Remove old RULE privilege completely.

[pgsql.git] / doc / src / sgml / xfunc.sgml

<!-- doc/src/sgml/xfunc.sgml -->

 <sect1 id="xfunc">
  <title>User-Defined Functions</title>

  <indexterm zone="xfunc">
   <primary>function</primary>
   <secondary>user-defined</secondary>
  </indexterm>

  <para>
   <productname>PostgreSQL</productname> provides four kinds of
   functions:

   <itemizedlist>
    <listitem>
     <para>
      query language functions (functions written in
      <acronym>SQL</acronym>) (<xref linkend="xfunc-sql"/>)
     </para>
    </listitem>
    <listitem>
     <para>
      procedural language functions (functions written in, for
      example, <application>PL/pgSQL</application> or <application>PL/Tcl</application>)
      (<xref linkend="xfunc-pl"/>)
     </para>
    </listitem>
    <listitem>
     <para>
      internal functions (<xref linkend="xfunc-internal"/>)
     </para>
    </listitem>
    <listitem>
     <para>
      C-language functions (<xref linkend="xfunc-c"/>)
     </para>
    </listitem>
   </itemizedlist>
  </para>

  <para>
   Every kind
   of  function  can take base types, composite types, or
   combinations of these as arguments (parameters). In addition,
   every kind of function can return a base type or
   a composite type.  Functions can also be defined to return
   sets of base or composite values.
  </para>

  <para>
   Many kinds of functions can take or return certain pseudo-types
   (such as polymorphic types), but the available facilities vary.
   Consult the description of each kind of function for more details.
  </para>

  <para>
   It's easiest to define <acronym>SQL</acronym>
   functions, so we'll start by discussing those.
   Most of the concepts presented for <acronym>SQL</acronym> functions
   will carry over to the other types of functions.
  </para>

  <para>
   Throughout this chapter, it can be useful to look at the reference
   page of the <link linkend="sql-createfunction"><command>CREATE
   FUNCTION</command></link> command to
   understand the examples better.  Some examples from this chapter
   can be found in <filename>funcs.sql</filename> and
   <filename>funcs.c</filename> in the <filename>src/tutorial</filename>
   directory in the <productname>PostgreSQL</productname> source
   distribution.
  </para>
  </sect1>

  <sect1 id="xproc">
   <title>User-Defined Procedures</title>

  <indexterm zone="xproc">
   <primary>procedure</primary>
   <secondary>user-defined</secondary>
  </indexterm>

   <para>
    A procedure is a database object similar to a function.
    The key differences are:

    <itemizedlist>
     <listitem>
      <para>
       Procedures are defined with
       the <link linkend="sql-createprocedure"><command>CREATE
       PROCEDURE</command></link> command, not <command>CREATE
       FUNCTION</command>.
      </para>
     </listitem>
     <listitem>
      <para>
       Procedures do not return a function value; hence <command>CREATE
       PROCEDURE</command> lacks a <literal>RETURNS</literal> clause.
       However, procedures can instead return data to their callers via
       output parameters.
      </para>
     </listitem>
     <listitem>
      <para>
       While a function is called as part of a query or DML command, a
       procedure is called in isolation using
       the <link linkend="sql-call"><command>CALL</command></link> command.
      </para>
     </listitem>
     <listitem>
      <para>
       A procedure can commit or roll back transactions during its
       execution (then automatically beginning a new transaction), so long
       as the invoking <command>CALL</command> command is not part of an
       explicit transaction block.  A function cannot do that.
      </para>
     </listitem>
     <listitem>
      <para>
       Certain function attributes, such as strictness, don't apply to
       procedures.  Those attributes control how the function is
       used in a query, which isn't relevant to procedures.
      </para>
     </listitem>
    </itemizedlist>
   </para>

   <para>
    The explanations in the following sections about how to define
    user-defined functions apply to procedures as well, except for the
    points made above.
   </para>

   <para>
    Collectively, functions and procedures are also known
    as <firstterm>routines</firstterm><indexterm><primary>routine</primary></indexterm>.
    There are commands such as <link linkend="sql-alterroutine"><command>ALTER ROUTINE</command></link>
    and <link linkend="sql-droproutine"><command>DROP ROUTINE</command></link> that can operate on functions and
    procedures without having to know which kind it is.  Note, however, that
    there is no <literal>CREATE ROUTINE</literal> command.
   </para>
  </sect1>

  <sect1 id="xfunc-sql">
   <title>Query Language (<acronym>SQL</acronym>) Functions</title>

   <indexterm zone="xfunc-sql">
    <primary>function</primary>
    <secondary>user-defined</secondary>
    <tertiary>in SQL</tertiary>
   </indexterm>

   <para>
    SQL functions execute an arbitrary list of SQL statements, returning
    the result of the last query in the list.
    In the simple (non-set)
    case, the first row of the last query's result will be returned.
    (Bear in mind that <quote>the first row</quote> of a multirow
    result is not well-defined unless you use <literal>ORDER BY</literal>.)
    If the last query happens
    to return no rows at all, the null value will be returned.
   </para>

   <para>
    Alternatively, an SQL function can be declared to return a set (that is,
    multiple rows) by specifying the function's return type as <literal>SETOF
    <replaceable>sometype</replaceable></literal>, or equivalently by declaring it as
    <literal>RETURNS TABLE(<replaceable>columns</replaceable>)</literal>.  In this case
    all rows of the last query's result are returned.  Further details appear
    below.
   </para>

   <para>
    The body of an SQL function must be a list of SQL
    statements separated by semicolons.  A semicolon after the last
    statement is optional.  Unless the function is declared to return
    <type>void</type>, the last statement must be a <command>SELECT</command>,
    or an <command>INSERT</command>, <command>UPDATE</command>,
    <command>DELETE</command>, or <command>MERGE</command>
    that has a <literal>RETURNING</literal> clause.
   </para>

    <para>
     Any collection of commands in the  <acronym>SQL</acronym>
     language can be packaged together and defined as a function.
     Besides <command>SELECT</command> queries, the commands can include data
     modification queries (<command>INSERT</command>,
     <command>UPDATE</command>, <command>DELETE</command>, and
     <command>MERGE</command>), as well as
     other SQL commands. (You cannot use transaction control commands, e.g.,
     <command>COMMIT</command>, <command>SAVEPOINT</command>, and some utility
     commands, e.g.,  <literal>VACUUM</literal>, in <acronym>SQL</acronym> functions.)
     However, the final command
     must be a <command>SELECT</command> or have a <literal>RETURNING</literal>
     clause that returns whatever is
     specified as the function's return type.  Alternatively, if you
     want to define an SQL function that performs actions but has no
     useful value to return, you can define it as returning <type>void</type>.
     For example, this function removes rows with negative salaries from
     the <literal>emp</literal> table:

<screen>
CREATE FUNCTION clean_emp() RETURNS void AS '
    DELETE FROM emp
        WHERE salary &lt; 0;
' LANGUAGE SQL;

SELECT clean_emp();

 clean_emp
-----------

(1 row)
</screen>
    </para>

    <para>
     You can also write this as a procedure, thus avoiding the issue of the
     return type.  For example:
<screen>
CREATE PROCEDURE clean_emp() AS '
    DELETE FROM emp
        WHERE salary &lt; 0;
' LANGUAGE SQL;

CALL clean_emp();
</screen>
     In simple cases like this, the difference between a function returning
     <type>void</type> and a procedure is mostly stylistic.  However,
     procedures offer additional functionality such as transaction control
     that is not available in functions.  Also, procedures are SQL standard
     whereas returning <type>void</type> is a PostgreSQL extension.
    </para>

    <note>
     <para>
      The entire body of an SQL function is parsed before any of it is
      executed.  While an SQL function can contain commands that alter
      the system catalogs (e.g., <command>CREATE TABLE</command>), the effects
      of such commands will not be visible during parse analysis of
      later commands in the function.  Thus, for example,
      <literal>CREATE TABLE foo (...); INSERT INTO foo VALUES(...);</literal>
      will not work as desired if packaged up into a single SQL function,
      since <structname>foo</structname> won't exist yet when the <command>INSERT</command>
      command is parsed.  It's recommended to use <application>PL/pgSQL</application>
      instead of an SQL function in this type of situation.
     </para>
   </note>

   <para>
    The syntax of the <command>CREATE FUNCTION</command> command requires
    the function body to be written as a string constant.  It is usually
    most convenient to use dollar quoting (see <xref
    linkend="sql-syntax-dollar-quoting"/>) for the string constant.
    If you choose to use regular single-quoted string constant syntax,
    you must double single quote marks (<literal>'</literal>) and backslashes
    (<literal>\</literal>) (assuming escape string syntax) in the body of
    the function (see <xref linkend="sql-syntax-strings"/>).
   </para>

