8 Error Handling and Diagnostics

SQLCODE and SQLSTATE

With Release 1.5 of the Oracle Precompilers, the SQLCODE status variable was introduced as the SQL89 standard ANSI/ISO error reporting mechanism. The SQL92 standard listed SQLCODE as a deprecated feature and defined a new status variable, SQLSTATE (introduced with Release 1.6 of the Oracle Precompilers), as the preferred ANSI/ISO error reporting mechanism.

SQLCODE stores error codes and the "not found" condition. It is retained only for compatibility with SQL89 and is likely to be removed from future versions of the standard.

Unlike SQLCODE, SQLSTATE stores error and warning codes and uses a standardized coding scheme. After executing a SQL statement, the Oracle server returns a status code to the SQLSTATE variable currently in scope. The status code indicates whether a SQL statement executed successfully or raised an exception (error or warning condition). To promote interoperability (the ability of systems to exchange information easily), SQL92 predefines all the common SQL exceptions.

SQLCA

The SQLCA is a record-like, host-language data structure. Oracle updates the SQLCA after every executable SQL statement. (SQLCA values are undefined after a declarative statement.) By checking Oracle return codes stored in the SQLCA, your program can determine the outcome of a SQL statement. This can be done in two ways:

• Implicit checking with the WHENEVER statement

• Explicit checking of SQLCA variables

You can use WHENEVER statements, code explicit checks on SQLCA variables, or do both. Generally, using WHENEVER statements is preferable because it is easier, more portable, and ANSI-compliant.

ORACA

When more information is needed about run-time errors than the SQLCA provides, you can use the ORACA, which contains cursor statistics, SQL statement data, option settings, and system statistics.

The ORACA is optional and can be declared regardless of the MODE setting. For more information about the ORACA status variable, refer to "Using the Oracle Communications Area.

Some Historical Information

The treatment of status variables and variable combinations by the Oracle Precompilers has evolved beginning with Release 1.5.

Release 1.5

The Oracle Precompiler, Release 1.5, presumed there was a status variable SQLCODE whether it was declared in a Declare Section; in fact, the precompiler never bothered to note whether there was a declaration for SQLCODE or not -- it just presumed that the declaration exists. SQLCA would be used as a status variable only if there was an INCLUDE of the SQLCA.

Release 1.6

Beginning with Oracle Precompilers, Release 1.6, the precompilers no longer presume that there is a SQLCODE status variable and it is not required. The precompiler requires that at least one of SQLCA, SQLCODE, or SQLSTATE be declared.

SQLCODE is recognized as a status variable if and only if at least one of the following criteria is satisfied:

• It is declared in a Declare Section with exactly the correct datatype.

• The precompiler finds no other status variable.

If the precompiler finds a SQLSTATE declaration (of exactly the correct type of course) in a Declare Section or finds an INCLUDE of the SQLCA, it will not presume SQLCODE is declared.

Release 1.7

Because Release 1.5 of the Oracle Precompilers allowed the SQLCODE variable to be declared outside a Declare Section while declaring SQLCA at the same time, precompilers Release 1.6 and greater are presented with a compatibility problem. A new option, ASSUME_SQLCODE={YES|NO} (default NO), was added to fix this in Release 1.6.7 and is documented as a new feature in Release 1.7.

When ASSUME_SQLCODE=YES, and when SQLSTATE or SQLCA (Pro*FORTRAN only) are declared as status variables, the precompiler presumes SQLCODE is declared irrepsective of whether it is declared in a Declare Section or of the proper type. This causes Releases 1.6.7 and later to act like Release 1.5 in this regard. For information about the precompiler option ASSUME_SQLCODE, refer to "ASSUME_SQLCODE" .

Declaring Status Variables

This section describes how to declare SQLCODE and SQLSTATE. For information about declaring the SQLCA status variable, refer to "Declaring the SQLCA".

Declaring SQLCODE

SQLCODE (SQLCOD in Pro*FORTRAN) must be declared as a 4-byte integer variable either inside or outside the Declare Section, as shown in Table 8-1.

If declared outside the Declare Section, SQLCODE is recognized as a status variable only if ASSUME_SQLCODE=YES. SQLCODE declarations are ignored when MODE={ORACLE|ANSI13}.

By using host languages that allow both local and global declarations, you can declare more than one SQLCODE variable. Access to a local SQLCODE is limited by its scope within your program. After every SQL operation, Oracle returns a status code to the SQLCODE currently in scope. So, your program can learn the outcome of the most recent SQL operation by checking SQLCODE explicitly, or implicitly with the WHENEVER statement.

When you declare SQLCODE instead of the SQLCA in a particular compilation unit, the precompiler allocates an internal SQLCA for that unit. Your host program cannot access the internal SQLCA. If you declare the SQLCA and SQLCODE (not supported in Pro*COBOL), Oracle returns the same status code to both after every SQL operation.

