Chapter 39 – Embedded SQL Binding Style

Note

You are reading a digital copy of SQL-99 Complete, Really, a book that documents the SQL-99 standard.

The book and the standard does not reflect the features of CrateDB, we are just publishing it as a service for the community and for reference purposes.

On the one hand, CrateDB does not implement the SQL-99 standard thoroughly, on the other hand, it extends the standard for implementing cluster features and others.

For more information specific to CrateDB, check out the CrateDB Reference documentation.

SQL DBMSs communicate with SQL applications through a common programming language interface that is invoked through one of the SQL Standard-defined binding styles, or interface options. The programming language used – either Ada, C, COBOL, Fortran, MUMPS, Pascal or PL/I – is called a host language. An SQL DBMS must support the use of at least one host language, either for embedded SQL programming or for invoking external routines and/or procedures. Depending on your DBMS then, you can use SQL to set up and maintain SQL-data from an application program in one of two ways: by incorporating SQL statements or standard routines in a specifically designed programming module or by embedding SQL statements directly into your program.

Table of Contents

Binding Styles

The SQL Standard defines these binding styles, at least one of which must be supported by an SQL DBMS:

Module Binding Style

The SQL-client Module binding style is defined in SQL/Foundation. In this binding style, SQL-client Modules are run, executing SQL operations. One special version of this binding style is known as the Direct SQL Invocation binding style, where SQL statements are directly executed through a front-end. (The SQL statements are gathered together into a default SQL-session Module whose existence is usually not apparent to the user.) Direct SQL is defined in SQL/Bindings. Another special version of the Module binding style is known as the SQL/CLI binding style, where SQL-data operations are performed when SQL applications invoke one of a number of standard routines, passing appropriate arguments. (The routines are gathered together into a default SQL-server Module whose existence is usually not apparent to the user.) The SQL/CLI binding style is defined in SQL/CLI.

Embedded SQL Binding Style

The embedded SQL binding style is defined in SQL/Bindings. In this binding style, SQL statements are directly coded into a host language program and some method (usually a precompiler)is then used to (a) generate an externally-invoked procedure from each embedded SQL statement and (b) replace each embedded SQL statement with an invocation of that generated procedure. (The generated procedures are collected together into a programming module which can then be used as an SQL-client Module by your DBMS, so this ia also a special version of the Module binding style.)

There are thus three main approaches to writing complete programs with SQL:

  1. With embedded SQL, you can put SQL statements directly into “host programs” (programs written in Ada, C, COBOL, FORTRAN, MUMPS, Pascal or PL/I) – making them part of the host program’s source code. Since the host language’s compiler won’t recognize SQL statements as valid statements of its language, some sort of preprocessor is required to make this work.

  2. With SQL/CLI, you can call a SQL DBMS’s library from a host program. The SQL statements you want to execute are parameters of the call. Because SQL’s modus operandi won’t quite match the host language’s, helper functions are required for the interface definition.

  3. With the Module language, you can dispense with host programs and write entire Modules in SQL. You’ll still have to call these Modules from one of the standard host languages though.

There is one other binding style — Direct SQL — which involves no host language. The entry of SQL statements on a keyboard, for example, is Direct SQL. Direct SQL is, under the covers, merely an implementation of a program written with one of the main three binding styles (Module, Embedded, or SQL/CLI).

Reflecting what we believe is their relative importance, this book contains one chapter on embedded SQL, then several chapters on SQL/CLI, then one chapter on th Module binding style. This is the embedded SQL chapter.

Because we believe that SQL/CLI will quickly become the SQL interface of choice, this chapter omits large amounts of detail, in favour of providing the necessary detail in our chapters on SQL/CLI. Still, this chapter is a good opportunity to introduce the basic concepts in a simple way. Embedded SQL is easy to read, so we have a trouble-free first viewing of the problems that a program writer must solve. Watch in particular for solutions of the “impedance mismatch” problem (getting SQL objects to link with host-language analogs), the “Weltanschauung” problem (handling one-row-at-a-time rather than whole-set-at-a-time) and the “control” problem (adjusting host program flow based on SQL execution information).

What is embedded SQL?

Embedded SQL was once the predominant standard way to mix SQL statements with host languages. It lets you mix SQL statements directly into an application program written in some common computer programming language. It is especially associated with COBOL or PL/I programs and IBM’s DB2 and big iron; however. Most of the big DBMS vendors support it on microcomputers too. Support is weak among small DBMS vendors, and especially weak for computer host languages that aren’t currently in vogue.

SQL can be embedded into many host languages, but support varies depending on the vendor and depending on the language. The following are the standard host languages (that is, the ones mentioned in the SQL Standard). A DBMS that supports the embedded SQL binding style must support SQL embedded into at least one of these languages:

Ada: Standard. Weakly supported.

C: Standard. Well supported.

COBOL: Standard. Packages are sometimes supplied by the COBOL vendor. For example, Micro Focus offers XDB SQL as an optional add-on, along with appropriate interfacing.

Fortran: Standard. Well supported, but interfacing is sometimes awkward.

MUMPS: Standard. Weakly supported.

Pascal: Standard. Not everybody supports Borland Delphi’s special characteristics such as Pchar null-terminated string references.

PL/I: Standard. Weakly supported.

Here are some other host languages that have some importance but are not (yet) standard:

BASIC: Not supported by the SQL Standard. The vendor support which exists is usually restricted to a single dialect (such as Microsoft BASIC or PowerBASIC).

Java: Not supported by the SQL Standard. Probably Java will achieve recognition as an “official” host language for SQL, but most current efforts are for a CLI standard called JDBC.

For all of these languages, support is better via the SQL/CLI.

So, to begin with, you start with a host language, writing your code in the normal way. When you get to a point where a database operation needs to be carried out, use embedded SQL statements (rather than routines written in the host language; an executable SQL statement can be put into a program anywhere that an executable host language statement can be) to carry them out – the host code “calls” SQL statements. You can pass data between the host and SQL: SQL-data values go to variables in the host code, host program values go to SQL Columns or functions or some other appropriate database target. Before compiling your program, you’ll need to go through a preprocessing step – hence the name: Precompiler.

To make it all possible, you have to follow rigid conventions:

  • Begin every SQL statement with the SQL prefix appropriate for the host language.

  • End every SQL statement with the SQL terminator appropriate for the host language.

  • Declare all host-language program variables which will be shared with SQL in a special DECLARE section.

  • Declare additional program variables for error handling.

Precompilers

Most precompilers are standalone utility programs. A few are integrated with the compiler itself. Regardless of their startup method, all precompilers must do a certain series of operations which turn an embedded-SQL module into a module that the host compiler can compile. These are the operations which the sample Precompiler that comes with this book does for a C program:

  1. Open the Input File (the file containing embedded SQL).

    Open the Output File (the file which the compiler needs).

  2. Read a token from the Input File. At end: exit.

  3. If the token is not EXEC, or the next token is not SQL: Write the token to the output file. Goto 2.

  4. If the next tokens are BEGIN DECLARE SECTION:

/* The following tokens must be variable definitions */

Read and write the following tokens. While doing so, keep track of and store: variable name, variable size, variable type. Since this is supposed to be a C program, certain keywords are expected, such as char or short. Stop when the next tokens are END DECLARE SECTION.