   <sect2 id="xfunc-sql-function-arguments">
    <title>Arguments for <acronym>SQL</acronym> Functions</title>

   <indexterm>
    <primary>function</primary>
    <secondary>named argument</secondary>
   </indexterm>

    <para>
     Arguments of an SQL function can be referenced in the function
     body using either names or numbers.  Examples of both methods appear
     below.
    </para>

    <para>
     To use a name, declare the function argument as having a name, and
     then just write that name in the function body.  If the argument name
     is the same as any column name in the current SQL command within the
     function, the column name will take precedence.  To override this,
     qualify the argument name with the name of the function itself, that is
     <literal><replaceable>function_name</replaceable>.<replaceable>argument_name</replaceable></literal>.
     (If this would conflict with a qualified column name, again the column
     name wins.  You can avoid the ambiguity by choosing a different alias for
     the table within the SQL command.)
    </para>

    <para>
     In the older numeric approach, arguments are referenced using the syntax
     <literal>$<replaceable>n</replaceable></literal>: <literal>$1</literal> refers to the first input
     argument, <literal>$2</literal> to the second, and so on.  This will work
     whether or not the particular argument was declared with a name.
    </para>

    <para>
     If an argument is of a composite type, then the dot notation,
     e.g., <literal><replaceable>argname</replaceable>.<replaceable>fieldname</replaceable></literal> or
     <literal>$1.<replaceable>fieldname</replaceable></literal>, can be used to access attributes of the
     argument.  Again, you might need to qualify the argument's name with the
     function name to make the form with an argument name unambiguous.
    </para>

    <para>
     SQL function arguments can only be used as data values,
     not as identifiers.  Thus for example this is reasonable:
<programlisting>
INSERT INTO mytable VALUES ($1);
</programlisting>
but this will not work:
<programlisting>
INSERT INTO $1 VALUES (42);
</programlisting>
    </para>

    <note>
     <para>
      The ability to use names to reference SQL function arguments was added
      in <productname>PostgreSQL</productname> 9.2.  Functions to be used in
      older servers must use the <literal>$<replaceable>n</replaceable></literal> notation.
     </para>
    </note>
   </sect2>

   <sect2 id="xfunc-sql-base-functions">
    <title><acronym>SQL</acronym> Functions on Base Types</title>

    <para>
     The simplest possible <acronym>SQL</acronym> function has no arguments and
     simply returns a base type, such as <type>integer</type>:

<screen>
CREATE FUNCTION one() RETURNS integer AS $$
    SELECT 1 AS result;
$$ LANGUAGE SQL;

-- Alternative syntax for string literal:
CREATE FUNCTION one() RETURNS integer AS '
    SELECT 1 AS result;
' LANGUAGE SQL;

SELECT one();

 one
-----
   1
</screen>
    </para>

    <para>
     Notice that we defined a column alias within the function body for the result of the function
     (with  the  name <literal>result</literal>),  but this column alias is not visible
     outside the function.  Hence,  the  result  is labeled <literal>one</literal>
     instead of <literal>result</literal>.
    </para>

    <para>
     It is almost as easy to define <acronym>SQL</acronym> functions
     that take base types as arguments:

<screen>
CREATE FUNCTION add_em(x integer, y integer) RETURNS integer AS $$
    SELECT x + y;
$$ LANGUAGE SQL;

SELECT add_em(1, 2) AS answer;

 answer
--------
      3
</screen>
    </para>

    <para>
     Alternatively, we could dispense with names for the arguments and
     use numbers:

<screen>
CREATE FUNCTION add_em(integer, integer) RETURNS integer AS $$
    SELECT $1 + $2;
$$ LANGUAGE SQL;

SELECT add_em(1, 2) AS answer;

 answer
--------
      3
</screen>
    </para>

    <para>
     Here is a more useful function, which might be used to debit a
     bank account:

<programlisting>
CREATE FUNCTION tf1 (accountno integer, debit numeric) RETURNS numeric AS $$
    UPDATE bank
        SET balance = balance - debit
        WHERE accountno = tf1.accountno;
    SELECT 1;
$$ LANGUAGE SQL;
</programlisting>

     A user could execute this function to debit account 17 by $100.00 as
     follows:

<programlisting>
SELECT tf1(17, 100.0);
</programlisting>
    </para>

    <para>
     In this example, we chose the name <literal>accountno</literal> for the first
     argument, but this is the same as the name of a column in the
     <literal>bank</literal> table.  Within the <command>UPDATE</command> command,
     <literal>accountno</literal> refers to the column <literal>bank.accountno</literal>,
     so <literal>tf1.accountno</literal> must be used to refer to the argument.
     We could of course avoid this by using a different name for the argument.
    </para>

    <para>
     In practice one would probably like a more useful result from the
     function than a constant 1, so a more likely definition
     is:

<programlisting>
CREATE FUNCTION tf1 (accountno integer, debit numeric) RETURNS numeric AS $$
    UPDATE bank
        SET balance = balance - debit
        WHERE accountno = tf1.accountno;
    SELECT balance FROM bank WHERE accountno = tf1.accountno;
$$ LANGUAGE SQL;
</programlisting>

     which adjusts the balance and returns the new balance.
     The same thing could be done in one command using <literal>RETURNING</literal>:

<programlisting>
CREATE FUNCTION tf1 (accountno integer, debit numeric) RETURNS numeric AS $$
    UPDATE bank
        SET balance = balance - debit
        WHERE accountno = tf1.accountno
    RETURNING balance;
$$ LANGUAGE SQL;
</programlisting>
    </para>

    <para>
     If the final <literal>SELECT</literal> or <literal>RETURNING</literal>
     clause in an <acronym>SQL</acronym> function does not return exactly
     the function's declared result
     type, <productname>PostgreSQL</productname> will automatically cast
     the value to the required type, if that is possible with an implicit
     or assignment cast.  Otherwise, you must write an explicit cast.
     For example, suppose we wanted the
     previous <function>add_em</function> function to return
     type <type>float8</type> instead.  It's sufficient to write

<programlisting>
CREATE FUNCTION add_em(integer, integer) RETURNS float8 AS $$
    SELECT $1 + $2;
$$ LANGUAGE SQL;
</programlisting>

     since the <type>integer</type> sum can be implicitly cast
     to <type>float8</type>.
     (See <xref linkend="typeconv"/> or <xref linkend="sql-createcast"/>
     for more about casts.)
    </para>
   </sect2>

   <sect2 id="xfunc-sql-composite-functions">
    <title><acronym>SQL</acronym> Functions on Composite Types</title>

    <para>
     When writing functions with arguments of composite types, we must not
     only specify which argument we want but also the desired attribute
     (field) of that argument.  For example, suppose that
     <type>emp</type> is a table containing employee data, and therefore
     also the name of the composite type of each row of the table.  Here
     is a function <function>double_salary</function> that computes what someone's
     salary would be if it were doubled:

<screen>
CREATE TABLE emp (
    name        text,
    salary      numeric,
    age         integer,
    cubicle     point
);

INSERT INTO emp VALUES ('Bill', 4200, 45, '(2,1)');

CREATE FUNCTION double_salary(emp) RETURNS numeric AS $$
    SELECT $1.salary * 2 AS salary;
$$ LANGUAGE SQL;

SELECT name, double_salary(emp.*) AS dream
    FROM emp
    WHERE emp.cubicle ~= point '(2,1)';

 name | dream
------+-------
 Bill |  8400
</screen>
    </para>

    <para>
     Notice the use of the syntax <literal>$1.salary</literal>
     to select one field of the argument row value.  Also notice
     how the calling <command>SELECT</command> command
     uses <replaceable>table_name</replaceable><literal>.*</literal> to select
     the entire current row of a table as a composite value.  The table
     row can alternatively be referenced using just the table name,
     like this:
<screen>
SELECT name, double_salary(emp) AS dream
    FROM emp
    WHERE emp.cubicle ~= point '(2,1)';
</screen>
     but this usage is deprecated since it's easy to get confused.
     (See <xref linkend="rowtypes-usage"/> for details about these
     two notations for the composite value of a table row.)
    </para>

    <para>
     Sometimes it is handy to construct a composite argument value
     on-the-fly.  This can be done with the <literal>ROW</literal> construct.
     For example, we could adjust the data being passed to the function:
<screen>
SELECT name, double_salary(ROW(name, salary*1.1, age, cubicle)) AS dream
    FROM emp;
</screen>
    </para>

    <para>
     It is also possible to build a function that returns a composite type.
     This is an example of a function
     that returns a single <type>emp</type> row:

