How extract table names from PL/SQL package body file? - sql

Question

I need to get the table names queried in a pl/sql package file.
I know that there is an option for this in Notepad++ by regex but I don't know what regex to apply for get the table names (I understand that must be some regex to take the keyword "FROM" and get the next string after space, I think so).
For the next example code:
CREATE OR REPLACE PACKAGE BODY pac_example AS
FUNCTION f1 RETURN NUMBER IS
BEGIN
SELECT * FROM table1;
RETURN 1;
END f1;
FUNCTION f2 RETURN NUMBER IS
BEGIN
SELECT * FROM table2;
RETURN 1;
END f2;
END pac_example;
And I expect replace all and get the file with only its table names:
table1
table2

Answer 1

If you're interested only in table names that are directly referred from the PACKAGE BODY, a simple and straight-forward method is to query all_dependencies or user_dependencies.
SELECT owner,
referenced_name as table_name
FROM all_dependencies
WHERE type IN (
'PACKAGE BODY'
) AND name IN (
'PAC_EXAMPLE'
) AND referenced_type = 'TABLE';
DEMO

Answer 2

To my knowledge no one has done this with 100% accuracy. The closest you get is the ALL/DBA_DEPENDENIES but does not tell you if the table is accessed in a SELECT, INSERT, UPDATE or DELETE.
It will however resolve synonyms.
The downside of this is that it will not include tables referenced in dynamic SQL.
If you have a database that uses particular naming convention for tables (e.g. Tnnn_XXXXX ) you could do:
SELECT DISTINCT c.text, c.name, c.type, t.table_name
FROM user_source c, user_tables t
WHERE UPPER(t.text) like '%' || t.name_name || '%' -- Maybe REGEXP_LIKE better
ORDER BY 2, 1, 4;
I worked on a project decades ago where they wanted a CRUD matrix of programs (PLSQL, SQL, Oracle Forms/Reports, ProC, ProCOBOL) and what tables each accessed.
The only solution available at the time was for me to write a parser (in C) that parsed the codebase looking for SQL and processing it. Mine even reported columns as well as tables. The C program parsed the source, looking for KEYWORDS and characters to control a state engine. It took a a couple of weeks to refine and get working across all the different codebase types.
By the end, the only thing it could not do was dynamic queries where the table name was built up from variable values. But the workaround here was to capture the tkprof files and process these.
Tragically, I do not have the source code for this anymore.
However, if I were to do it again, I would use Lex/Yacc/Bison to parser SQL and build a system around these tools.
A quick search found this:
https://github.com/jgarzik/sqlfun
https://www.oreilly.com/library/view/flex-bison/9780596805418/ch04.html
Not a small undertaking.

Answer 3

Ctrl+H
Find what: (?:\A(?:(?!FROM).)*|\G)FROM\s+(\w+(?:\s*,\s*\w+)*)(?:(?!FROM).)*
Replace with: " #a space and a double quote
check Wrap around
check Regular expression
CHECK . matches newline
Replace all
Explanation:
(?: # start non capture group
\A # beginning of file
(?:(?!FROM).)* # Tempered greedy token, make sure we haven't FROM before
| # OR
\G # restart from last match position
) # end group
FROM\s+ # literally FROM followed by 1 or more spaces
( # start group 1
\w+ # 1 or more word characters (table name)
(?:\s*,\s*\w+)* # non capture group spaces comma spaces and 1 or more word characters, optional more tables
) # end group
(?:(?!FROM).)* # Tempered greedy token, make sure we haven't FROM
Replacement:
$1 # content of group 1, table name
Screen capture:

Answer 4

You can use following regex to search for table names.
Regex: FROM\s([^;]+)
Replacement: \n%\1%\n
Then follow this answer for replacing other data in file.

Answer 5

the earlier mentioned tables
all_dependencies or user_dependencies
can list out the dependencies as mentioned earlier, but it wont cover the dynamic queries. and the search if done in notepad using key words like 'from' then only tables referred by statements after 'From' statements will be covered.
the below code snippet can be considered for full analysis :- line by line, word by word and analysis the tables (referred to the sample mentioned by you)
declare
l_line varchar2(2000);
ln_start_string number;
ln_last_string number;
ln_string_length number;
l_word varchar2(4000);
l_table_flag varchar(2):='N';
cursor l_pkg_body_cur
is
select TEXT from all_source where upper(name) like upper('pac_example') and type = 'PACKAGE BODY';
-- to get the source compiled in package boby, mentioned the package to be searched here
begin
for rec in l_pkg_body_cur
loop
-- line by line processing
select TRIM(rec.text) into l_line from dual;
ln_string_length := length(l_line);
loop
-- word by word processing
l_table_flag :='N';
select instr(l_line,' ') into ln_last_string from dual;
select substr(l_line,0,ln_last_string) into l_word from dual;
begin
select 'Y' into l_table_flag from all_tables where upper(table_name) like upper(trim(l_word)) and rownum=1; -- to validate it is table or not
exception
when others then
l_table_flag := 'N';
end;
IF l_table_flag = 'Y'
then
dbms_output.put_line(trim(l_word) ); -- table name
end if;
select length (l_word) into ln_start_string from dual;
select trim(substr(replace(l_line,';',null),ln_start_string)) into l_line from dual;
exit when l_line is NULL;
end loop;
end loop;
end;
--output:
Statement processed.
table1
table2
similar way this query can be altered as wanted to find views or synonyms deparated by changing the base table- all_views,all_synonyms, accordingly as per required.
this is like most straight forward approach- might take more processing time depends on the package size
same can be done using UNIX scripting if needed
if needed to check from the file directly UNIX scripting can be used (UTF_file operations can also be used ) to get line by line from the file and have sql session to do the above validation and to display the results
but hopefully this would provide most accurate results.