Goto 2.

  1. If the next token is CREATE or INSERT or some other SQL <keyword> indicating that this is the beginning of an executable SQL statement:

Write the words SQLExecDirect(-1,.

Read and write until “;” is seen. (“;” is the “terminator” of SQL in C.)

Write ,-3).

Goto 2.

NB: While reading and writing, a “host variable” token may be encountered, of the form: <colon><host-variable-name>. The precompiler must convert this to whatever the appropriate form is for a pass-by-address parameter, and output a statement like SQLBindCol(...,&hostvariable);.

If the precompiler finishes without severe errors, there is an output program. The output from the precompiler is the input for the compiler. From that, the compiler produces an object file. Then the linker comes into play. The linker will encounter some external references, for example the SQLExecDirect call that the precompiler produced. To resolve these references, the linker will look in a library – which is another essential part of the DBMS package for use with precompilers.

In theory, a precompiler could act differently provided it met the functional specification. We have described what we know best: this book’s sample Precompiler.

SQL Prefixes and Terminators

The precompiler lacks the smarts to figure out the syntax of the host language program. After all, host language code can get pretty complex. All the precompiler can do is look for signals that say “embedded SQL starts here” or “embedded SQL ends here”. Or, to use the formal terms, embedded SQL statements must have a prefix and a terminator. Usually the prefix is EXEC SQL and the terminator is a semicolon. There is a bit of variation among the standard host languages:

  • For Ada, use: EXEC SQL ... ;

  • For C, use: EXEC SQL ... ;

  • For COBOL, use: EXEC SQL ... END-EXEC

  • For Fortran, use: EXEC SQL ... <no end> (i.e.: no explicit terminator)

  • For MUMPS, use: &SQL( ... )

  • For Pascal, use: EXEC SQL ... ;

  • For PL/I, use: EXEC SQL ... ;

Our preferred languages for discussion are C and Pascal, so it’s just as easy to say: “all SQL statements must be prefixed by 'EXEC SQL' and terminated by ‘;’”. The case of the <keyword>s EXEC SQL (or the other prefixes) doesn’t matter, but it is compulsory that they be together (rather than on separate lines). Everything between EXEC SQL and ; must be legal SQL – host language statements or comments have no place here. Aside from that, format is fairly free, as this example snippet shows:

...
{                    /* Braces around SQL statements are a good idea. */
EXEC sql
   CREATE DOMAIN d5 INTEGER;
   EXEC SQL
   CREATE DOMAIN     /* This is a new-style comment not a C comment */
   d6 INTEGER
;
}
...

Host Variables

Embedded SQL statements can contain host language variables in the same places that any other scalar expression can be placed. The host variables allow you to pass data between the program and your SQL-data. In an SQL statement, a <host variable name> must be preceded by a colon, to distinguish it from an SQL Object name. Host variables may not be qualified or subscripted, and must return only scalar values.

An important and inevitable part of any embedded SQL program is the declare section – the variable declarations that appear between EXEC SQL BEGIN DECLARE SECTION; and EXEC SQL END DECLARE SECTION;, as in:

EXEC SQL BEGIN DECLARE SECTION;
   int   x;
   char  y[5];
EXEC SQL END DECLARE SECTION;

In this example, x and y are host variables. They are defined according to the rules of the host language (in this case, C). But the definitions must be comprehensible to the precompiler too, because the precompiler must know the data type, size and name of each host variable. The trick, for making a variable that both the host and SQL can understand, is to keep the definition simple. Only a few data types are legal, and tricks usually aren’t – even simple tricks like enumerations or constants or macros will confuse the precompiler.

SQL <data type> and Host Variable Correspondence

Host variable data types must be compatible with the <data type> of the SQL- data values they’ll be receiving or sending. The appropriate data type analogs for each host language, as defined by the SQL Standard, are as follows (L stands for length, P for precision, S for scale, T for time fractional seconds precision, Q for <interval qualifier> and N for the implementation-defined size of a structured type reference; assume an 8-bit character set is in use):

For Ada:

SQL <data type>

Ada data type

SQLSTATE

SQL_STANDARD.SQLSTATE_TYPE

CHAR(L)

SQL_STANDARD.CHAR, with P'LENGTH of L

VARCHAR(L)

None

CLOB(L)

None

BIT(L)

SQL_STANDARD.BIT, with P'LENGTH of L

BIT VARYING(L)

None

BLOB(L)

None

BOOLEAN

SQL_STANDARD.BOOLEAN

SMALLINT

SQL_STANDARD.SMALLINT

INTEGER

SQL_STANDARD.INT

DECIMAL(P,S)

None

NUMERIC(P,S)

None

REAL

SQL_STANDARD.REAL

DOUBLE PRECISION

SQL_STANDARD.DOUBLE_PRECISION

FLOAT(P)

None

DATE

None

TIME(T)

None

TIMESTAMP(T)

None

INTERVAL(Q)

None

UDT

None

REF

SQL_STANDARD.CHAR, with P'LENGTH of N

For C:

SQL <data type>

C Data Type

SQLSTATE

char, with length 6

CHAR(L)

char, with length (L+1)

VARCHAR(L)

char, with length (L+1)

CLOB(L)

struct {
long x_reserved
unsigned long x_length
char x_data[L];
}

BIT(L)

char, with length L/8

BIT VARYING(L)``

None

BLOB(L)``

struct {
long x_reserved
unsigned long x_length
char x_data[L];
}

BOOLEAN

pointer to long

SMALLINT

pointer to short

INTEGER

pointer to long

DECIMAL(P,S)

None

NUMERIC(P,S)

None

REAL

pointer to float

DOUBLE PRECISION

pointer to double

FLOAT(P)

None

DATE

None

TIME(T)

None

TIMESTAMP(T)

None

INTERVAL(Q)

None

UDT

None

REF

char, with length N

ARRAY

None

ROW

None

For COBOL:

SQL <data type>``

COBOL Data Type

SQLSTATE

PICTURE X(5)

CHAR(L)

alphanumeric, with length L

VARCHAR(L)

None

CLOB(L)

01 XXXX.
49 XXXX-RESERVED PIC S9(9) USAGE IS BINARY.
49 XXXX-LENGTH PIC S9(9) USAGE IS BINARY.
49 XXXX-DATA PIC X(L).

BIT(L)

alphanumeric, with length L/8+1

BIT VARYING(L)

None

BLOB(L)

01 XXXX.
49 XXXX-RESERVED PIC S9(9) USAGE IS BINARY.
49 XXXX-LENGTH PIC S9(9) USAGE IS BINARY.
49 XXXX-DATA PIC X(L).

BOOLEAN

PICTURE X

SMALLINT

PICTURE S9(SPI) USAGE BINARY, where SPI is implementation-defined

INTEGER

PICTURE S9(PI) USAGE BINARY, where PI is implementation-defined

DECIMAL(P,S)

None

NUMERIC(P,S)

USAGE DISPLAY SIGN LEADING SEPARATE, with PICTURE as specified [note]

REAL

None

DOUBLE PRECISION

None

FLOAT(P)

None

DATE

None

TIME(T)

None

TIMESTAMP(T)

None

INTERVAL(Q)

None

UDT

None

REF

alphanumeric with length N

ARRAY

None

ROW

None

Note

If S=P, then a PICTURE with an 'S' followed by a 'V' followed by P '9's. If P>S>0, then a PICTURE with an 'S' followed by P-S '9's followed by a 'V' followed by S '9's. If S=0, then a PICTURE with an 'S' followed by P '9's optionally followed by a 'V'.