<programlisting>
CREATE FUNCTION new_emp() RETURNS emp AS $$
    SELECT text 'None' AS name,
        1000.0 AS salary,
        25 AS age,
        point '(2,2)' AS cubicle;
$$ LANGUAGE SQL;
</programlisting>

     In this example we have specified each of  the  attributes
     with  a  constant value, but any computation
     could have been substituted for these constants.
    </para>

    <para>
     Note two important things about defining the function:

     <itemizedlist>
      <listitem>
       <para>
        The select list order in the query must be exactly the same as
        that in which the columns appear in the composite type.
        (Naming the columns, as we did above,
        is irrelevant to the system.)
       </para>
      </listitem>
      <listitem>
       <para>
        We must ensure each expression's type can be cast to that of
        the corresponding column of the composite type.
        Otherwise we'll get errors like this:
<screen>
<computeroutput>
ERROR:  return type mismatch in function declared to return emp
DETAIL:  Final statement returns text instead of point at column 4.
</computeroutput>
</screen>
        As with the base-type case, the system will not insert explicit
        casts automatically, only implicit or assignment casts.
       </para>
      </listitem>
     </itemizedlist>
    </para>

    <para>
     A different way to define the same function is:

<programlisting>
CREATE FUNCTION new_emp() RETURNS emp AS $$
    SELECT ROW('None', 1000.0, 25, '(2,2)')::emp;
$$ LANGUAGE SQL;
</programlisting>

     Here we wrote a <command>SELECT</command> that returns just a single
     column of the correct composite type.  This isn't really better
     in this situation, but it is a handy alternative in some cases
     &mdash; for example, if we need to compute the result by calling
     another function that returns the desired composite value.
     Another example is that if we are trying to write a function that
     returns a domain over composite, rather than a plain composite type,
     it is always necessary to write it as returning a single column,
     since there is no way to cause a coercion of the whole row result.
    </para>

    <para>
     We could call this function directly either by using it in
     a value expression:

<screen>
SELECT new_emp();

         new_emp
--------------------------
 (None,1000.0,25,"(2,2)")
</screen>

     or by calling it as a table function:

<screen>
SELECT * FROM new_emp();

 name | salary | age | cubicle
------+--------+-----+---------
 None | 1000.0 |  25 | (2,2)
</screen>

     The second way is described more fully in <xref
     linkend="xfunc-sql-table-functions"/>.
    </para>

    <para>
     When you use a function that returns a composite type,
     you might want only one field (attribute) from its result.
     You can do that with syntax like this:

<screen>
SELECT (new_emp()).name;

 name
------
 None
</screen>

     The extra parentheses are needed to keep the parser from getting
     confused.  If you try to do it without them, you get something like this:

<screen>
SELECT new_emp().name;
ERROR:  syntax error at or near "."
LINE 1: SELECT new_emp().name;
                        ^
</screen>
    </para>

    <para>
     Another option is to use functional notation for extracting an attribute:

<screen>
SELECT name(new_emp());

 name
------
 None
</screen>

     As explained in <xref linkend="rowtypes-usage"/>, the field notation and
     functional notation are equivalent.
    </para>

    <para>
     Another way to use a function returning a composite type is to pass the
     result to another function that accepts the correct row type as input:

<screen>
CREATE FUNCTION getname(emp) RETURNS text AS $$
    SELECT $1.name;
$$ LANGUAGE SQL;

SELECT getname(new_emp());
 getname
---------
 None
(1 row)
</screen>
    </para>
   </sect2>

   <sect2 id="xfunc-output-parameters">
    <title><acronym>SQL</acronym> Functions with Output Parameters</title>

   <indexterm>
    <primary>function</primary>
    <secondary>output parameter</secondary>
   </indexterm>

    <para>
     An alternative way of describing a function's results is to define it
     with <firstterm>output parameters</firstterm>, as in this example:

<screen>
CREATE FUNCTION add_em (IN x int, IN y int, OUT sum int)
AS 'SELECT x + y'
LANGUAGE SQL;

SELECT add_em(3,7);
 add_em
--------
     10
(1 row)
</screen>

     This is not essentially different from the version of <literal>add_em</literal>
     shown in <xref linkend="xfunc-sql-base-functions"/>.  The real value of
     output parameters is that they provide a convenient way of defining
     functions that return several columns.  For example,

<screen>
CREATE FUNCTION sum_n_product (x int, y int, OUT sum int, OUT product int)
AS 'SELECT x + y, x * y'
LANGUAGE SQL;

 SELECT * FROM sum_n_product(11,42);
 sum | product
-----+---------
  53 |     462
(1 row)
</screen>

     What has essentially happened here is that we have created an anonymous
     composite type for the result of the function.  The above example has
     the same end result as

<screen>
CREATE TYPE sum_prod AS (sum int, product int);

CREATE FUNCTION sum_n_product (int, int) RETURNS sum_prod
AS 'SELECT $1 + $2, $1 * $2'
LANGUAGE SQL;
</screen>

     but not having to bother with the separate composite type definition
     is often handy.  Notice that the names attached to the output parameters
     are not just decoration, but determine the column names of the anonymous
     composite type.  (If you omit a name for an output parameter, the
     system will choose a name on its own.)
    </para>

    <para>
     Notice that output parameters are not included in the calling argument
     list when invoking such a function from SQL.  This is because
     <productname>PostgreSQL</productname> considers only the input
     parameters to define the function's calling signature.  That means
     also that only the input parameters matter when referencing the function
     for purposes such as dropping it.  We could drop the above function
     with either of

<screen>
DROP FUNCTION sum_n_product (x int, y int, OUT sum int, OUT product int);
DROP FUNCTION sum_n_product (int, int);
</screen>
    </para>

    <para>
     Parameters can be marked as <literal>IN</literal> (the default),
     <literal>OUT</literal>, <literal>INOUT</literal>, or <literal>VARIADIC</literal>.
     An <literal>INOUT</literal>
     parameter serves as both an input parameter (part of the calling
     argument list) and an output parameter (part of the result record type).
     <literal>VARIADIC</literal> parameters are input parameters, but are treated
     specially as described below.
    </para>
   </sect2>

   <sect2 id="xfunc-output-parameters-proc">
    <title><acronym>SQL</acronym> Procedures with Output Parameters</title>

    <indexterm>
     <primary>procedures</primary>
     <secondary>output parameter</secondary>
    </indexterm>

    <para>
     Output parameters are also supported in procedures, but they work a bit
     differently from functions.  In <command>CALL</command> commands,
     output parameters must be included in the argument list.
     For example, the bank account debiting routine from earlier could be
     written like this:
<programlisting>
CREATE PROCEDURE tp1 (accountno integer, debit numeric, OUT new_balance numeric) AS $$
    UPDATE bank
        SET balance = balance - debit
        WHERE accountno = tp1.accountno
    RETURNING balance;
$$ LANGUAGE SQL;
</programlisting>
     To call this procedure, an argument matching the <literal>OUT</literal>
     parameter must be included.  It's customary to write
     <literal>NULL</literal>:
<programlisting>
CALL tp1(17, 100.0, NULL);
</programlisting>
     If you write something else, it must be an expression that is implicitly
     coercible to the declared type of the parameter, just as for input
     parameters.  Note however that such an expression will not be evaluated.
    </para>

    <para>
     When calling a procedure from <application>PL/pgSQL</application>,
     instead of writing <literal>NULL</literal> you must write a variable
     that will receive the procedure's output.  See <xref
     linkend="plpgsql-statements-calling-procedure"/> for details.
    </para>
   </sect2>

   <sect2 id="xfunc-sql-variadic-functions">
    <title><acronym>SQL</acronym> Functions with Variable Numbers of Arguments</title>

    <indexterm>
     <primary>function</primary>
     <secondary>variadic</secondary>
    </indexterm>

    <indexterm>
     <primary>variadic function</primary>
    </indexterm>

    <para>
     <acronym>SQL</acronym> functions can be declared to accept
     variable numbers of arguments, so long as all the <quote>optional</quote>
     arguments are of the same data type.  The optional arguments will be
     passed to the function as an array.  The function is declared by
     marking the last parameter as <literal>VARIADIC</literal>; this parameter
     must be declared as being of an array type.  For example:

<screen>
CREATE FUNCTION mleast(VARIADIC arr numeric[]) RETURNS numeric AS $$
    SELECT min($1[i]) FROM generate_subscripts($1, 1) g(i);
$$ LANGUAGE SQL;

SELECT mleast(10, -1, 5, 4.4);
 mleast
--------
     -1
(1 row)
</screen>

     Effectively, all the actual arguments at or beyond the
     <literal>VARIADIC</literal> position are gathered up into a one-dimensional
     array, as if you had written