Declaring SQLSTATE

SQLSTATE (SQLSTA in Pro*FORTRAN) must be declared as a five-character alphanumeric string inside the Declare Section, as shown inTable 8-2. Declaring the SQLCA is optional.

When MODE={ORACLE|ANSI13}, declarations of the SQLSTATE variable are ignored.

Status Variable Combinations

When MODE={ANSI|ANSI14}, the behavior of the status variables depends on the following:

• Which variables are declared

• Declaration placement (inside or outside the Declare Section)

• ASSUME_SQLCODE setting

Table 8-3 and Table 8-4 describe the resulting behavior of each status variable combination when ASSUME_SQLCODE=NO and when ASSUME_SQLCODE=YES, respectively.

Status Variable Values

This section describes the values for the SQLCODE and SQLSTATE status variables. For information about the SQLCA status variable, refer to "Key Components of Error Reporting".

SQLCODE Values

After every SQL operation, Oracle returns a status code to the SQLCODE variable currently in scope. The status code, which indicates the outcome of the SQL operation, can be any of the following numbers:

0

Oracle executed the SQL statement without detecting an error or exception.

> 0

Oracle executed the statement but detected an exception. This occurs when Oracle cannot find a row that meets the condition in your WHERE clause or when a SELECT INTO or FETCH returns no rows.

When MODE={ANSI|ANSI14|ANSI13}, +100 is returned to SQLCODE after an INSERT of no rows. This can happen when a subquery returns no rows to process.

< 0

Oracle did not execute the statement because of a database, system, network, or application error. Such errors are irrecoverable. When they occur, the current transaction should, in most cases, be rolled back. Negative return codes correspond to error codes listed in Oracle Database Error Messages.

You can learn the outcome of the most recent SQL operation by checking SQLCODE explicitly with your own code or implicitly with the WHENEVER statement.

When you declare SQLCODE instead of the SQLCA in a particular precompilation unit, the precompiler allocates an internal SQLCA for that unit. Your host program cannot access the internal SQLCA. If you declare the SQLCA and SQLCODE (Pro*FORTRAN only), Oracle returns the same status code to both after every SQL operation.

SQLSTATE Values

SQLSTATE status codes consist of a two-character class code followed by a three-character subclass code. Aside from class code 00 (successful completion), the class code denotes a category of exceptions. Aside from subclass code 000 (not applicable), the subclass code denotes a specific exception within that category. For example, the SQLSTATE value `22012' consists of class code 22 (data exception) and subclass code 012 (division by zero).

Each of the five characters in a SQLSTATE value is a digit (0..9) or an uppercase Latin letter (A..Z). Class codes that begin with a digit in the range 0..4 or a letter in the range A..H are reserved for predefined conditions (those defined in SQL92). All other class codes are reserved for implementation-defined conditions. Within predefined classes, subclass codes that begin with a digit in the range 0..4 or a letter in the range A..H are reserved for predefined subconditions. All other subclass codes are reserved for implementation-defined subconditions. Figure 8-1 shows the coding scheme.

Figure 8-1 SQLSTATE Coding Scheme

Description of "Figure 8-1 SQLSTATE Coding Scheme"

Table 8-5 shows the classes predefined by SQL92.

Table 8-6 shows how Oracle errors map to SQLSTATE status codes. In some cases, several Oracle errors map to the status code. In other cases, no Oracle error maps to the status code (so the last column is empty). Status codes in the range 60000 .. 99999 are implementation-defined.

Declaring the SQLCA

To declare the SQLCA, simply include it (using an EXEC SQL INCLUDE statement) in your host-language source file as follows:

* Include the Oracle Communications Area (ORACA).
 EXEC SQL INCLUDE ORACA
EXEC SQL INCLUDE SQLCA;

The SQLCA is used if and only if there is an INCLUDE of the SQLCA.

When you precompile your program, the INCLUDE SQLCA statement is replaced by several variable declarations that allow Oracle to communicate with the program.

Declaring the SQLCA in Pro*COBOL

In Pro*COBOL, it makes no difference whether the INCLUDE is inside or outside of a Declare Section. For more information about declaring the SQLCA in Pro*COBOL, refer to Pro*COBOL Programmer's Guide.

Declaring the SQLCA in Pro*FORTRAN

In Pro*FORTRAN, the SQLCA must be declared outside the Declare Section, because it is a COMMON block. Furthermore, the SQLCA must come before the CONNECT statement and the first executable FORTRAN statement.

You must declare the SQLCA in each subroutine and function that contains SQL statements. Every time a SQL statement in one of the subroutines or functions is executed, Oracle updates the SQLCA held in the COMMON block.

Ordinarily, only the order and datatypes of variables in a COMMON-list matter, not their names. However, you cannot rename the SQLCA variables because the precompiler generates code that refers to them. Thus, all declarations of the SQLCA must be identical. For more information about declaring the SQLCA in Pro*FORTRAN, refer to Pro*FORTRAN Supplement to the Oracle Precompilers Guide.

