def start_debugger()

in agents/DebugSQLAgent.py [0:0]
56 lines of code
10 McCabe index (conditional complexity)

    def start_debugger  (self,
                        source_type,
                        user_grouping,
                        query,
                        user_question, 
                        SQLChecker,
                        tables_schema, 
                        columns_schema,
                        AUDIT_TEXT, 
                        similar_sql="-No examples provided..-", 
                        DEBUGGING_ROUNDS = 2,
                        LLM_VALIDATION=False):
        i = 0  
        STOP = False 
        invalid_response = False 
        chat_session = self.init_chat(source_type,user_grouping,tables_schema,columns_schema,similar_sql)
        sql = query.replace("```sql","").replace("```","").replace("EXPLAIN ANALYZE ","")

        AUDIT_TEXT=AUDIT_TEXT+"\n\nEntering the debugging steps!"
        while (not STOP):

            json_syntax_result={ "valid":True, "errors":"None"}
            # Check if LLM Validation is enabled 
            if LLM_VALIDATION: 
                # sql = query.replace("```sql","").replace("```","").replace("EXPLAIN ANALYZE ","")
                json_syntax_result = SQLChecker.check(source_type,user_question,tables_schema,columns_schema, sql) 

            else: 
                json_syntax_result['valid'] = True
                AUDIT_TEXT=AUDIT_TEXT+"\nLLM Validation is deactivated. Jumping directly to dry run execution."
 

            if json_syntax_result['valid'] is True:
                AUDIT_TEXT=AUDIT_TEXT+"\nGenerated SQL is syntactically correct as per LLM Validation!"
                   
                # print(AUDIT_TEXT)
                if source_type=='bigquery':
                    connector=bqconnector
                else:
                    connector=pgconnector
                    
                correct_sql, exec_result_df = connector.test_sql_plan_execution(sql)
                
                if not correct_sql:
                        AUDIT_TEXT=AUDIT_TEXT+"\nGenerated SQL failed on execution! Here is the feedback from bigquery dryrun/ explain plan:  \n" + str(exec_result_df)
                        rewrite_result = self.rewrite_sql_chat(chat_session, sql, user_question, exec_result_df)
                        print('\n Rewritten and Cleaned SQL: ' + str(rewrite_result))
                        AUDIT_TEXT=AUDIT_TEXT+"\nRewritten and Cleaned SQL: \n' + str({rewrite_result})"
                        sql = str(rewrite_result).replace("```sql","").replace("```","").replace("EXPLAIN ANALYZE ","")

                else: STOP = True
            else:
                print(f'\nGenerated qeury failed on syntax check as per LLM Validation!\nError Message from LLM:  {json_syntax_result} \nRewriting the query...')
                AUDIT_TEXT=AUDIT_TEXT+'\nGenerated qeury failed on syntax check as per LLM Validation! \nError Message from LLM:  '+ str(json_syntax_result) + '\nRewriting the query...'
                
                syntax_err_df = pd.read_json(json.dumps(json_syntax_result))
                rewrite_result=self.rewrite_sql_chat(chat_session, sql, user_question, syntax_err_df)
                print(rewrite_result)
                AUDIT_TEXT=AUDIT_TEXT+'\n Rewritten SQL: ' + str(rewrite_result)
                sql=str(rewrite_result).replace("```sql","").replace("```","").replace("EXPLAIN ANALYZE ","")
            i+=1
            if i > DEBUGGING_ROUNDS:
                AUDIT_TEXT=AUDIT_TEXT+ "Exceeded the number of iterations for correction!"
                AUDIT_TEXT=AUDIT_TEXT+ "The generated SQL can be invalid!"
                STOP = True
                invalid_response=True
            # After the while is completed
        if i > DEBUGGING_ROUNDS:
            invalid_response=True
        return sql, invalid_response, AUDIT_TEXT