<screen>
SELECT mleast(ARRAY[10, -1, 5, 4.4]);    -- doesn't work
</screen>

     You can't actually write that, though &mdash; or at least, it will
     not match this function definition.  A parameter marked
     <literal>VARIADIC</literal> matches one or more occurrences of its element
     type, not of its own type.
    </para>

    <para>
     Sometimes it is useful to be able to pass an already-constructed array
     to a variadic function; this is particularly handy when one variadic
     function wants to pass on its array parameter to another one.  Also,
     this is the only secure way to call a variadic function found in a schema
     that permits untrusted users to create objects; see
     <xref linkend="typeconv-func"/>.  You can do this by
     specifying <literal>VARIADIC</literal> in the call:

<screen>
SELECT mleast(VARIADIC ARRAY[10, -1, 5, 4.4]);
</screen>

     This prevents expansion of the function's variadic parameter into its
     element type, thereby allowing the array argument value to match
     normally.  <literal>VARIADIC</literal> can only be attached to the last
     actual argument of a function call.
    </para>

    <para>
     Specifying <literal>VARIADIC</literal> in the call is also the only way to
     pass an empty array to a variadic function, for example:

<screen>
SELECT mleast(VARIADIC ARRAY[]::numeric[]);
</screen>

     Simply writing <literal>SELECT mleast()</literal> does not work because a
     variadic parameter must match at least one actual argument.
     (You could define a second function also named <literal>mleast</literal>,
     with no parameters, if you wanted to allow such calls.)
    </para>

    <para>
     The array element parameters generated from a variadic parameter are
     treated as not having any names of their own.  This means it is not
     possible to call a variadic function using named arguments (<xref
     linkend="sql-syntax-calling-funcs"/>), except when you specify
     <literal>VARIADIC</literal>.  For example, this will work:

<screen>
SELECT mleast(VARIADIC arr =&gt; ARRAY[10, -1, 5, 4.4]);
</screen>

     but not these:

<screen>
SELECT mleast(arr =&gt; 10);
SELECT mleast(arr =&gt; ARRAY[10, -1, 5, 4.4]);
</screen>
    </para>
   </sect2>

   <sect2 id="xfunc-sql-parameter-defaults">
    <title><acronym>SQL</acronym> Functions with Default Values for Arguments</title>

    <indexterm>
     <primary>function</primary>
     <secondary>default values for arguments</secondary>
    </indexterm>

    <para>
     Functions can be declared with default values for some or all input
     arguments.  The default values are inserted whenever the function is
     called with insufficiently many actual arguments.  Since arguments
     can only be omitted from the end of the actual argument list, all
     parameters after a parameter with a default value have to have
     default values as well.  (Although the use of named argument notation
     could allow this restriction to be relaxed, it's still enforced so that
     positional argument notation works sensibly.)  Whether or not you use it,
     this capability creates a need for precautions when calling functions in
     databases where some users mistrust other users; see
     <xref linkend="typeconv-func"/>.
    </para>

    <para>
     For example:
<screen>
CREATE FUNCTION foo(a int, b int DEFAULT 2, c int DEFAULT 3)
RETURNS int
LANGUAGE SQL
AS $$
    SELECT $1 + $2 + $3;
$$;

SELECT foo(10, 20, 30);
 foo
-----
  60
(1 row)

SELECT foo(10, 20);
 foo
-----
  33
(1 row)

SELECT foo(10);
 foo
-----
  15
(1 row)

SELECT foo();  -- fails since there is no default for the first argument
ERROR:  function foo() does not exist
</screen>
     The <literal>=</literal> sign can also be used in place of the
     key word <literal>DEFAULT</literal>.
    </para>
   </sect2>

   <sect2 id="xfunc-sql-table-functions">
    <title><acronym>SQL</acronym> Functions as Table Sources</title>

    <para>
     All SQL functions can be used in the <literal>FROM</literal> clause of a query,
     but it is particularly useful for functions returning composite types.
     If the function is defined to return a base type, the table function
     produces a one-column table.  If the function is defined to return
     a composite type, the table function produces a column for each attribute
     of the composite type.
    </para>

    <para>
     Here is an example:

<screen>
CREATE TABLE foo (fooid int, foosubid int, fooname text);
INSERT INTO foo VALUES (1, 1, 'Joe');
INSERT INTO foo VALUES (1, 2, 'Ed');
INSERT INTO foo VALUES (2, 1, 'Mary');

CREATE FUNCTION getfoo(int) RETURNS foo AS $$
    SELECT * FROM foo WHERE fooid = $1;
$$ LANGUAGE SQL;

SELECT *, upper(fooname) FROM getfoo(1) AS t1;

 fooid | foosubid | fooname | upper
-------+----------+---------+-------
     1 |        1 | Joe     | JOE
(1 row)
</screen>

     As the example shows, we can work with the columns of the function's
     result just the same as if they were columns of a regular table.
    </para>

    <para>
     Note that we only got one row out of the function.  This is because
     we did not use <literal>SETOF</literal>.  That is described in the next section.
    </para>
   </sect2>

   <sect2 id="xfunc-sql-functions-returning-set">
    <title><acronym>SQL</acronym> Functions Returning Sets</title>

    <indexterm>
     <primary>function</primary>
     <secondary>with SETOF</secondary>
    </indexterm>

    <para>
     When an SQL function is declared as returning <literal>SETOF
     <replaceable>sometype</replaceable></literal>, the function's final
     query is executed to completion, and each row it
     outputs is returned as an element of the result set.
    </para>

    <para>
     This feature is normally used when calling the function in the <literal>FROM</literal>
     clause.  In this case each row returned by the function becomes
     a row of the table seen by the query.  For example, assume that
     table <literal>foo</literal> has the same contents as above, and we say:

<programlisting>
CREATE FUNCTION getfoo(int) RETURNS SETOF foo AS $$
    SELECT * FROM foo WHERE fooid = $1;
$$ LANGUAGE SQL;

SELECT * FROM getfoo(1) AS t1;
</programlisting>

     Then we would get:
<screen>
 fooid | foosubid | fooname
-------+----------+---------
     1 |        1 | Joe
     1 |        2 | Ed
(2 rows)
</screen>
    </para>

    <para>
     It is also possible to return multiple rows with the columns defined by
     output parameters, like this:

<programlisting>
CREATE TABLE tab (y int, z int);
INSERT INTO tab VALUES (1, 2), (3, 4), (5, 6), (7, 8);

CREATE FUNCTION sum_n_product_with_tab (x int, OUT sum int, OUT product int)
RETURNS SETOF record
AS $$
    SELECT $1 + tab.y, $1 * tab.y FROM tab;
$$ LANGUAGE SQL;

SELECT * FROM sum_n_product_with_tab(10);
 sum | product
-----+---------
  11 |      10
  13 |      30
  15 |      50
  17 |      70
(4 rows)
</programlisting>

     The key point here is that you must write <literal>RETURNS SETOF record</literal>
     to indicate that the function returns multiple rows instead of just one.
     If there is only one output parameter, write that parameter's type
     instead of <type>record</type>.
    </para>

    <para>
     It is frequently useful to construct a query's result by invoking a
     set-returning function multiple times, with the parameters for each
     invocation coming from successive rows of a table or subquery.  The
     preferred way to do this is to use the <literal>LATERAL</literal> key word,
     which is described in <xref linkend="queries-lateral"/>.
     Here is an example using a set-returning function to enumerate
     elements of a tree structure:

<screen>
SELECT * FROM nodes;
   name    | parent
-----------+--------
 Top       |
 Child1    | Top
 Child2    | Top
 Child3    | Top
 SubChild1 | Child1
 SubChild2 | Child1
(6 rows)

CREATE FUNCTION listchildren(text) RETURNS SETOF text AS $$
    SELECT name FROM nodes WHERE parent = $1
$$ LANGUAGE SQL STABLE;

SELECT * FROM listchildren('Top');
 listchildren
--------------
 Child1
 Child2
 Child3
(3 rows)

SELECT name, child FROM nodes, LATERAL listchildren(name) AS child;
  name  |   child
--------+-----------
 Top    | Child1
 Top    | Child2
 Top    | Child3
 Child1 | SubChild1
 Child1 | SubChild2
(5 rows)
</screen>

     This example does not do anything that we couldn't have done with a
     simple join, but in more complex calculations the option to put
     some of the work into a function can be quite convenient.
    </para>

    <para>
     Functions returning sets can also be called in the select list
     of a query.  For each row that the query
     generates by itself, the set-returning function is invoked, and an output
     row is generated for each element of the function's result set.
     The previous example could also be done with queries like
     these:

<screen>
SELECT listchildren('Top');
 listchildren
--------------
 Child1
 Child2
 Child3
(3 rows)