For Fortran:

SQL <data type>

Fortran Data Type

SQLSTATE

CHARACTER, with length 5

CHAR(L)

CHARACTER, with length L

VARCHAR(L)

None

CLOB(L)

CHARACTER XXXX(L+8)
INTEGER*4 XXXX_RESERVED
INTEGER*4 XXXX_LENGTH
CHARACTER XXXX_DATA
EQUIVALENCE(XXXX(5), XXXX_LENGTH)
EQUIVALENCE(XXXX(9), XXXX_DATA)

BIT(L)

CHARACTER, with length L/8+1

BIT VARYING(L)

None

BLOB(L)

CHARACTER XXXX(L+8)
INTEGER*4 XXXX_RESERVED
INTEGER*4 XXXX_LENGTH
CHARACTER XXXX_DATA
EQUIVALENCE(XXXX(5), XXXX_LENGTH)
EQUIVALENCE(XXXX(9), XXXX_DATA)

BOOLEAN

LOGICAL

SMALLINT

None

INTEGER

INTEGER

DECIMAL(P,S)

None

NUMERIC(P,S)

None

REAL

REAL

DOUBLE PRECISION

DOUBLE PRECISION

FLOAT(P)

None

DATE

None

TIME(T)

None

TIMESTAMP(T)

None

INTERVAL(Q)

None

UDT

None

REF

CHARACTER with length N

ARRAY

None

ROW

None

For MUMPS:

SQL <data type>

MUMPS Data Type

SQLSTATE

character, with maximum length at least 5

CHAR(L)

None

VARCHAR(L)

character with maximum length L

CLOB(L)

None

BIT(L)

None

BIT VARYING(L)

None

BLOB(L)

None

BOOLEAN

None

SMALLINT

None

INTEGER

character

DECIMAL(P,S)

character

NUMERIC(P,S)

character

REAL

character

DOUBLE PRECISION

None

FLOAT(P)

None

DATE

None

TIME(T)

None

TIMESTAMP(T)

None

INTERVAL(Q)

None

UDT

None

REF

character

ARRAY

None

ROW

None

For Pascal:

SQL <data type>

Pascal Data Type

SQLSTATE

PACKED ARRAY [1..5] OF CHAR

CHARACTER(1)

CHAR

CHAR(L), L>1

PACKED ARRAY [1..L] OF CHAR

VARCHAR(L)

None

CLOB(L)

None

BIT(L)

PACKED ARRAY [L/8] OF CHAR

BIT VARYING(L)

None

BLOB(L)

None

BOOLEAN

BOOLEAN

SMALLINT

None

INTEGER

INTEGER

DECIMAL(P,S)

None

NUMERIC(P,S)

None

REAL

REAL

DOUBLE PRECISION

None

FLOAT(P)

None

DATE

None

TIME(T)

None

TIMESTAMP(T)

None

INTERVAL(Q)

None

UDT

None

REF

PACKED ARRAY[1..N] OF CHAR

ARRAY

None

ROW

None

For PL/I:

SQL <data type>

PL/I Data Type

SQLSTATE

CHARACTER(5)

CHAR(L)

CHARACTER(L)

VARCHAR(L)

CHARACTER VARYING(L)

CLOB(L)

DCL level. lvchar
49 len1 FIXED BINARY (31)
49 len2 FIXED BINARY (31)
49 data CHAR (n)

BIT(L)

BIT(L)

BIT VARYING(L)

BIT VARYING(L)

BLOB(L)

DCL level. lvchar
49 len1 FIXED BINARY (31)
49 len2 FIXED BINARY (31)
49 data CHAR (n)

BOOLEAN

BIT(1)

SMALLINT

FIXED BINARY(SPI), where SPI is
implementation-defined

INTEGER

FIXED BINARY(PI), where PI is
implementation-defined

DECIMAL(P,S)

FIXED DECIMAL(P,S)

NUMERIC(P,S)

None

REAL

None

DOUBLE PRECISION

None

FLOAT(P)

FLOAT BINARY (P)

DATE

None

TIME(T)

None

TIMESTAMP(T)

None

INTERVAL(Q)

None

UDT

None

REF

CHARACTER VARYING(N)

ARRAY

None

ROW

None

Input and Output Variables

A host variable can be either an input host variable or an output host variable. (We always speak from the perspective of the DBMS, so “input” means “input to the DBMS” and “output” means “output from the DBMS”.)

Here is a set of embedded SQL and C statements; x and y appear as input host variables:

EXEC SQL CREATE TABLE Table_1 (
            column_1 INT, column_2 CHAR(4));

x = 1; strcpy(y,"1234");

EXEC SQL INSERT INTO Table_1
         VALUES (:x,:y);

EXEC SQL DELETE FROM Table_1
         WHERE column_2 = :y;

EXEC SQL UPDATE Table_1 SET
            column_2 = :x;           /* ERROR! */

In each case, it’s easy to see the host variables – they’re the names with colons in front. In the last SQL statement in the example, the use of :x is an error, because x is an “int” and COLUMN_2 has a CHAR <data type>. Usually the precompiler will catch that type of error – if it doesn’t, there will be an error message at runtime. The subject of data type compatibility is a tortuous one, but if you paid attention while reading our chapters on the various SQL predefined <data type>s, it should hold no terrors for you. Nevertheless, it is difficult to write language-independent embedded SQL code – inevitably, one has the host language’s data type in mind when putting together the SQL code.

Indicator Variables

No host language can handle NULLs. So how can a host program receive an SQL output host variable with a NULL value? And how can a host program pass an input host variable with a NULL value? In both cases, the answer is that you have to pass two parameters by address: one to the data itself and one to a numeric variable containing a flag for “is NULL” or “is not NULL”. This numeric variable is called an indicator variable.

Indicator variables are “signed numeric with scale zero”. In practice, they are almost always 32-bit integers. They too should appear within a declare section. For example:

EXEC SQL BEGIN DECLARE SECTION;
   var host : Integer;           {Variable, intended for use as data}
   var host_indicator : Longint; {Variable, intended for use as indicator}
EXEC SQL END DECLARE SECTION;

In an embedded SQL statement, indicators follow host variables. The required syntax is:

:<host variable name> [ [ INDICATOR ] :<indicator name> ]

Here are two equivalent examples:

EXEC SQL INSERT INTO Table_1
         VALUES (:host INDICATOR :host_indicator);

EXEC SQL INSERT INTO Table_1
         VALUES (:host :host_indicator);

For input host variables: any indicator value less than zero means the value passed was NULL and any indicator value greater than or equal to zero means the value passed was a non-null value. For output host variables, the DBMS uses the specific indicator value -1 to mean the value passed was NULL and the specific value zero to mean the value passed was a non-null value. If the indicator variable value means NULL, then the host variable’s contents are irrelevant.