What's in the SQLCA?

The SQLCA contains the following run-time information about the outcome of SQL statements:

• Oracle error codes

• Warning flags

• Event information

• Rows-processed count

• Diagnostics

Figure 8-2 shows all the variables in the SQLCA. To see the SQLCA structure and variable names in a particular host language, refer to your supplement to this Guide.

Figure 8-2 SQLCA Variables

Description of "Figure 8-2 SQLCA Variables"

Key Components of Error Reporting

Error reporting depends on variables in the SQLCA. This section highlights the key components of error reporting. The next section takes a close look at the SQLCA.

Status Codes

Every executable SQL statement returns a status code to the SQLCA variable SQLCODE, which you can check implicitly with the WHENEVER statement or explicitly with your own code.

Status codes can be zero, less than zero, or greater than zero. Refer to Declaring SQLCODE" for complete SQLCODE descriptions.

Warning Flags

Warning flags are returned in the SQLCA variables SQLWARN(0) through SQLWARN(7), which you can check implicitly or explicitly. These warning flags are useful for run-time conditions not considered errors by Oracle. For example, when DBMS=V6, if an indicator variable is available, Oracle signals a warning after assigning a truncated column value to a host variable. (If no indicator variable is available, Oracle issues an error message.)

Rows-Processed Count

The number of rows processed by the most recently executed SQL statement is returned in the SQLCA variable SQLERRD(3), which you can check explicitly.

Speaking strictly, this variable is not for error reporting, but it can help you avoid mistakes. For example, suppose you expect to delete about ten rows from a table. After the deletion, you check SQLERRD(3) and find that 75 rows were processed. To be safe, you might want to roll back the deletion and examine the search condition in your WHERE clause.

Parse Error Offset

Before executing a SQL statement, Oracle must parse it, that is, examine it to make sure it follows syntax rules and refers to valid database objects. If Oracle finds an error, an offset is stored in the SQLCA variable SQLERRD(5), which you can check explicitly. The offset specifies the character position in the SQL statement at which the parse error begins. The first character occupies position zero. For example, if the offset is 9, the parse error begins at the tenth character.

By default, static SQL statements are checked for syntactic errors at precompile time. So, SQLERRD(5) is most useful for debugging dynamic SQL statements, which your program accepts or builds at run time.

Parse errors arise from missing, misplaced, or misspelled keywords, invalid options, nonexistent tables, and the like. For example, the dynamic SQL statement

UPDATE EMP SET JIB = :job_title WHERE EMPNO = :emp_number

causes the parse error

ORA-00904: invalid column name

because the column name JOB is misspelled. The value of SQLERRD(5) is 15 because the erroneous column name JIB begins at the sixteenth character.

If your SQL statement does not cause a parse error, Oracle sets SQLERRD(5) to zero. Oracle also sets SQLERRD(5) to zero if a parse error begins at the first character (which occupies position zero). So, check SQLERRD(5) only if SQLCODE is negative, which means that an error has occurred.

Error Message Text

The error code and message for Oracle errors are available in the SQLCA variable SQLERRMC. At most, the first 70 characters of text are stored. To get the full text of messages longer than 70 characters, you use the SQLGLM function. Refer to "Getting the Full Text of Error Messages".

SQLCA Structure

This section describes the structure of the SQLCA, its fields, and the values they can store.

SQLCAID

This string field is initialized to "SQLCA" to identify the SQL Communications Area.

SQLCABC

This integer field holds the length, in bytes, of the SQLCA structure.

SQLCODE

This integer field holds the status code of the most recently executed SQL statement. The status code, which indicates the outcome of the SQL operation, can be any of the following numbers:

0

Oracle executed the statement without detecting an error or exception.

> 0

Oracle executed the statement but detected an exception. This occurs when Oracle cannot find a row that meets your WHERE-clause search condition or when a SELECT INTO or FETCH returns no rows.

< 0

When MODE={ANSI|ANSI14|ANSI13}, +100 is returned to SQLCODE after an INSERT of no rows. This can happen when a subquery returns no rows to process.

Oracle did not execute the statement because of a database, system, network, or application error. Such errors are irrecoverable. When they occur, the current transaction should, in most cases, be rolled back.

Negative return codes correspond to error codes listed in Oracle Database Error Messages.

SQLERRM

This subrecord contains the following two fields:

SQLERRML

This integer field holds the length of the message text stored in SQLERRMC.

SQLERRMC

This string field holds the message text for the error code stored in SQLCODE and can store up to 70 characters. For the full text of messages longer than 70 characters, use the SQLGLM function.

Verify SQLCODE is negative

before you reference SQLERRMC. If you reference SQLERRMC when SQLCODE is zero, you get the message text associated with a prior SQL statement.

SQLERRP

This string field is reserved for future use.