SELECT name, listchildren(name) FROM nodes;
  name  | listchildren
--------+--------------
 Top    | Child1
 Top    | Child2
 Top    | Child3
 Child1 | SubChild1
 Child1 | SubChild2
(5 rows)
</screen>

     In the last <command>SELECT</command>,
     notice that no output row appears for <literal>Child2</literal>, <literal>Child3</literal>, etc.
     This happens because <function>listchildren</function> returns an empty set
     for those arguments, so no result rows are generated.  This is the same
     behavior as we got from an inner join to the function result when using
     the <literal>LATERAL</literal> syntax.
    </para>

    <para>
     <productname>PostgreSQL</productname>'s behavior for a set-returning function in a
     query's select list is almost exactly the same as if the set-returning
     function had been written in a <literal>LATERAL FROM</literal>-clause item
     instead.  For example,
<programlisting>
SELECT x, generate_series(1,5) AS g FROM tab;
</programlisting>
     is almost equivalent to
<programlisting>
SELECT x, g FROM tab, LATERAL generate_series(1,5) AS g;
</programlisting>
     It would be exactly the same, except that in this specific example,
     the planner could choose to put <structname>g</structname> on the outside of the
     nested-loop join, since <structname>g</structname> has no actual lateral dependency
     on <structname>tab</structname>.  That would result in a different output row
     order.  Set-returning functions in the select list are always evaluated
     as though they are on the inside of a nested-loop join with the rest of
     the <literal>FROM</literal> clause, so that the function(s) are run to
     completion before the next row from the <literal>FROM</literal> clause is
     considered.
    </para>

    <para>
     If there is more than one set-returning function in the query's select
     list, the behavior is similar to what you get from putting the functions
     into a single <literal>LATERAL ROWS FROM( ... )</literal> <literal>FROM</literal>-clause
     item.  For each row from the underlying query, there is an output row
     using the first result from each function, then an output row using the
     second result, and so on.  If some of the set-returning functions
     produce fewer outputs than others, null values are substituted for the
     missing data, so that the total number of rows emitted for one
     underlying row is the same as for the set-returning function that
     produced the most outputs.  Thus the set-returning functions
     run <quote>in lockstep</quote> until they are all exhausted, and then
     execution continues with the next underlying row.
    </para>

    <para>
     Set-returning functions can be nested in a select list, although that is
     not allowed in <literal>FROM</literal>-clause items.  In such cases, each level
     of nesting is treated separately, as though it were
     a separate <literal>LATERAL ROWS FROM( ... )</literal> item.  For example, in
<programlisting>
SELECT srf1(srf2(x), srf3(y)), srf4(srf5(z)) FROM tab;
</programlisting>
     the set-returning functions <function>srf2</function>, <function>srf3</function>,
     and <function>srf5</function> would be run in lockstep for each row
     of <structname>tab</structname>, and then <function>srf1</function> and <function>srf4</function>
     would be applied in lockstep to each row produced by the lower
     functions.
    </para>

    <para>
     Set-returning functions cannot be used within conditional-evaluation
     constructs, such as <literal>CASE</literal> or <literal>COALESCE</literal>.  For
     example, consider
<programlisting>
SELECT x, CASE WHEN x &gt; 0 THEN generate_series(1, 5) ELSE 0 END FROM tab;
</programlisting>
     It might seem that this should produce five repetitions of input rows
     that have <literal>x &gt; 0</literal>, and a single repetition of those that do
     not; but actually, because <function>generate_series(1, 5)</function> would be
     run in an implicit <literal>LATERAL FROM</literal> item before
     the <literal>CASE</literal> expression is ever evaluated, it would produce five
     repetitions of every input row.  To reduce confusion, such cases produce
     a parse-time error instead.
    </para>

    <note>
     <para>
      If a function's last command is <command>INSERT</command>,
      <command>UPDATE</command>, <command>DELETE</command>, or
      <command>MERGE</command> with <literal>RETURNING</literal>, that command will
      always be executed to completion, even if the function is not declared
      with <literal>SETOF</literal> or the calling query does not fetch all the
      result rows.  Any extra rows produced by the <literal>RETURNING</literal>
      clause are silently dropped, but the commanded table modifications
      still happen (and are all completed before returning from the function).
     </para>
    </note>

    <note>
     <para>
      Before <productname>PostgreSQL</productname> 10, putting more than one
      set-returning function in the same select list did not behave very
      sensibly unless they always produced equal numbers of rows.  Otherwise,
      what you got was a number of output rows equal to the least common
      multiple of the numbers of rows produced by the set-returning
      functions.  Also, nested set-returning functions did not work as
      described above; instead, a set-returning function could have at most
      one set-returning argument, and each nest of set-returning functions
      was run independently.  Also, conditional execution (set-returning
      functions inside <literal>CASE</literal> etc.) was previously allowed,
      complicating things even more.
      Use of the <literal>LATERAL</literal> syntax is recommended when writing
      queries that need to work in older <productname>PostgreSQL</productname> versions,
      because that will give consistent results across different versions.
      If you have a query that is relying on conditional execution of a
      set-returning function, you may be able to fix it by moving the
      conditional test into a custom set-returning function.  For example,
<programlisting>
SELECT x, CASE WHEN y &gt; 0 THEN generate_series(1, z) ELSE 5 END FROM tab;
</programlisting>
      could become
<programlisting>
CREATE FUNCTION case_generate_series(cond bool, start int, fin int, els int)
  RETURNS SETOF int AS $$
BEGIN
  IF cond THEN
    RETURN QUERY SELECT generate_series(start, fin);
  ELSE
    RETURN QUERY SELECT els;
  END IF;
END$$ LANGUAGE plpgsql;

SELECT x, case_generate_series(y &gt; 0, 1, z, 5) FROM tab;
</programlisting>
      This formulation will work the same in all versions
      of <productname>PostgreSQL</productname>.
     </para>
    </note>
   </sect2>

   <sect2 id="xfunc-sql-functions-returning-table">
    <title><acronym>SQL</acronym> Functions Returning <literal>TABLE</literal></title>

    <indexterm>
     <primary>function</primary>
     <secondary>RETURNS TABLE</secondary>
    </indexterm>

    <para>
     There is another way to declare a function as returning a set,
     which is to use the syntax
     <literal>RETURNS TABLE(<replaceable>columns</replaceable>)</literal>.
     This is equivalent to using one or more <literal>OUT</literal> parameters plus
     marking the function as returning <literal>SETOF record</literal> (or
     <literal>SETOF</literal> a single output parameter's type, as appropriate).
     This notation is specified in recent versions of the SQL standard, and
     thus may be more portable than using <literal>SETOF</literal>.
    </para>

    <para>
     For example, the preceding sum-and-product example could also be
     done this way:

<programlisting>
CREATE FUNCTION sum_n_product_with_tab (x int)
RETURNS TABLE(sum int, product int) AS $$
    SELECT $1 + tab.y, $1 * tab.y FROM tab;
$$ LANGUAGE SQL;
</programlisting>

     It is not allowed to use explicit <literal>OUT</literal> or <literal>INOUT</literal>
     parameters with the <literal>RETURNS TABLE</literal> notation &mdash; you must
     put all the output columns in the <literal>TABLE</literal> list.
    </para>
   </sect2>

   <sect2 id="xfunc-sql-polymorphic-functions">
    <title>Polymorphic <acronym>SQL</acronym> Functions</title>

    <para>
     <acronym>SQL</acronym> functions can be declared to accept and
     return the polymorphic types described in <xref
     linkend="extend-types-polymorphic"/>.  Here is a polymorphic
     function <function>make_array</function> that builds up an array
     from two arbitrary data type elements:
<screen>
CREATE FUNCTION make_array(anyelement, anyelement) RETURNS anyarray AS $$
    SELECT ARRAY[$1, $2];
$$ LANGUAGE SQL;

SELECT make_array(1, 2) AS intarray, make_array('a'::text, 'b') AS textarray;
 intarray | textarray
----------+-----------
 {1,2}    | {a,b}
(1 row)
</screen>
    </para>

    <para>
     Notice the use of the typecast <literal>'a'::text</literal>
     to specify that the argument is of type <type>text</type>. This is
     required if the argument is just a string literal, since otherwise
     it would be treated as type
     <type>unknown</type>, and array of <type>unknown</type> is not a valid
     type.
     Without the typecast, you will get errors like this:
<screen>
ERROR:  could not determine polymorphic type because input has type unknown
</screen>
    </para>

