could i get some help with syntax or understanding as to better way to do this.
SELECT
owner OWNER,
table_name,
column_name,
'select max(length('||column_name||')) from '||table_name||';' max_data_length
FROM
dba_tab_columns
WHERE
AND ( data_type = 'NUMBER'
OR data_type = 'INTEGER' )
the 4th column in above query spits out a sql string rather than computing the value and returning it.

Answer 1

Here is some food for thought. Note that I am only looking for numeric columns that don't already have precision specified in the catalog. (If you prefer, you can audit all numeric columns and compare the declared precision against the actual precision used by your data.)
I am also looking only in specific schemas. Instead, you may give a list of schemas to be ignored; I hope you are not seriously considering making any changes to SYS, for example, even if it does (and it does!) have numeric columns without specified precision.
The catalog doesn't store INTEGER in the data type; instead, it stores that as NUMBER(38) So I am not searching for data type INTEGER in DBA_TAB_COLUMNS. But this raises an interesting question - perhaps you should search for all columns where DATA_PRECISION is null (as in my code below), but also for DATA_PRECISION = 38.
In the code below I use DBMS_OUTPUT to display the findings directly to the screen. You will probably want to do something smarter with this; either create a table function, or create a table and insert the findings in it, or perhaps even issue DDL already (note that those also require dynamic SQL).
This still leaves you to deal with scale. Perhaps you can get around that with a specification like NUMBER(prec, *) - not sure if that will meet your needs. But the idea is similar; you will just need to write code carefully, like I did for precision (accounting for the decimal point and the minus sign, for example).
Long story short, here is what I ran on my system, and the output it produced.
declare
prec number;
begin
for rec in (
select owner, table_name, column_name
from all_tab_columns
where owner in ('SCOTT', 'HR')
and data_type = 'NUMBER'
and data_precision is null
)
loop
execute immediate
'select max(length(translate(to_char(' || rec.column_name ||
'), ''0-.'', ''0'')))
from ' || rec.owner || '.' || rec.table_name
into prec;
dbms_output.put_line('owner: ' || lpad(rec.owner, 12, ' ') ||
' table name: ' || lpad(rec.table_name, 12, ' ') ||
' column_name: ' || lpad(rec.column_name, 12, ' ') ||
' precision: ' || prec);
end loop;
end;
/
PL/SQL procedure successfully completed.
owner: HR table name: REGIONS column_name: REGION_ID precision: 1
owner: HR table name: COUNTRIES column_name: REGION_ID precision: 1
owner: SCOTT table name: SALGRADE column_name: GRADE precision: 1
owner: SCOTT table name: SALGRADE column_name: LOSAL precision: 4
owner: SCOTT table name: SALGRADE column_name: HISAL precision: 4
PL/SQL procedure successfully completed.
EDIT
Here are several additional points (mostly, corrections) based on extended conversations with Sayan Malakshinov in comments to my answer and to his.
Most importantly, even if we can figure out max precision of numeric columns, that doesn't seem directly related to the ultimate goal of this whole thing, which is to determine the correct Postgre data types for the existing Oracle columns. For example in Postgre, unlike Oracle, it is important to distinguish between integer and non-integer. Unless scale is explicitly 0 in Oracle, we don't know that the column is "integers only"; we could find that out, through a similar dynamic SQL approach, but we would be checking for non-integer values, not precision.
Various corrections: My query is careless with regard to quoted identifiers (schema name, table name, column name). See the proper use of double-quotes in the dynamic query in Sayan's answer; my dynamic query should be modified to use double-quotes in the same way his does.
In my approach I pass numbers through TO_CHAR and then remove minus sign and decimal period. Of course, one's system may use comma, or other symbols, for decimal separator; the safer approach is to remove everything that is not a digit. That can be done with
translate(col_name, '0123456789' || col_name, '0123456789')
The query also doesn't handle very large or very small numbers, which can be stored in the Oracle database, but can only be represented in scientific notation when passed through TO_CHAR().
In any case, since "max precision" doesn't seem directly related to the ultimate goal of mapping to correct data types in Postgre, I am not changing the code - leaving it in the original form.
Thanks to Sayan for pointing out all these issues.
One more thing - *_TAB_COLUMNS contains information about view columns too; very likely those should be ignored for the task at hand. Very easy to do, as long as we realize it needs to be done.