Use of indicators is optional. There is no need for them if there is no chance that NULL values will occur. However, since “nullability” is a transient characteristic, many programmers use indicators as a matter of course after every host variable.

Here is an embedded SQL statement in which x appears as an output host variable:

EXEC SQL SELECT column_1 INTO :x FROM Table_1;

but it’s not a particularly good example, because it only works if you know in advance that there’s a maximum of one row in TABLE_1. (Think: if there were two rows, there would be two values for COLUMN_1, but there’s only one host variable for the output: x.) Because of this assumption that there will never be more than one row retrieved, this form of SELECT is referred to as a “singleton SELECT”. A more flexible approach exists for getting output SELECT results, as output host variables, from SQL to the host language: the SQL Cursor.

Cursors

The most famous of the impedance-mismatch problems is that SQL operates on sets, while host languages operate on set members. Therefore an SQL query – by which we almost always mean a SELECT statement – is returning more data than a host variable can store. It’s not enough to reply “use an array then”. Although sometimes arrays are indeed helpful, we have to keep in mind that (a) the set might be gigantic and (b) there might be a two-way requirement (that is, rows which go out might also come back in). Either of these considerations make array definitions inappropriate for a general solution. The true and standard solution is the Cursor. With a Cursor, you can get only one row at a time – namely the row that the Cursor is “positioned on”.

To understand the role of a Cursor, remember that your DBMS builds a result Table that contains all of the rows retrieved by an SQL query executed in an application program. Your DBMS uses a Cursor to make the rows of the result Table available to the program: the Cursor identifies, or points to, the current row of the result Table. When a Cursor is pointing to a row, it is said to be positioned on that row – and you can UPDATE or DELETE that row using the “positioned” forms of the UPDATE and DELETE statements.

To get it all working, you need to do four things: DECLARE a Cursor, OPEN the Cursor, FETCH (repeatedly) from the Cursor (manipulating the rows one by one) and CLOSE the Cursor. This Pascal program does all four things, in the form of a classic FETCH loop which is familiar to all SQL programmers.

{ Pascal program, using fetch loop }
var
EXEC SQL BEGIN DECLARE SECTION;
   sample_integer : Integer;
   sample_boolean : Boolean;
EXEC SQL END DECLARE SECTION;
j : Integer;

begin
  EXEC SQL CONNECT TO DEFAULT;
  EXEC SQL DECLARE sample_cursor CURSOR FOR SELECT * FROM sample_table;
  EXEC SQL OPEN sample_cursor;
  for j := 1 to 100 do begin
    EXEC SQL FETCH sample_cursor INTO :sample_integer, :sample_boolean;
    end;
  EXEC SQL CLOSE sample_cursor;
  EXEC SQL DISCONNECT DEFAULT;
end.

The DECLARE CURSOR statement defines the query (usually a SELECT statement) that will get the SQL-data you want. It is not an executable statement, it is only (as the name implies) declarative. The action begins with the OPEN statement. When OPEN happens, the SELECT actually takes place, so now the DBMS has a set of rows waiting to be fetched – these rows are called the current active set, or the result Table. The FETCH statement is the host program’s way of asking the DBMS “may I have the next record please”. In this example we have assumed that the DBMS will always answer YES and put values into the host variables SAMPLE_INTEGER and SAMPLE_BOOLEAN – realistically, of course, we should also check whether the FETCH in fact succeeds. This is a check which we’ll perform in a later example. Anyway, once the loop is done, the CLOSE statement will throw out the set of rows that the OPEN statement produced, thus closing the Cursor.

Some of the things you can do with Cursors, besides merely FETCH from them, include:

  • DELETE and UPDATE rows at the current Cursor position.

  • FETCH something other than the “next” row (this option is called the SCROLL Cursor option).

DECLARE CURSOR Statement

The DECLARE CURSOR statement defines a Cursor. The required syntax for the DECLARE CURSOR statement is:

DECLARE <Cursor name> [ <cursor sensitivity> ]
   [ SCROLL ] CURSOR [ WITH HOLD ] [ WITH RETURN ]
   FOR <cursor specification>

   <cursor sensitivity> ::=
   SENSITIVE |
   INSENSITIVE |
   ASENSITIVE

   <cursor specification> ::=
   <query expression> [ <order by clause> ]
   [ FOR {READ ONLY | UPDATE [ OF <Column name> list ]} ]

Each Cursor you define must have a unique name in the Module it’s defined in. Its <cursor specification> results in a Table when evaluated. An INSENSITIVE Cursor is a Cursor that effectively causes a separate copy of its result Table to be created; the Cursor accesses that copy, rather than the original result, so any changes made to the original result by other methods won’t be visible to this Cursor. A SENSITIVE Cursor is a Cursor that works directly on its result Table: it makes no copy, so other changes made to the result Table will be visible to this Cursor. An ASENSITIVE Cursor may or may not make a copy of its result Table; whether other changes to its result Table will be visible is implementation-defined. The default is an ASENSITIVE Cursor.

Normally, you can access the Cursor’s result Table, one row at a time, only in the order that your DBMS gets the rows. A SCROLL Cursor is a Cursor that can hop around in its result set. In the first case, then, only FETCH NEXT is allowed, while for a SCROLL Cursor, all forms of FETCH are allowed.

The optional updatability clause defaults to FOR READ ONLY if the Cursor’s definition includes either INSENSITIVE, SCROLL or an ORDER BY clause, or if the Cursor’s result Table is not an updatable Table. If none of these are true, you can either specify FOR UPDATE OF, followed by a list of the result Columns you want to update, or the Cursor definition will default to that state.

FOR READ ONLY means the Cursor is not updatable – that is, UPDATE and DELETE operations won’t be allowed on this Cursor. FOR UPDATE OF, permitted only on an updatable Cursor, means that the Columns specified may be the target of an UPDATE operation. If you include a Column list, only those Columns can be updated by the Cursor. If you omit the Column list, the effect is the same as if you included a list that names every Column of the result Table.

If DECLARE CURSOR includes the optional WITH HOLD clause, the Cursor is a holdable Cursor. This means that if the Cursor is open when a transaction is terminated with a COMMIT statement, it won’t be closed – it will stay open into the next transaction. A holdable Cursor is always closed by ROLLBACK.

If your Cursor definition includes the optional WITH RETURN clause, the Cursor is called a result set Cursor. A result set Cursor that is declared in an SQL-invoked procedure returns a result set if it is open when the procedure ends.