    <para>
     With <function>make_array</function> declared as above, you must
     provide two arguments that are of exactly the same data type; the
     system will not attempt to resolve any type differences.  Thus for
     example this does not work:
<screen>
SELECT make_array(1, 2.5) AS numericarray;
ERROR:  function make_array(integer, numeric) does not exist
</screen>
     An alternative approach is to use the <quote>common</quote> family of
     polymorphic types, which allows the system to try to identify a
     suitable common type:
<screen>
CREATE FUNCTION make_array2(anycompatible, anycompatible)
RETURNS anycompatiblearray AS $$
    SELECT ARRAY[$1, $2];
$$ LANGUAGE SQL;

SELECT make_array2(1, 2.5) AS numericarray;
 numericarray
--------------
 {1,2.5}
(1 row)
</screen>
     Because the rules for common type resolution default to choosing
     type <type>text</type> when all inputs are of unknown types, this
     also works:
<screen>
SELECT make_array2('a', 'b') AS textarray;
 textarray
-----------
 {a,b}
(1 row)
</screen>
    </para>

    <para>
     It is permitted to have polymorphic arguments with a fixed
     return type, but the converse is not. For example:
<screen>
CREATE FUNCTION is_greater(anyelement, anyelement) RETURNS boolean AS $$
    SELECT $1 &gt; $2;
$$ LANGUAGE SQL;

SELECT is_greater(1, 2);
 is_greater
------------
 f
(1 row)

CREATE FUNCTION invalid_func() RETURNS anyelement AS $$
    SELECT 1;
$$ LANGUAGE SQL;
ERROR:  cannot determine result data type
DETAIL:  A result of type anyelement requires at least one input of type anyelement, anyarray, anynonarray, anyenum, or anyrange.
</screen>
    </para>

    <para>
     Polymorphism can be used with functions that have output arguments.
     For example:
<screen>
CREATE FUNCTION dup (f1 anyelement, OUT f2 anyelement, OUT f3 anyarray)
AS 'select $1, array[$1,$1]' LANGUAGE SQL;

SELECT * FROM dup(22);
 f2 |   f3
----+---------
 22 | {22,22}
(1 row)
</screen>
    </para>

    <para>
     Polymorphism can also be used with variadic functions.
     For example:
<screen>
CREATE FUNCTION anyleast (VARIADIC anyarray) RETURNS anyelement AS $$
    SELECT min($1[i]) FROM generate_subscripts($1, 1) g(i);
$$ LANGUAGE SQL;

SELECT anyleast(10, -1, 5, 4);
 anyleast
----------
       -1
(1 row)

SELECT anyleast('abc'::text, 'def');
 anyleast
----------
 abc
(1 row)

CREATE FUNCTION concat_values(text, VARIADIC anyarray) RETURNS text AS $$
    SELECT array_to_string($2, $1);
$$ LANGUAGE SQL;

SELECT concat_values('|', 1, 4, 2);
 concat_values
---------------
 1|4|2
(1 row)
</screen>
    </para>
   </sect2>

   <sect2 id="xfunc-sql-collations">
    <title><acronym>SQL</acronym> Functions with Collations</title>

    <indexterm>
     <primary>collation</primary>
     <secondary>in SQL functions</secondary>
    </indexterm>

    <para>
     When an SQL function has one or more parameters of collatable data types,
     a collation is identified for each function call depending on the
     collations assigned to the actual arguments, as described in <xref
     linkend="collation"/>.  If a collation is successfully identified
     (i.e., there are no conflicts of implicit collations among the arguments)
     then all the collatable parameters are treated as having that collation
     implicitly.  This will affect the behavior of collation-sensitive
     operations within the function.  For example, using the
     <function>anyleast</function> function described above, the result of
<programlisting>
SELECT anyleast('abc'::text, 'ABC');
</programlisting>
     will depend on the database's default collation.  In <literal>C</literal> locale
     the result will be <literal>ABC</literal>, but in many other locales it will
     be <literal>abc</literal>.  The collation to use can be forced by adding
     a <literal>COLLATE</literal> clause to any of the arguments, for example
<programlisting>
SELECT anyleast('abc'::text, 'ABC' COLLATE "C");
</programlisting>
     Alternatively, if you wish a function to operate with a particular
     collation regardless of what it is called with, insert
     <literal>COLLATE</literal> clauses as needed in the function definition.
     This version of <function>anyleast</function> would always use <literal>en_US</literal>
     locale to compare strings:
<programlisting>
CREATE FUNCTION anyleast (VARIADIC anyarray) RETURNS anyelement AS $$
    SELECT min($1[i] COLLATE "en_US") FROM generate_subscripts($1, 1) g(i);
$$ LANGUAGE SQL;
</programlisting>
     But note that this will throw an error if applied to a non-collatable
     data type.
    </para>

    <para>
     If no common collation can be identified among the actual arguments,
     then an SQL function treats its parameters as having their data types'
     default collation (which is usually the database's default collation,
     but could be different for parameters of domain types).
    </para>

    <para>
     The behavior of collatable parameters can be thought of as a limited
     form of polymorphism, applicable only to textual data types.
    </para>
   </sect2>
  </sect1>

  <sect1 id="xfunc-overload">
   <title>Function Overloading</title>

   <indexterm zone="xfunc-overload">
    <primary>overloading</primary>
    <secondary>functions</secondary>
   </indexterm>

   <para>
    More than one function can be defined with the same SQL name, so long
    as the arguments they take are different.  In other words,
    function names can be <firstterm>overloaded</firstterm>.  Whether or not
    you use it, this capability entails security precautions when calling
    functions in databases where some users mistrust other users; see
    <xref linkend="typeconv-func"/>.  When a query is executed, the server
    will determine which function to call from the data types and the number
    of the provided arguments.  Overloading can also be used to simulate
    functions with a variable number of arguments, up to a finite maximum
    number.
   </para>

   <para>
    When creating a family of overloaded functions, one should be
    careful not to create ambiguities.  For instance, given the
    functions:
<programlisting>
CREATE FUNCTION test(int, real) RETURNS ...
CREATE FUNCTION test(smallint, double precision) RETURNS ...
</programlisting>
    it is not immediately clear which function would be called with
    some trivial input like <literal>test(1, 1.5)</literal>.  The
    currently implemented resolution rules are described in
    <xref linkend="typeconv"/>, but it is unwise to design a system that subtly
    relies on this behavior.
   </para>

   <para>
    A function that takes a single argument of a composite type should
    generally not have the same name as any attribute (field) of that type.
    Recall that <literal><replaceable>attribute</replaceable>(<replaceable>table</replaceable>)</literal>
    is considered equivalent
    to <literal><replaceable>table</replaceable>.<replaceable>attribute</replaceable></literal>.
    In the case that there is an
    ambiguity between a function on a composite type and an attribute of
    the composite type, the attribute will always be used.  It is possible
    to override that choice by schema-qualifying the function name
    (that is, <literal><replaceable>schema</replaceable>.<replaceable>func</replaceable>(<replaceable>table</replaceable>)
    </literal>) but it's better to
    avoid the problem by not choosing conflicting names.
   </para>

   <para>
    Another possible conflict is between variadic and non-variadic functions.
    For instance, it is possible to create both <literal>foo(numeric)</literal> and
    <literal>foo(VARIADIC numeric[])</literal>.  In this case it is unclear which one
    should be matched to a call providing a single numeric argument, such as
    <literal>foo(10.1)</literal>.  The rule is that the function appearing
    earlier in the search path is used, or if the two functions are in the
    same schema, the non-variadic one is preferred.
   </para>

   <para>
    When overloading C-language functions, there is an additional
    constraint: The C name of each function in the family of
    overloaded functions must be different from the C names of all
    other functions, either internal or dynamically loaded.  If this
    rule is violated, the behavior is not portable.  You might get a
    run-time linker error, or one of the functions will get called
    (usually the internal one).  The alternative form of the
    <literal>AS</literal> clause for the SQL <command>CREATE
    FUNCTION</command> command decouples the SQL function name from
    the function name in the C source code.  For instance:
<programlisting>
CREATE FUNCTION test(int) RETURNS int
    AS '<replaceable>filename</replaceable>', 'test_1arg'
    LANGUAGE C;
CREATE FUNCTION test(int, int) RETURNS int
    AS '<replaceable>filename</replaceable>', 'test_2arg'
    LANGUAGE C;
</programlisting>
    The names of the C functions here reflect one of many possible conventions.
   </para>
  </sect1>

  <sect1 id="xfunc-volatility">
   <title>Function Volatility Categories</title>

   <indexterm zone="xfunc-volatility">
    <primary>volatility</primary>
    <secondary>functions</secondary>
   </indexterm>
   <indexterm zone="xfunc-volatility">
    <primary>VOLATILE</primary>
   </indexterm>
   <indexterm zone="xfunc-volatility">
    <primary>STABLE</primary>
   </indexterm>
   <indexterm zone="xfunc-volatility">
    <primary>IMMUTABLE</primary>
   </indexterm>