Answer 2

Reading carefully that AWS article and since both previous answers (including mine) use rough estimate (length+to_char without format model and vsize operate decimal length, not bytes), I decided to write another full answer.
Look at this simple example:
with
function f_bin(x number) return varchar2 as
bi binary_integer;
e_overflow exception;
pragma exception_init(e_overflow, -1426);
begin
bi:=x;
return case when bi=x then 'ok' else 'error' end;
exception when e_overflow then return 'error';
end;
function f_check(x number, f varchar2) return varchar2 as
begin
return case when to_number(to_char(abs(x),f),f) = abs(x) then 'ok' else 'error' end;
exception when VALUE_ERROR then return 'error';
end;
a(a1) as (
select * from table(sys.odcinumberlist(
1,
0.1,
-0.1,
-7,
power(2,15)-1,
power(2,16)-1,
power(2,31)-1,
power(2,32)-1
))
)
select
a1,
f_check(a1,'fm0XXX') byte2,
f_check(a1,'fm0XXXXXXX') byte4,
f_bin(a1) ff_signed_binary_int,
to_char(abs(a1),'fm0XXXXXXXXXXXXXXX') f_byte8,
f_check(a1,'fm0XXXXXXXXXXXXXXX') byte8,
vsize(a1) vs,
dump(a1) dmp
from a;
Result:
A1 BYTE2 BYTE4 FF_SIGNED_ F_BYTE8 BYTE8 VS DMP
---------- ---------- ---------- ---------- ---------------- ---------- ---------- ----------------------------------------
1 ok ok ok 0000000000000001 ok 2 Typ=2 Len=2: 193,2
.1 error error error 0000000000000000 error 2 Typ=2 Len=2: 192,11
-.1 error error error 0000000000000000 error 3 Typ=2 Len=3: 63,91,102
-7 ok ok ok 0000000000000007 ok 3 Typ=2 Len=3: 62,94,102
32767 ok ok ok 0000000000007FFF ok 4 Typ=2 Len=4: 195,4,28,68
65535 ok ok ok 000000000000FFFF ok 4 Typ=2 Len=4: 195,7,56,36
2147483647 error ok ok 000000007FFFFFFF ok 6 Typ=2 Len=6: 197,22,48,49,37,48
4294967295 error ok error 00000000FFFFFFFF ok 6 Typ=2 Len=6: 197,43,95,97,73,96
Here I used PL/SQL functions for readability and to make it more clear.
Function f_bin casts an input number parameter to PL/SQL binary_integer (signed int4) and compares the result with input parameter, ie it checks if it loses accuracy. Defined exception shows that it can raise an exception 1426 "numeric overflow".
Function f_check does double conversion to_number(to_char(...)) of the input value and checks if it's still equal to the input value. Here I use hexadecimal format mask (XX = 1 byte), so it checks if an input number can fit an in this format mask. Hexadecimal format model works with non-negative numbers, so we need to use abs() here.
F_BYTE8 shows formatted value that uses a function from the column BYTE8, so you can easily see the loss of accuracy here.
All the above were just for readability, but we can make the same using just pure SQL:
with
a(a1) as (
select * from table(sys.odcinumberlist(
1,
0.1,
-0.1,
-7,
power(2,15)-1,
power(2,16)-1,
power(2,31)-1,
power(2,32)-1
))
)
select
a1,
case when abs(a1) = to_number(to_char(abs(a1),'fmXXXXXXXXXXXXXXX') default null on conversion error,'fmXXXXXXXXXXXXXXX')
then ceil(length(to_char(abs(a1),'fmXXXXXXXXXXXXXXX'))/2)
else -1
end xx,
vsize(a1) vs,
dump(a1) dmp
from a;
Result:
A1 XX VS DMP
---------- ---------- ---------- ----------------------------------------
1 1 2 Typ=2 Len=2: 193,2
.1 -1 2 Typ=2 Len=2: 192,11
-.1 -1 3 Typ=2 Len=3: 63,91,102
-7 1 3 Typ=2 Len=3: 62,94,102
32767 2 4 Typ=2 Len=4: 195,4,28,68
65535 2 4 Typ=2 Len=4: 195,7,56,36
2147483647 4 6 Typ=2 Len=6: 197,22,48,49,37,48
4294967295 4 6 Typ=2 Len=6: 197,43,95,97,73,96
As you can see, here I return -1 in case of conversion errors to byte8 and number of non-zero bytes otherwize.
Obviusly it can be simplified even more: you can just check range limits and that x=trunc(x) or mod(x,1)=0.

Answer 3

Looks like that is what you need:
VSIZE returns the number of bytes in the internal representation of expr. If expr is null, then this function returns null.
https://docs.oracle.com/en/database/oracle/oracle-database/19/sqlrf/VSIZE.html
In Oracle INTEGER is just a number(*,0): http://orasql.org/2012/11/10/differences-between-integerint-in-sql-and-plsql/
select
owner,table_name,column_name,
data_type,data_length,data_precision,data_scale,avg_col_len
,x.vs
from (select/*+ no_merge */ c.*
from dba_tab_columns c
where data_type='NUMBER'
and owner not in (select username from dba_users where ORACLE_MAINTAINED='Y')
) c
,xmltable(
'/ROWSET/ROW/VSIZE'
passing dbms_xmlgen.getxmltype('select nvl(max(vsize("'||c.column_name||'")),0) as VSIZE from "'||c.owner||'"."'||c.table_name||'"')
columns vs int path '.'
) x
;
Update: if you read oracle internal number format (exponent+bcd-mantissa and look at a result of dump(x) function, you can see that Oracle stores numbers as a decimal value(4 bits per 1 decimal digit, 2 digits in 1 byte), so for such small ranges you can just take their maximum BCD mantissa+1 exponent as a rough estimation.