SQLERRD

This array of binary integers has six elements. Descriptions of the fields in SQLERRD (called SQLERD in FORTRAN) follow:

SQLERRD(1)

This field is reserved for future use.

SQLERRD(2)

This field is reserved for future use.

SQLERRD(3)

This field holds the number of rows processed by the most recently executed SQL statement. However, if the SQL statement failed, the value of SQLERRD(3) is undefined, with one exception. If the error occurred during an array operation, processing stops at the row that caused the error, so SQLERRD(3) gives the number of rows processed successfully.

The rows-processed count is zeroed after an OPEN statement and incremented after a FETCH statement. For the EXECUTE, INSERT, UPDATE, DELETE, and SELECT INTO statements, the count reflects the number of rows processed successfully. The count does not include rows processed by an update or delete cascade. For example, if 20 rows are deleted because they meet WHERE-clause criteria, and 5 more rows are deleted because they now (after the primary delete) violate column constraints, the count is 20 not 25.

SQLERRD(4)

This field is reserved for future use.

SQLERRD(5)

This field holds an offset that specifies the character position at which a parse error begins in the most recently executed SQL statement. The first character occupies position zero.

SQLERRD(6)

This field is reserved for future use.

SQLWARN

This array of single characters has eight elements. They are used as warning flags. Oracle sets a flag by assigning it a "W" (for warning) character value. The flags warn of exceptional conditions.

For example, a warning flag is set when Oracle assigns a truncated column value to an output host variable.

Also note that, while Figure 8-2 illustrates SQLWARN as an array, it is implemented in Pro*COBOL as a group item with elementary PIC X items named SQLWARN0 through SQLWARN7. The Pro*FORTRAN implementation is composed of the LOGICAL variables, SQLWN0 through SQLWN7.

Descriptions of the fields in SQLWARN follow:

SQLWARN(0)

This flag is set if another warning flag is set.

SQLWARN(1)

This flag is set if a truncated column value was assigned to an output host variable. This applies only to character data. Oracle truncates certain numeric data without setting a warning or returning a negative SQLCODE value.

To find out if a column value was truncated and by how much, check the indicator variable associated with the output host variable. The (positive) integer returned by an indicator variable is the original length of the column value. You can increase the length of the host variable accordingly.

SQLWARN(2)

This flag is set if one or more nulls were ignored in the evaluation of a SQL group function such as AVG, COUNT, or MAX. This behavior is expected because, except for COUNT(*), all group functions ignore nulls. If necessary, you can use the SQL function NVL to temporarily assign values (zeros, for example) to the null column entries.

SQLWARN(3)

This flag is set if the number of columns in a query select list does not equal the number of host variables in the INTO clause of the SELECT or FETCH statement. The number of items returned is the lesser of the two.

SQLWARN(4)

This flag is set if every row in a table was processed by an UPDATE or DELETE statement without a WHERE clause. An update or deletion is called unconditional if no search condition restricts the number of rows processed. Such updates and deletions are unusual, so Oracle sets this warning flag. That way, you can roll back the transaction if necessary

SQLWARN(5)

This flag is set when an EXEC SQL CREATE {PROCEDURE|FUNCTION|PACKAGE|PACKAGE BODY} statement fails because of a PL/SQL compilation error.

SQLWARN(6)

This flag is no longer in use.

SQLWARN(7)

This flag is no longer in use.

SQLEXT

This string field is reserved for future use.

PL/SQL Considerations

When your precompiler program executes an embedded PL/SQL block, not all fields in the SQLCA are set. For example, if the block fetches several rows, the rows-processed count, SQLERRD(3), is set to 1, not the actual number of rows fetched. So, you should rely only on the SQLCODE and SQLERRM fields in the SQLCA after executing a PL/SQL block.

Getting the Full Text of Error Messages

The SQLCA can accommodate error messages up to 70 characters long. To get the full text of longer (or nested) error messages, you need the SQLGLM function. If connected to Oracle, you can call SQLGLM using the syntax

SQLGLM(message_buffer, buffer_size, message_length);

where:

message_buffer

is the text buffer in which you want Oracle to store the error message (Oracle blank-pads to the end of this buffer).

buffer_size

is an integer variable that specifies the maximum size of the buffer in bytes.

message_length

is an integer variable in which Oracle stores the actual length of the error message.

The maximum length of an Oracle error message is 512 characters including the error code, nested messages, and message inserts such as table and column names. The maximum length of an error message returned by SQLGLM depends on the value you specify for buffer_size.

In the following example, you call SQLGLM to get an error message of up to 100 characters in length:

-- declare variables for function call
msg_buffer CHARACTER(100);
buf_size INTEGER;
msg_length INTEGER;
set buf_size = 100;
EXEC SQL WHENEVER SQLERROR DO sql_error;
-- other statements
ROUTINE sql_error
BEGIN
 -- get full text of error message
 SQLGLM(msg_buffer, buf_size, msg_length);
 display contents of msg_buffer;
 exit program with an error
END sql_error;

Notice that SQLGLM is called only when a SQL error has occurred. Always make sure SQLCODE is negative before calling SQLGLM. If you call SQLGLM when SQLCODE is zero, you get the message text associated with a prior SQL statement.