   <para>
    Every function has a <firstterm>volatility</firstterm> classification, with
    the possibilities being <literal>VOLATILE</literal>, <literal>STABLE</literal>, or
    <literal>IMMUTABLE</literal>.  <literal>VOLATILE</literal> is the default if the
    <link linkend="sql-createfunction"><command>CREATE FUNCTION</command></link>
    command does not specify a category.  The volatility category is a
    promise to the optimizer about the behavior of the function:

   <itemizedlist>
    <listitem>
     <para>
      A <literal>VOLATILE</literal> function can do anything, including modifying
      the database.  It can return different results on successive calls with
      the same arguments.  The optimizer makes no assumptions about the
      behavior of such functions.  A query using a volatile function will
      re-evaluate the function at every row where its value is needed.
     </para>
    </listitem>
    <listitem>
     <para>
      A <literal>STABLE</literal> function cannot modify the database and is
      guaranteed to return the same results given the same arguments
      for all rows within a single statement. This category allows the
      optimizer to optimize multiple calls of the function to a single
      call. In particular, it is safe to use an expression containing
      such a function in an index scan condition. (Since an index scan
      will evaluate the comparison value only once, not once at each
      row, it is not valid to use a <literal>VOLATILE</literal> function in an
      index scan condition.)
     </para>
    </listitem>
    <listitem>
     <para>
      An <literal>IMMUTABLE</literal> function cannot modify the database and is
      guaranteed to return the same results given the same arguments forever.
      This category allows the optimizer to pre-evaluate the function when
      a query calls it with constant arguments.  For example, a query like
      <literal>SELECT ... WHERE x = 2 + 2</literal> can be simplified on sight to
      <literal>SELECT ... WHERE x = 4</literal>, because the function underlying
      the integer addition operator is marked <literal>IMMUTABLE</literal>.
     </para>
    </listitem>
   </itemizedlist>
   </para>

   <para>
    For best optimization results, you should label your functions with the
    strictest volatility category that is valid for them.
   </para>

   <para>
    Any function with side-effects <emphasis>must</emphasis> be labeled
    <literal>VOLATILE</literal>, so that calls to it cannot be optimized away.
    Even a function with no side-effects needs to be labeled
    <literal>VOLATILE</literal> if its value can change within a single query;
    some examples are <literal>random()</literal>, <literal>currval()</literal>,
    <literal>timeofday()</literal>.
   </para>

   <para>
    Another important example is that the <function>current_timestamp</function>
    family of functions qualify as <literal>STABLE</literal>, since their values do
    not change within a transaction.
   </para>

   <para>
    There is relatively little difference between <literal>STABLE</literal> and
    <literal>IMMUTABLE</literal> categories when considering simple interactive
    queries that are planned and immediately executed: it doesn't matter
    a lot whether a function is executed once during planning or once during
    query execution startup.  But there is a big difference if the plan is
    saved and reused later.  Labeling a function <literal>IMMUTABLE</literal> when
    it really isn't might allow it to be prematurely folded to a constant during
    planning, resulting in a stale value being re-used during subsequent uses
    of the plan.  This is a hazard when using prepared statements or when
    using function languages that cache plans (such as
    <application>PL/pgSQL</application>).
   </para>

   <para>
    For functions written in SQL or in any of the standard procedural
    languages, there is a second important property determined by the
    volatility category, namely the visibility of any data changes that have
    been made by the SQL command that is calling the function.  A
    <literal>VOLATILE</literal> function will see such changes, a <literal>STABLE</literal>
    or <literal>IMMUTABLE</literal> function will not.  This behavior is implemented
    using the snapshotting behavior of MVCC (see <xref linkend="mvcc"/>):
    <literal>STABLE</literal> and <literal>IMMUTABLE</literal> functions use a snapshot
    established as of the start of the calling query, whereas
    <literal>VOLATILE</literal> functions obtain a fresh snapshot at the start of
    each query they execute.
   </para>

   <note>
    <para>
     Functions written in C can manage snapshots however they want, but it's
     usually a good idea to make C functions work this way too.
    </para>
   </note>

   <para>
    Because of this snapshotting behavior,
    a function containing only <command>SELECT</command> commands can safely be
    marked <literal>STABLE</literal>, even if it selects from tables that might be
    undergoing modifications by concurrent queries.
    <productname>PostgreSQL</productname> will execute all commands of a
    <literal>STABLE</literal> function using the snapshot established for the
    calling query, and so it will see a fixed view of the database throughout
    that query.
   </para>

   <para>
    The same snapshotting behavior is used for <command>SELECT</command> commands
    within <literal>IMMUTABLE</literal> functions.  It is generally unwise to select
    from database tables within an <literal>IMMUTABLE</literal> function at all,
    since the immutability will be broken if the table contents ever change.
    However, <productname>PostgreSQL</productname> does not enforce that you
    do not do that.
   </para>

   <para>
    A common error is to label a function <literal>IMMUTABLE</literal> when its
    results depend on a configuration parameter.  For example, a function
    that manipulates timestamps might well have results that depend on the
    <xref linkend="guc-timezone"/> setting.  For safety, such functions should
    be labeled <literal>STABLE</literal> instead.
   </para>

   <note>
    <para>
     <productname>PostgreSQL</productname> requires that <literal>STABLE</literal>
     and <literal>IMMUTABLE</literal> functions contain no SQL commands other
     than <command>SELECT</command> to prevent data modification.
     (This is not a completely bulletproof test, since such functions could
     still call <literal>VOLATILE</literal> functions that modify the database.
     If you do that, you will find that the <literal>STABLE</literal> or
     <literal>IMMUTABLE</literal> function does not notice the database changes
     applied by the called function, since they are hidden from its snapshot.)
    </para>
   </note>
  </sect1>

  <sect1 id="xfunc-pl">
   <title>Procedural Language Functions</title>

   <para>
    <productname>PostgreSQL</productname> allows user-defined functions
    to be written in other languages besides SQL and C.  These other
    languages are generically called <firstterm>procedural
    languages</firstterm> (<acronym>PL</acronym>s).
    Procedural languages aren't built into the
    <productname>PostgreSQL</productname> server; they are offered
    by loadable modules.
    See <xref linkend="xplang"/> and following chapters for more
    information.
   </para>
  </sect1>

  <sect1 id="xfunc-internal">
   <title>Internal Functions</title>

   <indexterm zone="xfunc-internal"><primary>function</primary><secondary>internal</secondary></indexterm>

   <para>
    Internal functions are functions written in C that have been statically
    linked into the <productname>PostgreSQL</productname> server.
    The <quote>body</quote> of the function definition
    specifies the C-language name of the function, which need not be the
    same as the name being declared for SQL use.
    (For reasons of backward compatibility, an empty body
    is accepted as meaning that the C-language function name is the
    same as the SQL name.)
   </para>

   <para>
    Normally, all internal functions present in the
    server are declared during the initialization of the database cluster
    (see <xref linkend="creating-cluster"/>),
    but a user could use <command>CREATE FUNCTION</command>
    to create additional alias names for an internal function.
    Internal functions are declared in <command>CREATE FUNCTION</command>
    with language name <literal>internal</literal>.  For instance, to
    create an alias for the <function>sqrt</function> function:
<programlisting>
CREATE FUNCTION square_root(double precision) RETURNS double precision
    AS 'dsqrt'
    LANGUAGE internal
    STRICT;
</programlisting>
    (Most internal functions expect to be declared <quote>strict</quote>.)
   </para>

   <note>
    <para>
     Not all <quote>predefined</quote> functions are
     <quote>internal</quote> in the above sense.  Some predefined
     functions are written in SQL.
    </para>
   </note>
  </sect1>

  <sect1 id="xfunc-c">
   <title>C-Language Functions</title>

   <indexterm zone="xfunc-c">
    <primary>function</primary>
    <secondary>user-defined</secondary>
    <tertiary>in C</tertiary>
   </indexterm>

   <para>
    User-defined functions can be written in C (or a language that can
    be made compatible with C, such as C++).  Such functions are
    compiled into dynamically loadable objects (also called shared
    libraries) and are loaded by the server on demand.  The dynamic
    loading feature is what distinguishes <quote>C language</quote> functions
    from <quote>internal</quote> functions &mdash; the actual coding conventions
    are essentially the same for both.  (Hence, the standard internal
    function library is a rich source of coding examples for user-defined
    C functions.)
   </para>

   <para>
    Currently only one calling convention is used for C functions
    (<quote>version 1</quote>). Support for that calling convention is
    indicated by writing a <literal>PG_FUNCTION_INFO_V1()</literal> macro
    call for the function, as illustrated below.
   </para>