Here’s an example of DECLARE CURSOR, based on the sample Tables we defined in our chapter on simple search conditions:

DECLARE emps_cursor SCROLL CURSOR FOR
       SELECT   dept,empnum,surname,gname,address
       FROM     Employee WHERE dept <= 'C'
       ORDER BY dept,empnum
    FOR UPDATE OF surname,address;

This Cursor definition will allow you to get all the rows of the EMPLOYEE Table where the DEPT value is A, B or C. It will also allow you to UPDATE the values of the DEPT, SURNAME and ADDRESS Columns of the retrieved rows.

If you want to restrict your code to Core SQL, don’t declare a Cursor with either SENSITIVE, INSENSITIVE or ASENSITIVE, don’t declare a Cursor with SCROLL, don’t declare a Cursor with WITH RETURN and don’t declare a Cursor with ORDER BY if you’ve declared it with FOR UPDATE with, or without, a Column list.

OPEN Statement

The OPEN statement opens a Cursor. The required syntax for the OPEN statement is:

OPEN <Cursor name>

The OPEN statement processes the Cursor’s query (it’s an error if the Cursor is already open). In order to OPEN a Cursor, then, your current <AuthorizationID> has to have all the Privileges required to process that query.

While it is open, a Cursor identifies a result Table, as well as a certain position within the rows of that Table: it can be one some row of the Table, before some row of the Table or after the last row of the Table. Immediately after you execute OPEN, the Cursor is positioned before the first row of its result Table.

Here’s an example of OPEN, based on the sample Cursor we defined in the last section:

OPEN emps_cursor;

This SQL statement will evaluate the Cursor query:

SELECT   dept,empnum,surname,gname,address
FROM     Employee
WHERE    dept<='C'
ORDER BY dept,empnum

and position the Cursor before the first row of the result Table.

FETCH Statement

The FETCH statement positions a Cursor on a specific row of its result Table, and retrieves that row’s values into host variables. The required syntax for the FETCH statement is:

FETCH [ [ <fetch orientation> ] FROM ] <Cursor name>
INTO <fetch target list>

   <fetch orientation> ::=
   NEXT |
   PRIOR |
   FIRST |
   LAST |
   {ABSOLUTE | RELATIVE} <simple value specification>

   <fetch target list> ::=
   <target specification> [ {,<target specification>}... ]

The FETCH orientation specification is one of the <keyword>s listed before the <Cursor name>: it defines which row of the Cursor’s result Table FETCH will get next. FETCH NEXT (row), FETCH PRIOR (row), FETCH FIRST (row) and FETCH LAST (row) are self-explanatory. FETCH NEXT is the default, and is the only legal option if the Cursor is not a SCROLL Cursor. For FETCH ABSOLUTE and FETCH RELATIVE, let n be the <simple value specification> (it must be a <literal>, SQL parameter or host variable that represents an integer): FETCH ABSOLUTE n moves the Cursor to the nth row of its result Table (counting backward from the end if n is negative) and FETCH RELATIVE n moves the Cursor to the nth row from the current position (again, counting backward if n is negative). Assuming that the fetch orientation does identify a row of the result Table (e.g.: that there is a NEXT row), then the Cursor is positioned on that row and the SQL-data values are retrieved from that row. If the fetch orientation has gone beyond the result Table – that is, if there is no NEXT row because FETCH already retrieved the final row – then the Cursor is positioned either after the last row or before the first row of the result Table (depending on which way the fetch orientation was moving) and no SQL-data is retrieved.

<Cursor name> is the name of the OPEN Cursor whose result Table you want to FETCH. The values retrieved from each row are placed into a list of host variables by FETCH. You provide your DBMS with the comma-delimited list of output host variables (and their optional indicators) in FETCH's INTO clause. The first value in the row that the Cursor is positioned on is assigned to the first host variable in the list, the second value is assigned to the second host variable, and so on (the number of Columns in the result Table and the number of host variables must, of course, match – as must their <data type>s).

Here’s some examples of FETCH, based on the sample Cursor we opened in the last section:

FETCH NEXT FROM emps_cursor
  INTO :dept,
       :empnum,
       :surname :surname_indicator,
       :gname :gname_indicator,
       :address :address_indicator;

This SQL statement will position the Cursor on the next row of the result Table (which happens to be the first row in this example), and assign the SQL-data values from each Column in that row to the corresponding host variables. Since we declared the Cursor as a SCROLL Cursor, we could use this FETCH statement instead:

FETCH ABSOLUTE 4 FROM emps_cursor
  INTO :dept,
       :empnum,
       :surname :surname_indicator,
       :gname :gname_indicator,
       :address :address_indicator;

This SQL statement will position the Cursor on the fourth row of the result Table, and assign the SQL-data values from each Column in that row to the corresponding host variables.

If you want to restrict your code to Core SQL, don’t use FETCH with a <fetch orientation>: always let it default to FETCH NEXT.

Singleton SELECT Statement

If you know that a result Table will contain only one row of SQL-data, you don’t need to declare a Cursor to get that result into your application program. Instead, you can use the form of the SELECT statement called the singleton SELECT: it puts the values found in a single row into a set of host variables. The required syntax for a singleton SELECT is:

SELECT [ ALL | DISTINCT ] <select list>
INTO <target specification> [ {,<target specification>}... ]
<table expression>

The singleton SELECT statement can only be embedded in an application program. It can only retrieve one row of SQL-data. It is an error if more than one row might satisfy the query – you must use a Cursor to manipulate the data instead.

The singleton SELECT looks exactly like a regular SELECT statement except for the INTO clause that comes between the select list and the <table expression> (i.e.: FROMWHERE … etc.). The INTO clause works like the INTO clause in the FETCH statement: you provide your DBMS with a comma-delimited list of output host variables (and their optional indicators) therein and your DBMS places the first value in the result row into the first host variable in the list, the second value in the result row into the second host variable, and so on (the number of Columns in the result row and the number of host variables must, of course, match – as must their <data type>s).

Here’s an example of a singleton SELECT, based on the sample Tables we defined in our chapter on simple search conditions:

SELECT dept,empnum,surname,gname,address
INTO   :dept,
       :empnum,
       :surname :surname_indicator,
       :gname :gname_indicator,
       :address :address_indicator;
FROM   Employee
WHERE  empnum=10;

This SQL statement will evaluate the query:

SELECT dept,empnum,surname,gname,address
FROM   Employee
WHERE  empnum=10

and assign the SQL-data values from each Column in that row to the corresponding host variables.

INSERT Statement

A special form of the INSERT statement exists for use with a Cursor. The required syntax is:

INSERT INTO <Cursor name> [ (<Column name> [ , ... ]) ]
{<query expression> | DEFAULT VALUES}

This form of the INSERT statement works exactly the same as the INSERT we described in our chapter on the SQL-data change statements, except that you put a <Cursor name> instead of a <Table reference> in the INTO clause. That is, the target of the INSERT operation is the Cursor’s result Table. Here’s an example:

INSERT INTO emps_cursor
VALUES (:dept,
        :empnum,
        :surname :surname_indicator,
        :gname :gname_indicator,
        :address :address_indicator);

If you want to restrict your code to Core SQL, don’t use this form of the INSERT statement.