Using the WHENEVER Statement

By default, precompiled programs ignore Oracle error and warning conditions and continue processing if possible. To perform automatic condition checking and error handling, use the WHENEVER statement.

With the WHENEVER statement, you can specify actions to be taken when Oracle detects an error, warning condition, or "not found" condition. These actions include continuing with the next statement, calling a routine, branching to a labeled statement, or stopping.

You code the WHENEVER statement by using the following syntax:

EXEC SQL WHENEVER <condition> <action>;

You can have Oracle automatically check the SQLCA for any of the following conditions.

SQLWARNING

SQLWARN(0) is set because Oracle returned a warning (one of the warning flags, SQLWARN(1) through SQLWARN(7), is also set) or SQLCODE has a positive value other than +1403. For example, SQLWARN(1) is set when Oracle assigns a truncated column value to an output host variable.

Declaring the SQLCA is optional when MODE={ANSI|ANSI14}. To use WHENEVER SQLWARNING, however, you must declare the SQLCA.

SQLERROR

SQLCODE has a negative value because Oracle returned an error.

NOT FOUND

SQLCODE has a value of +1403 (+100 when MODE={ANSI|ANSI14| ANSI13}), because Oracle could not find a row that meets the search condition of a WHERE clause, or a SELECT INTO or FETCH returned no rows. When MODE={ANSI|ANSI14|ANSI13}, +100 is returned to SQLCODE after an INSERT of no rows.

When Oracle detects one of the preceding conditions, you can have your program take any of the following actions.

CONTINUE

Your program continues to run with the next statement if possible. This is the default action, equivalent to not using the WHENEVER statement. You can use it to "turn off" condition checking.

DO

Your program transfers control to an internal routine. When the end of the routine is reached, control transfers to the statement that follows the failed SQL statement.

A routine is any functional program unit that can be invoked, such as a COBOL paragraph or FORTRAN subroutine. In this context, separately compiled programs, such as COBOL subroutines, are not routines.

The usual rules for entering and exiting a routine apply. However, passing parameters to the routine is not allowed. Furthermore, the routine must not return a value.

The parameter routine_call is a host language invocation, as in

EXEC SQL -- COBOL
 WHENEVER <condition> DO PERFORM <paragraph_name> -- COBOL
END-EXEC. -- COBOL

or

EXEC SQL -- FORTRAN
 WHENEVER <condition> DO CALL <subroutine_name> -- FORTRAN

GOTO

Your program branches to a labeled statement.

STOP

Your program stops running and uncommitted work is rolled back.

Be careful. The STOP action displays no messages before logging off Oracle. In Pascal, the STOP action is illegal because Pascal has no equivalent command.

Some Examples

If you want your program to

• go to close_cursor if a "no data found" condition occurs,

• continue with the next statement if a warning occurs, and

• go to error_handler if an error occurs

simply code the following WHENEVER statements before the first executable SQL statement:

EXEC SQL WHENEVER NOT FOUND GOTO close_cursor;
EXEC SQL WHENEVER SQLWARNING CONTINUE;
EXEC SQL WHENEVER SQLERROR GOTO error_handler;

The following Pro*C example uses WHENEVER...DO statements to handle specific errors:

EXEC SQL WHENEVER SQLERROR DO handle_insert_error;
EXEC SQL INSERT INTO EMP (EMPNO, ENAME, DEPTNO)
 VALUES (:emp_number, :emp_name, :dept_number);
EXEC SQL WHENEVER SQLERROR DO handle_delete_error;
EXEC SQL DELETE FROM DEPT WHERE DEPTNO = :dept_number;
...
ROUTINE handle_insert_error;
 BEGIN
 IF sqlca.sqlcode = -1 THEN -- duplicate key value
 ...
 ELSEIF sqlca.sqlcode = -1401 THEN -- value too large
 ...
 ENDIF;
 ...
 END;
ROUTINE handle_delete_error;
 BEGIN
 IF sqlca.sqlerrd(3) = 0 THEN -- no rows deleted
 ...
 ELSE
 ...
 ENDIF;
 ...
 END;
...

Notice how the procedures check variables in the SQLCA to determine a course of action.

Scope

Because WHENEVER is a declarative statement, its scope is positional, not logical. It tests all executable SQL statements that follow it in the source file, not in the flow of program logic. Therefore, you should code the WHENEVER statement before the first executable SQL statement you want to test.

A WHENEVER statement stays in effect until superseded by another WHENEVER statement checking for the same condition.

In the example , the first WHENEVER SQLERROR statement is superseded by a second, and so applies only to the CONNECT statement. The second WHENEVER SQLERROR statement applies to both the UPDATE and DROP statements, despite the flow of control from step1 to step3.

step1:
EXEC SQL WHENEVER SQLERROR STOP;
EXEC SQL CONNECT :username IDENTIFIED BY :password;
 ...
 GOTO step3;
step2:
EXEC SQL WHENEVER SQLERROR CONTINUE;
EXEC SQL UPDATE EMP SET SAL = SAL * 1.10; 
 ...
step3:
 EXEC SQL DROP INDEX EMP_INDEX;
 ...

Guidelines

The following guidelines will help you avoid some common pitfalls.

Placing the Statements. In general, code a WHENEVER statement before the first executable SQL statement in your program. This ensures that all ensuing errors are trapped because WHENEVER statements stay in effect to the end of a file.

Handling End-of-Data Conditions. Your program should be prepared to handle an end-of-data condition when using a cursor to fetch rows. If a FETCH returns no data, the program should branch to a labeled section of code where a CLOSE command is issued, as follows:

SQL WHENEVER NOT FOUND GOTO no_more;
...
no_more:
 ...
 EXEC SQL CLOSE my_cursor;
 ...

Avoiding Infinite Loops. If a WHENEVER SQLERROR GOTO statement branches to an error handling routine that includes an executable SQL statement, your program might enter an infinite loop if the SQL statement fails with an error. You can avoid this by coding WHENEVER SQLERROR CONTINUE before the SQL statement, as shown in the following example:

EXEC SQL WHENEVER SQLERROR GOTO sql_error;
...
sql_error:
 EXEC SQL WHENEVER SQLERROR CONTINUE;
 EXEC SQL ROLLBACK WORK RELEASE;
 ...

Without the WHENEVER SQLERROR CONTINUE statement, a ROLLBACK error would invoke the routine again, starting an infinite loop.

Careless use of WHENEVER can cause problems. For example, the following code enters an infinite loop if the DELETE statement sets NOT FOUND because no rows meet the search condition:

-- improper use of WHENEVER
...
EXEC SQL WHENEVER NOT FOUND GOTO no_more;
LOOP
 EXEC SQL FETCH emp_cursor INTO :emp_name, :salary;
 ...
ENDLOOP;
no_more:
 EXEC SQL DELETE FROM EMP WHERE EMPNO = :emp_number;
 ...

In the next example, you handle the NOT FOUND condition properly by resetting the GOTO target:

-- proper use of WHENEVER
...
EXEC SQL WHENEVER NOT FOUND GOTO no_more;
LOOP
 EXEC SQL FETCH emp_cursor INTO :emp_name, :salary;
 ...
ENDLOOP;
no_more:
 EXEC SQL WHENEVER NOT FOUND GOTO no_match;
 EXEC SQL DELETE FROM EMP WHERE EMPNO = :emp_number;
 ...
no_match:
 ...

Maintaining Addressability. With host languages that allow local and global identifiers, make sure all SQL statements governed by a WHENEVER GOTO statement can branch to the GOTO label. The following code results in a compile-time error because labelA in FUNC1 is not within the scope of the INSERT statement in FUNC2:

FUNC1
 BEGIN
 EXEC SQL WHENEVER SQLERROR GOTO labelA;
 EXEC SQL DELETE FROM EMP WHERE DEPTNO = :dept_number;
 ...
 labelA:
 ...
 END;
FUNC2
 BEGIN
 EXEC SQL INSERT INTO EMP (JOB) VALUES (:job_title); 
 ...
 END;

The label to which a WHENEVER GOTO statement branches must be in the same precompilation file as the statement.

Returning after an Error. If your program must return after handling an error, use the DO routine_call action. Alternatively, you can test the value of SQLCODE, as shown in the following example:

EXEC SQL UPDATE EMP SET SAL = SAL * 1.10;
IF sqlca.sqlcode < 0 THEN
 -- handle error
EXEC SQL DROP INDEX EMP_INDEX;
...

Just make sure no WHENEVER GOTO or WHENEVER STOP statement is active.

Getting the Text of SQL Statements

In many precompiler applications, it is convenient to know the text of the statement being processed, its length, and the SQL command (such as INSERT or SELECT) that it contains. This is especially true for applications that use dynamic SQL.

The routine SQLGLS, which is part of the SQLLIB run-time library, returns the following information:

• The text of the most recently parsed SQL statement

• The Length of the statement

• A Function code (refer to Table 8-8 for the SQL command used in the statement

You can call SQLGLS after issuing a static SQL statement. With dynamic SQL Method 1, you can call SQLGLS after the SQL statement is executed. With dynamic SQL Method 2, 3, or 4, you can call SQLGLS after the statement is prepared.

To call SQLGLS, you use the following syntax:

SQLGLS(SQLSTM, STMLEN, SQLFC)

Table 8-7 shows the host-language datatypes available for the parameters in the SQLGLS argument list.