Positioned UPDATE Statement

The positioned UPDATE statement lets you UPDATE the Cursor’s current row. The required syntax for the positioned UPDATE statement is:

UPDATE [ <Table reference> ] SET
{<Column name>=scalar_expression [ {, ... ] | ROW=row_expression}
WHERE CURRENT OF <Cursor name>

This form of the UPDATE statement works exactly the same as the UPDATE we described in our chapter on the SQL-data change statements, except that you put “WHERE CURRENT OF <Cursor name>” after the SET assignments instead of the (optional) “WHERE condition” clause and that you can omit the <Table reference> after UPDATE, since the <Cursor name> in the WHERE clause identifies the UPDATE target Table anyway.

The Cursor must be open, must be an updatable Cursor and must be positioned on a row of its result Table. That row (the current row) is the row that will be updated. Each Column that is a target of the SET clause must have been mentioned in the FOR UPDATE OF clause of the Cursor’s definition. Any updated Column may not be named in the Cursor’s ORDER BY clause.

Here’s an example of a positioned UPDATE, based on the sample Cursor we fetched with earlier:

UPDATE Employee SET
  surname = :new_surname :new_surname_indicator,
  address = :new_address :new_address_indicator
WHERE CURRENT OF emps_cursor;

This SQL statement changes the values of the SURNAME and ADDRESS Columns in the current row of the Cursor’s result Table. After the UPDATE, the Cursor remains on its current row. It won’t move until another FETCH statement is executed.

If you want to restrict your code to Core SQL, don’t omit the <Table reference> from a positioned UPDATE statement, don’t add an ORDER BY clause to your Cursor definition, and don’t have an UPDATE target that is a <literal>, host variable or SQL parameter.

Positioned DELETE Statement

The positioned delete statement lets you DELETE the Cursor’s current row. The required syntax for the positioned DELETE statement is:

DELETE [ FROM <Table reference> ] WHERE CURRENT OF <Cursor name>

This form of the DELETE statement works exactly the same as the DELETE we described in our chapter on the SQL-data change statements, except that you use “WHERE CURRENT OF <Cursor name>” instead of the (optional) “WHERE condition” clause and that you can omit the FROM <Table reference> clause after DELETE, since the <Cursor name> in the WHERE clause identifies the DELETE target Table anyway.

The Cursor must be open, must be an updatable Cursor and must be positioned on a row of its result Table. That row (the current row) is the row that will be deleted. Here’s an example of a positioned DELETE, based on the sample Cursor we fetched with earlier:

DELETE FROM Employee WHERE CURRENT OF emps_cursor;

This SQL statement deletes the current row of the Cursor’s result Table. After the DELETE, the Cursor is positioned before the row that follows the deleted row (or after the last row, if the deleted row was the last row). It won’t be positioned on another row until another FETCH statement is executed.

CLOSE Statement

The CLOSE statement closes a Cursor. The required syntax for the CLOSE statement is:

CLOSE <Cursor name>

The CLOSE statement destroys the Cursor’s result Table (it’s an error if the Cursor isn’t open). Closing a Cursor causes your DBMS to immediately check all Constraints that were affected by Cursor operations for violation and to execute any triggered actions that were deferred during the Cursor operations. Here’s an example:

CLOSE emps_cursor;

A Cursor is also closed by ROLLBACK and (unless it’s a holdable Cursor) by COMMIT.

Embedded SQL Examples

Here is an example program which selects and fetches ten rows in Table T, and displays the contents. This program watches for NULL values, which it indicates by displaying a question mark rather than a number. (The use of “?” to mean “NULL” is conventional in displays and printouts, but sometimes blanks are used instead.)

 #include <stdio.h>                         /* Example <xx> */
 EXEC SQL BEGIN DECLARE SECTION;
    int x;
    int x_indicator;
 EXEC SQL END DECLARE SECTION;
 int i;
 void main ()
{
   EXEC SQL CONNECT TO database_or_server USER Josephine;
   EXEC SQL DECLARE example_cursor CURSOR FOR SELECT col_1 FROM T;
   EXEC SQL OPEN example_cursor;
   for (i=0; i<10; ++i) {
     EXEC SQL FETCH example_cursor INTO :x INDICATOR :x_indicator;
     if (x_indicator < 0) printf("?\n");
     else printf("%d\n",x); }
   EXEC SQL CLOSE example_cursor;
   EXEC SQL DISCONNECT database_or_server; }

Diagnostics

“I beseech you in the bowels of Christ, consider that you may be wrong.” – Cromwell

After any embedded SQL statement, you can add code to ask the DBMS: “has anything gone wrong?”, and if so “what has gone wrong?”. The DBMS’s answers – diagnostics – are categorized and coded in a standard way, so you have some advance knowledge about what the possible scenarios are. The basic piece of diagnostic information is SQLSTATE.

Originally, the basic item of diagnostic information was an integer variable called sqlcode (or, in Fortran: SQLCOD). After every SQL statement, the DBMS would put a value in sqlcode. If the value was less than zero, that meant “error”. If the value was zero, that meant “success”. If the value was greater than zero, that meant “success with additional information”, that is, a warning. The most common warning was +100, which meant “no data”. This was an easy system to follow, and it was common practice to add lines like these after every SQL statement:

if (sqlcode < 0) {
  printf("Error!\n"); /*or some more sophisticated error-handling action*/
  exit(1); }

It is still common today to see sqlcode checking, especially since all DBMS vendors still allow for it. However, the SQL-92 Standard “deprecated” use of sqlcode and the SQL3 Standard doesn’t recognize it at all. SQLSTATE is the modern basic item of diagnostic information.

A big advantage of SQLSTATE is that the possible values are standardized for each error category, a level of agreement that was never achieved with the sqlcode values. SQLSTATE is a 5-character variable which you should define in the declare section. After every SQL statement, the DBMS will put a value (the status code) into SQLSTATE. The status code consists of digits or upper case letters between A and Z. The first two letters or digits are the “class” (the general category), the next three letters or digits are the “subclass” (the specific category, if any). You can see a complete list of status codes, and descriptions thereof, in our chapter on SQL/CLI diagnostics. For now, content yourself with the knowledge of the most important classes:

  • Class ‘00’ is ‘SUCCESS’. When you see this, everything’s going fine.

  • Class ‘01’ is ‘SUCCESS WITH INFO’, or ‘WARNING’. For example, perhaps some precision was lost during an arithmetic calculation. There is usually no cause to change the course of your program’s flow, but displaying a message is often an appropriate action for your program to take.

  • Class ‘02’ is ‘NO DATA’. Every FETCH loop should watch for this one, since FETCH will cause ‘NO DATA’ if there is nothing more to fetch.

  • All other classes are ‘ERROR’. The SQL statement has failed. Check paranoidly. If SQLSTATE contains anything other than ‘00’ or ‘01’ or ‘02’, you might have to take some corrective action or even abort the job.

Let us now rewrite our FETCH loop program. This time we’ll check SQLSTATE.