All parameters must be passed by reference. This is usually the default parameter passing convention; you need not take special action.

The parameter SQLSTM is a blank-padded (not null-terminated) character buffer that holds the returned text of the SQL statement. Your program must statically declare the buffer or dynamically allocate memory for it.

The length parameter STMLEN is a four-byte integer. Before calling SQLGLS, set this parameter to the actual size (in bytes) of the SQLSTM buffer. When SQLGLS returns, the SQLSTM buffer contains the SQL statement text blank padded to the length of the buffer. STMLEN returns the actual number of bytes in the returned statement text, not counting the blank padding. However, STMLEN returns a zero if an error occurred.

Some possible errors follow:

• No SQL statement was parsed.

• You passed an invalid parameter (for example, a negative length value).

• An internal exception occurred in SQLLIB.

The parameter SQLFC is a four-byte integer that returns the SQL function code for the SQL command in the statement. Table 8-8 shows the function code for each SQL command.

SQLGLS does not return statements that contain the following commands:

• CONNECT

• COMMIT

• ROLLBACK

• RELEASE

• FETCH

There are no SQL function codes for these statements.

Declaring the ORACA

To declare the ORACA, simply include it (using an EXEC SQL INCLUDE statement) in your host-language source file as follows:

* Include the Oracle Communications Area (ORACA).
 EXEC SQL INCLUDE ORACA

The ORACA must be declared outside the Declare Section.

When you precompile your program, the INCLUDE ORACA statement is replaced by several program variable declarations. These declarations allow Oracle to communicate with your program.

Enabling the ORACA

To enable the ORACA, you must specify the ORACA option, either on the command line with

ORACA=YES

or inline with

EXEC ORACLE OPTION (ORACA=YES);

Then, you must choose appropriate run-time options by setting flags in the ORACA.

What's in the ORACA?

The ORACA contains option settings, system statistics, and extended diagnostics such as

• SQL statement text (you can specify when to save the text)

• name of the file in which an error occurred

• location of the error in a file

• cursor cache errors and statistics

Figure 8-3 shows all the variables in the ORACA. To see the ORACA structure and variable names in a particular host language, refer to your supplement to this Guide.

Figure 8-3 ORACA Variables

Description of "Figure 8-3 ORACA Variables"

Choosing Run-time Options

The ORACA includes several option flags. Setting these flags by assigning them nonzero values enables

• Save the text of SQL statements

• Enable DEBUG operations

• Check cursor cache consistency (the cursor cache is a continuously updated area of memory used for cursor management)

• Check heap consistency (the heap is an area of memory reserved for dynamic variables)

• Gather cursor statistics

The descriptions will help you choose the options you need.

ORACA Structure

This section describes the structure of the ORACA, its fields, and the values they can store.

ORACAID

This string field is initialized to "ORACA" to identify the Oracle Communications Area.

ORACABC

This integer field holds the length, expressed in bytes, of the ORACA data structure.

ORACCHF

If the master DEBUG flag (ORADBGF) is set, this flag lets you check the cursor cache for consistency before every cursor operation.

The Oracle run-time library does the consistency checking and might issue error messages, which are listed in Oracle Database Error Messages. They are returned to the SQLCA just like Oracle error messages.

This flag has the following settings:

0

Disable cache consistency checking (the default).

1

Enable cache consistency checking.

ORADBGF

This master flag lets you choose all the DEBUG options. It has the following settings:

0

Disable all DEBUG operations (the default).

1

Enable all DEBUG operations.

ORAHCHF

If the master DEBUG flag (ORADBGF) is set, this flag tells the Oracle run-time library to check the heap for consistency every time the precompiler dynamically allocates or frees memory. This is useful for detecting program bugs that upset memory.

This flag must be set before the CONNECT command is issued and, once set, cannot be cleared; subsequent change requests are ignored. It has the following settings:

0

Disable heap consistency checking (the default).

1

Enable heap consistency checking.

ORASTXTF

This flag lets you specify when the text of the current SQL statement is saved. It has the following settings:

0

Never save the SQL statement text (the default).

1

Save the SQL statement text on SQLERROR only.

2

Save the SQL statement text on SQLERROR or SQLWARNING.

3

Always save the SQL statement text.

The SQL statement text is saved in the ORACA subrecord named ORASTXT.

Diagnostics

The ORACA provides an enhanced set of diagnostics; the following variables help you to locate errors quickly.

ORASTXT

This subrecord helps you find faulty SQL statements. It lets you save the text of the last SQL statement parsed by Oracle. It contains the following two fields:

ORASTXTL

This integer field holds the length of the current SQL statement.

ORASTXTC

This string field holds the text of the current SQL statement. At most, the first 70 characters of text are saved.

Statements parsed by the precompiler, such as CONNECT, FETCH, and COMMIT, are not saved in the ORACA.