#include <stdio.h>                         /* Example <xx - b> */
EXEC SQL BEGIN DECLARE SECTION;
int x;
int x_indicator;
char sqlstate[6]; /* SQLSTATE is 5 characters. In C, allow for '\0' too. */
EXEC SQL END DECLARE SECTION;
int i;
void main ()
{
  EXEC SQL CONNECT TO database_or_server USER Josephine;
  if (... <> "00" && ... != "01")
    printf("Connection failed. sqlstate = %s.\n",sqlstate);
  EXEC SQL DECLARE example_cursor CURSOR FOR SELECT col_1 FROM T;
  /* There is no need to check for errors after a DECLARE statement. */
  EXEC SQL OPEN example_cursor;
  if (... <> "00" && ... != "01")
    printf("OPEN failed. sqlstate = %s.\n",sqlstate);
    goto disconnect_database; }
  for (;;) {  /* This loop is not infinite. There are break; statements. */
    EXEC SQL FETCH example_cursor INTO :x INDICATOR :x_indicator;
    if (... == '02') {
      /* The 'NO DATA' class just means there's nothing more to fetch. */
      break; }
    if (... == '01') printf("(Warning: sqlstate=%s.\n",sqlstate);
    if (... <> '00' && ... <> '01' && ... <> '02') {
      printf("FETCH failed. sqlstate = %s.\n",sqlstate);
      break; }
    if (x_indicator < 0) printf("?\n");
    else printf("%d\n",x); }
  EXEC SQL CLOSE example_cursor;
  /* Doubtless a CLOSE will always succeed, but let's check anyway. */
  if (... <> "00") printf("After close, sqlstate=%s.\n",sqlstate);
disconnect_database:
  EXEC SQL DISCONNECT database_or_server; }
  /* We're ending the program, but let's check anyway. */
  if (... <> "00") printf("After disconnect, sqlstate=%s.\n",sqlstate); }

If you find it tedious to check SQLSTATE after every single error, there is an SQL directive to automate the procedure: the WHENEVER statement.

WHENEVER Statement

The required syntax for the WHENEVER statement (omitting some details) is:

WHENEVER <condition> {GOTO | GO TO} <target of goto>

or

WHENEVER <condition> CONTINUE

The WHENEVER <condition> is one of:

  • SQLWARNING … true if SQLSTATE contains the ‘01’ status code class.

  • NOT FOUND … true for the SQLSTATE02’ class.

  • SQLEXCEPTION … true for any other SQLSTATE class. (The old SQL-92 name for this is SQLERROR.)

  • SQLSTATE (list) … an SQL3 innovation, rarely supported.

  • CONSTRAINT (name) … an SQL3 innovation, rarely supported.

The WHENEVER target is a label or address which would be legal as the argument of a GOTO statement in the host language. The WHENEVER directive is not an executable statement; it’s a signal to the precompiler. The precompiler responds to it by generating appropriate “if … goto …” statements after every SQL statement.

Tip

When you write a program for the first time, put EXEC SQL WHENEVER SQLEXCEPTION GOTO error_abort; at the start. After you’ve got the program working, edit it and put in error testing that’s more specific.

Serious programs contain considerably more than basic SQLSTATE checks. To begin with, they have different reactions depending on the specific class and subclass in the status code. They also make use of SQL’s GET DIAGNOSTICS statement. With GET DIAGNOSTICS, one can ask more about the context, get an implementation-defined error message, and perhaps retrieve several different diagnostics (it’s possible, for instance, that the DBMS generated several warnings before it ultimately gave up and generated an error).

GET DIAGNOSTICS Statement

The GET DIAGNOSTICS statement gets exception or completion condition information from the diagnostics area. The required syntax for the GET DIAGNOSTICS statement is:

GET DIAGNOSTICS {<statement information> | <condition information>}

   <statement information> ::=
   <statement information item> [ {,<statement information item>}... ]

       <statement information item> ::=
       <simple target specification> = <statement item name>

         <statement information item name> ::=
         NUMBER |
         MORE |
         COMMAND_FUNCTION |
         COMMAND_FUNCTION_CODE |
         ROW_COUNT |
         TRANSACTIONS_COMMITTED |
         TRANSACTIONS_ROLLED_BACK |
         TRANSACTION_ACTIVE

   <condition information> ::=
   EXCEPTION <condition number> <condition information item> [ {,<condition information item>}... ]

      <condition information item> ::=
      <simple target specification> = <condition item name>

         <condition information item name> ::=
         CONDITION_IDENTIFIER |
         CONDITION_NUMBER |
         RETURNED_SQLSTATE |
         CLASS_ORIGIN |
         SUBCLASS_ORIGIN |
         SERVER_NAME |
         CONNECTION_NAME |
         CONSTRAINT_CATALOG |
         CONSTRAINT_SCHEMA |
         CONSTRAINT_NAME |
         TRIGGER_CATALOG |
         TRIGGER_SCHEMA |
         TRIGGER_NAME |
         CATALOG_NAME |
         SCHEMA_NAME |
         TABLE_NAME |
         COLUMN_NAME |
         CURSOR_NAME |
         ROUTINE_CATALOG |
         ROUTINE_SCHEMA |
         ROUTINE_NAME |
         SPECIFIC_NAME |
         PARAMETER_NAME |
         MESSAGE_TEXT |
         MESSAGE_LENGTH |
         MESSAGE_OCTET_LENGTH

      <condition number> ::= <simple value specification>

There are two forms of the GET DIAGNOSTICS statement. The first gets information about the overall execution of the immediately preceding SQL statement, while the second form gets more specific information about one or more specific errors. Here are some examples of GET DIAGNOSTICS:

GET DIAGNOSTICS :smallint_host_variable = ROW_COUNT;

GET DIAGNOSTICS :char_host_variable = DYNAMIC_FUNCTION;

GET DIAGNOSTICS EXCEPTION 1 :char_host_variable = SUBCLASS_ORIGIN;

Not wanting to bore you twice, we’ll defer the detailed discussion of the diagnostics area’s fields till a later time – when we talk about the SQL/CLI’s SQLGetDiagRec and SQLGetDiagField functions.

Dynamic SQL

In the examples so far, we’ve assumed that we know some things in advance about the database and the SQL statements we’re going to execute. For example, in our FETCH loop examples, we make the assumptions that the <Table name> is T, that T a Column called COL_1, that COL_1 has a numeric <data type>, and so on. A program which contains such assumptions is called a static SQL program, because its SQL statements are changeless. Let us now suppose that we lack advance knowledge about the program. The classic supposition is: what if the user types in a SELECT command on the keyboard? We need to use a more flexible (and somewhat less efficient) tool set: the components of dynamic SQL.

Start with the basic supposition: that the user types in an SQL statement on the keyboard. There is a simple embedded SQL statement which parses and executes strings at runtime: EXECUTE IMMEDIATE. Here is an example:

...
gets(a_string);                           EXAMPLE A
for (i=0; i<1000; ++i) {
  EXEC SQL EXECUTE IMMEDIATE :a_string; }
...

In this example, a_string is a character string host variable. If it contains the string INSERT INTO T VALUES (5), then the above statement is equivalent to:

for (i=0; i<1000; ++i) {
  EXEC SQL INSERT INTO T VALUES (5); }

The good news about dynamic SQL is that you have an SQL interpreter available. That’s remarkable when you compare SQL with a host language. You can’t write equivalent code which will parse and execute a C statement from C, or a Pascal statement from Pascal! The result is that your SQL code is more capable than your host language code. The bad news is that interpreters are slow. Faced with repetitive operations like the ones in our example above, we would rather replace “parse and execute many times” with “parse once, execute many times”. Luckily, there are embedded SQL statements for that – PREPARE and EXECUTE. Here is a replacement example, which does the same thing but is maybe more efficient:

...
gets(a_string);                            EXAMPLE B
EXEC SQL PREPARE :a_string;
for (i=0; i<1000; ++i)
  EXEC SQL EXECUTE :a_string; }

Note

Some smart DBMSs can detect simple situations like the one in this example and optimize accordingly. For example, if you program with Oracle you will gain nothing by replacing Example A with Example B.)

The ultimate in embedded-SQL complexity is a dynamically executed query, such as a SELECT statement. We can’t define host variables and indicators in advance, because we don’t know what the <data type>s of the result Columns are – or even how many Columns the query will return. The first time you see dynamic SQL query code, you may have trouble understanding it. But the second time, you’ll have less trouble – because it will be the same. Most programmers use a boilerplate, so the program structure turns out to be similar in every program. In pseudocode that structure looks like this:

if (the statement is a query) do begin:
  Execute the statement.   || Variation: just prepare the statement.
  Ask the DBMS: "how many Columns were there?"
    Let the response be C.
  for (Column-number = 1 to C) do:
    Ask the DBMS: "for Column[Column-number], what is the data type etc.?"
    Using the DBMS's response, allocate appropriate RAM in host program.
    end.
  ...
  OPEN Cursor.
  Loop:
    FETCH. The fetch targets are the allocated areas in the host program.
    if (no-data) exit loop.

At this point, code varies depending on what we want to do with the dynamically retrieved data. For instance we could display it:

  display data.
  Next.
CLOSE Cursor.
end.

Principally, you’re asking the DBMS to DESCRIBE what it finds. The key statements for the purpose are:

  • ALLOCATE DESCRIPTOR ... which tells the DBMS to allocate an area internally (i.e. within the DBMS) for storing the description.

  • DESCRIBE <SQL statement name> <using descriptor> … which fills the descriptor with information about the SQL statement.

  • GET DESCRIPTOR ... which takes descriptor fields’s values and puts them into host program variables.

  • DEALLOCATE DESCRIPTOR ... which gets rid of what ALLOCATE DESCRIPTOR made.

As with GET DIAGNOSTICS, we’ll defer the detailed discussion of the descriptor area’s fields till a later time – when we talk about the SQL/CLI’s desc functions.

Summary

An embedded SQL host language program is an application program that contains both host language programming statements and SQL statements. Embedded SQL programs may be written in any of the high-level languages – ADA, C, COBOL, FORTRAN, MUMPS, PASCAL and PL/I – supported by the SQL Standard. Such programs use SQL, instead of routines written in the host language, to carry out database operations.

Embedded SQL statements are either static SQL statements or dynamic SQL statements. Static SQL is static in the sense that the embedded SQL statement is fully known, and coded, when your program is written. Executing the program does not change the SQL statement in any way. Dynamic SQL is so-called because the embedded SQL statement is not fully known when your program is written. Some part of it will be generated during the execution of the program.

SQL statements embedded in an application program must reference host variables to pass values between the program and SQL-data. Host variables are host language variables referenced in an embedded SQL statement. They must be declared within a “declare section” that is delimited by the BEGIN DECLARE SECTION and END DECLARE SECTION statements. The declare section must precede any use of the host variables. Since host variables must map to host language variables, they are not nullable unless they are coupled with an indicator variable.

An indicator variable is a host language variable that acts as an indicator for another host variable. The purpose of an indicator variable is to highlight null values and overflow conditions for the host variable it’s coupled with. Embedded SQL application programs must prepare for passing null values by pairing an indicator with any host variable that might be assigned a null value.

SQL provides a status host variableSQLSTATE – whose values indicate whether or not an SQL statement was successfully executed. All embedded SQL programs must contain at least one status host variable declaration.

The major problem in combining SQL with any host language is the inability of most other languages to handle multiple records at one time; one of the major SQL features. The SQL Object “Cursor” is provided by the SQL Standard to give the host languages a facility for dealing with a set of retrieved SQL-data records, one at a time.

An excellent summary of the embedded SQL specification appears in the SQL Standard. Part five says:

  • Embedded SQL is “syntax for embedding SQL-statements in a compilation unit that otherwise conforms to the standard for a particular programming language (host language).”

  • Embedded SQL specifies “how an equivalent compilation unit may be derived that conforms to the particular programming language standard. In that equivalent compilation unit, each embedded SQL-statement has been replaced by one or more statements in the host language, some of which invoke an SQL externally-invoked procedure that, when executed, has an effect equivalent [to] executing the SQL-statement.”

Dialects

Embedded SQL is well specified by part five of the SQL Standard: ISO/IEC 9075-2 Part 5: SQL/Bindings. The differences between SQL-92 and SQL3 in this area are minor, but you may find that older programs handle dynamic-SQL “descriptors” in a different way. Instead of declaring that descriptors are internal to the DBMS, old programs may allocate an area within the host program itself. This area, usually called SQLDA (SQL Dynamic Area) or something similar, is a structure like this:

#define  MV    64
struct sqlda {
  char sqldaid[8];     /* "SQLDA   " */
  long int sqldabc;    /* length of sqlda  */
  int sqln;            /* max sqlvar occurrences = MV */
  int sqld;            /* current sqlvar occurrences */
  struct sqlvar {      /* defined sqln times, actually occurs sqld times */
   int sqltype;        /* <data type> of Column  */
   int sqllen;         /* size of Column */
    char far *sqldata; /* pointer --> host variable */
    int far *sqlind;   /* pointer --> indicator */
    struct sqlname {   /* string, up to 30 bytes long */
     int length;       /* 1-word size of name */
     char data[30]; } sqlname; } sqlvar[MV]; };
struct sqlda sqlda={"SQLDA",0,64,0};

The job of the DESCRIBE statement is to fill in the entire SQLDA. Then the host program can access the fields by simply referring to the structure’s components.

As we mentioned, it is possible to integrate the precompiler with the compiler. Thus the process of precompiling and compiling appears to be a single step. IBM’s DB2 is particularly noteworthy for this convenient feature.

Another convenience is the allowance of “SQL” keywords for data type descriptions within the declare section. This usually goes hand in hand with macros or typedefs within a header file supplied by the vendor. Thus, one could have:

EXEC SQL BEGIN DECLARE SECTION;
  SQLCHAR x(5);
  SQLINTEGER y;
EXEC SQL END DECLARE SECTION;

This enables programmers to use similar declarations for a variety of host languages. Since the same header file can be used in the SQL/CLI, it’s easy to put together packages which use some embedded SQL modules and some SQL/CLI modules.