ORASFNM

This subrecord identifies the file containing the current SQL statement and so helps you find errors when multiple files are precompiled for one application. It contains the following two fields:

ORASFNML

This integer field holds the length of the filename stored in ORASFNMC.

ORASFNMC

This string field holds the filename. At most, the first 70 characters are stored.

ORASLNR

This integer field identifies the line at (or near) which the current SQL statement can be found.

Cursor Cache Statistics

The variables let you gather cursor cache statistics. They are automatically set by every COMMIT or ROLLBACK statement your program issues. Internally, there is a set of these variables for each CONNECTed database. The current values in the ORACA pertain to the database against which the last commit or rollback was executed.

ORAHOC

This integer field records the highest value to which MAXOPENCURSORS was set during program execution.

ORAMOC

This integer field records the maximum number of open Oracle cursors required by your program. This number can be higher than ORAHOC if MAXOPENCURSORS was set too low, which forced the precompiler to extend the cursor cache.

ORACOC

This integer field records the current number of open Oracle cursors required by your program.

ORANOR

This integer field records the number of cursor cache reassignments required by your program. This number shows the degree of "thrashing" in the cursor cache and should be kept as low as possible.

ORANPR

This integer field records the number of SQL statement parses required by your program.

ORANEX

This integer field records the number of SQL statement executions required by your program. The ratio of this number to the ORANPR number should be kept as high as possible. In other words, avoid unnecessary reparsing. For help, refer to Appendix C, "Performance Tuning".

An Example

The following program prompts for a department number, inserts the name and salary of each employee in that department into one of two tables, then displays diagnostic information from the ORACA:

EXEC SQL BEGIN DECLARE SECTION; 
 username CHARACTER(20);
 password CHARACTER(20);
 emp_name INTEGER;
 dept_number INTEGER;
 salary REAL;
EXEC SQL END DECLARE SECTION; 
EXEC SQL INCLUDE SQLCA;
EXEC SQL INCLUDE ORACA;
display 'Username? ';
read username;
display 'Password? ';
read password;
EXEC SQL WHENEVER SQLERROR DO sql_error;
EXEC SQL CONNECT :username IDENTIFIED BY :password;
display 'Connected to Oracle';
EXEC ORACLE OPTION (ORACA=YES);
-- set flags in the ORACA
set oraca.oradbgf = 1; -- enable debug operations
set oraca.oracchf = 1; -- enable cursor cache consistency check
set oraca.orastxtf = 3; -- always save the SQL statement
display 'Department number? ';
read dept_number;
EXEC SQL DECLARE emp_cursor CURSOR FOR
 SELECT ENAME, SAL + NVL(COMM,0)
 FROM EMP
 WHERE DEPTNO = :dept_number;
EXEC SQL OPEN emp_cursor;
EXEC SQL WHENEVER NOT FOUND DO no_more; 
rLOOP
 EXEC SQL FETCH emp_cursor INTO :emp_name, :salary;
 IF salary < 2500 THEN
 EXEC SQL INSERT INTO PAY1 VALUES (:emp_name, :salary);
 ELSE
 EXEC SQL INSERT INTO PAY2 VALUES (:emp_name, :salary);
 ENDIF;
ENDLOOP;
ROUTINE no_more
BEGIN
 EXEC SQL CLOSE emp_cursor;
 EXEC SQL WHENEVER SQLERROR CONTINUE;
 EXEC SQL COMMIT WORK RELEASE;
 display 'Last SQL statement: ', oraca.orastxt.orastxtc;
 display '... at or near line number: ', oraca.oraslnr;
 display
 display ' Cursor Cache Statistics';
 display '-------------------------------------------';
 display 'Maximum value of MAXOPENCURSORS ', oraca.orahoc;
 display 'Maximum open cursors required: ', oraca.oramoc;
 display 'Current number of open cursors: ', oraca.oracoc;
 display 'Number of cache reassignments: ', oraca.oranor;
 display 'Number of SQL statement parses: ', oraca.oranpr;
 display 'Number of SQL statement executions: ', oraca.oranex;
 exit program;
END no_more;
ROUTINE sql_error
BEGIN
 EXEC SQL WHENEVER SQLERROR CONTINUE;
 EXEC SQL ROLLBACK WORK RELEASE;
 display 'Last SQL statement: ', oraca.orastxt.orastxtc;
 display '... at or near line number: ', oraca.oraslnr;
 display
 display ' Cursor Cache Statistics';
 display '-------------------------------------------';
 display 'Maximum value of MAXOPENCURSORS ', oraca.orahoc;
 display 'Maximum open cursors required: ', oraca.oramoc;
 display 'Current number of open cursors: ', oraca.oracoc;
 display 'Number of cache reassignments: ', oraca.oranor;
 display 'Number of SQL statement parses: ', oraca.oranpr;
 display 'Number of SQL statement executions: ', oraca.oranex;
 exit program with an error;
END sql_error;