  <sect2 id="xfunc-c-dynload">
   <title>Dynamic Loading</title>

   <indexterm zone="xfunc-c-dynload">
    <primary>dynamic loading</primary>
   </indexterm>

   <para>
    The first time a user-defined function in a particular
    loadable object file is called in a session,
    the dynamic loader loads that object file into memory so that the
    function can be called.  The <command>CREATE FUNCTION</command>
    for a user-defined C function must therefore specify two pieces of
    information for the function: the name of the loadable
    object file, and the C name (link symbol) of the specific function to call
    within that object file.  If the C name is not explicitly specified then
    it is assumed to be the same as the SQL function name.
   </para>

   <para>
    The following algorithm is used to locate the shared object file
    based on the name given in the <command>CREATE FUNCTION</command>
    command:

    <orderedlist>
     <listitem>
      <para>
       If the name is an absolute path, the given file is loaded.
      </para>
     </listitem>

     <listitem>
      <para>
       If the name starts with the string <literal>$libdir</literal>,
       that part is replaced by the <productname>PostgreSQL</productname> package
        library directory
       name, which is determined at build time.<indexterm><primary>$libdir</primary></indexterm>
      </para>
     </listitem>

     <listitem>
      <para>
       If the name does not contain a directory part, the file is
       searched for in the path specified by the configuration variable
       <xref linkend="guc-dynamic-library-path"/>.<indexterm><primary>dynamic_library_path</primary></indexterm>
      </para>
     </listitem>

     <listitem>
      <para>
       Otherwise (the file was not found in the path, or it contains a
       non-absolute directory part), the dynamic loader will try to
       take the name as given, which will most likely fail.  (It is
       unreliable to depend on the current working directory.)
      </para>
     </listitem>
    </orderedlist>

    If this sequence does not work, the platform-specific shared
    library file name extension (often <filename>.so</filename>) is
    appended to the given name and this sequence is tried again.  If
    that fails as well, the load will fail.
   </para>

   <para>
    It is recommended to locate shared libraries either relative to
    <literal>$libdir</literal> or through the dynamic library path.
    This simplifies version upgrades if the new installation is at a
    different location.  The actual directory that
    <literal>$libdir</literal> stands for can be found out with the
    command <literal>pg_config --pkglibdir</literal>.
   </para>

   <para>
    The user ID the <productname>PostgreSQL</productname> server runs
    as must be able to traverse the path to the file you intend to
    load.  Making the file or a higher-level directory not readable
    and/or not executable by the <systemitem>postgres</systemitem>
    user is a common mistake.
   </para>

   <para>
    In any case, the file name that is given in the
    <command>CREATE FUNCTION</command> command is recorded literally
    in the system catalogs, so if the file needs to be loaded again
    the same procedure is applied.
   </para>

   <note>
    <para>
     <productname>PostgreSQL</productname> will not compile a C function
     automatically.  The object file must be compiled before it is referenced
     in a <command>CREATE
     FUNCTION</command> command.  See <xref linkend="dfunc"/> for additional
     information.
    </para>
   </note>

   <indexterm zone="xfunc-c-dynload">
    <primary>magic block</primary>
   </indexterm>

   <para>
    To ensure that a dynamically loaded object file is not loaded into an
    incompatible server, <productname>PostgreSQL</productname> checks that the
    file contains a <quote>magic block</quote> with the appropriate contents.
    This allows the server to detect obvious incompatibilities, such as code
    compiled for a different major version of
    <productname>PostgreSQL</productname>. To include a magic block,
    write this in one (and only one) of the module source files, after having
    included the header <filename>fmgr.h</filename>:

<programlisting>
PG_MODULE_MAGIC;
</programlisting>
   </para>

   <para>
    After it is used for the first time, a dynamically loaded object
    file is retained in memory.  Future calls in the same session to
    the function(s) in that file will only incur the small overhead of
    a symbol table lookup.  If you need to force a reload of an object
    file, for example after recompiling it, begin a fresh session.
   </para>

   <indexterm zone="xfunc-c-dynload">
    <primary>_PG_init</primary>
   </indexterm>
   <indexterm zone="xfunc-c-dynload">
    <primary>library initialization function</primary>
   </indexterm>

   <para>
    Optionally, a dynamically loaded file can contain an initialization
    function.  If the file includes a function named
    <function>_PG_init</function>, that function will be called immediately after
    loading the file.  The function receives no parameters and should
    return void.  There is presently no way to unload a dynamically loaded file.
   </para>

  </sect2>

   <sect2 id="xfunc-c-basetype">
    <title>Base Types in C-Language Functions</title>

    <indexterm zone="xfunc-c-basetype">
     <primary>data type</primary>
     <secondary>internal organization</secondary>
    </indexterm>

    <para>
     To know how to write C-language functions, you need to know how
     <productname>PostgreSQL</productname> internally represents base
     data types and how they can be passed to and from functions.
     Internally, <productname>PostgreSQL</productname> regards a base
     type as a <quote>blob of memory</quote>.  The user-defined
     functions that you define over a type in turn define the way that
     <productname>PostgreSQL</productname> can operate on it.  That
     is, <productname>PostgreSQL</productname> will only store and
     retrieve the data from disk and use your user-defined functions
     to input, process, and output the data.
    </para>

    <para>
     Base types can have one of three internal formats:

     <itemizedlist>
      <listitem>
       <para>
        pass by value, fixed-length
       </para>
      </listitem>
      <listitem>
       <para>
        pass by reference, fixed-length
       </para>
      </listitem>
      <listitem>
       <para>
        pass by reference, variable-length
       </para>
      </listitem>
     </itemizedlist>
    </para>

    <para>
     By-value  types  can  only be 1, 2, or 4 bytes in length
     (also 8 bytes, if <literal>sizeof(Datum)</literal> is 8 on your machine).
     You should be careful to define your types such that they will be the
     same size (in bytes) on all architectures.  For example, the
     <literal>long</literal> type is dangerous because it is 4 bytes on some
     machines and 8 bytes on others, whereas <type>int</type> type is 4 bytes
     on most Unix machines.  A reasonable implementation of the
     <type>int4</type> type on Unix machines might be:

<programlisting>
/* 4-byte integer, passed by value */
typedef int int4;
</programlisting>

     (The actual PostgreSQL C code calls this type <type>int32</type>, because
     it is a convention in C that <type>int<replaceable>XX</replaceable></type>
     means <replaceable>XX</replaceable> <emphasis>bits</emphasis>.  Note
     therefore also that the C type <type>int8</type> is 1 byte in size.  The
     SQL type <type>int8</type> is called <type>int64</type> in C.  See also
     <xref linkend="xfunc-c-type-table"/>.)
    </para>

    <para>
     On  the  other hand, fixed-length types of any size can
     be passed by-reference.  For example, here is a  sample
     implementation of a <productname>PostgreSQL</productname> type:

<programlisting>
/* 16-byte structure, passed by reference */
typedef struct
{
    double  x, y;
} Point;
</programlisting>

     Only  pointers  to  such types can be used when passing
     them in and out of <productname>PostgreSQL</productname> functions.
     To return a value of such a type, allocate the right amount of
     memory with <literal>palloc</literal>, fill in the allocated memory,
     and return a pointer to it.  (Also, if you just want to return the
     same value as one of your input arguments that's of the same data type,
     you can skip the extra <literal>palloc</literal> and just return the
     pointer to the input value.)
    </para>

    <para>
     Finally, all variable-length types must also be  passed
     by  reference.   All  variable-length  types must begin
     with an opaque length field of exactly 4 bytes, which will be set
     by <symbol>SET_VARSIZE</symbol>; never set this field directly! All data to
     be  stored within that type must be located in the memory
     immediately  following  that  length  field.   The
     length field contains the total length of the structure,
     that is,  it  includes  the  size  of  the  length  field
     itself.
    </para>

    <para>
     Another important point is to avoid leaving any uninitialized bits
     within data type values; for example, take care to zero out any
     alignment padding bytes that might be present in structs.  Without
     this, logically-equivalent constants of your data type might be
     seen as unequal by the planner, leading to inefficient (though not
     incorrect) plans.
    </para>

    <warning>
     <para>
      <emphasis>Never</emphasis> modify the contents of a pass-by-reference input
      value.  If you do so you are likely to corrupt on-disk data, since
      the pointer you are given might point directly into a disk buffer.
      The sole exception to this rule is explained in
      <xref linkend="xaggr"/>.
     </para>
    </warning>

    <para>
     As an example, we can define the type <type>text</type> as
     follows:

<programlisting>
typedef struct {
    int32 length;
    char data[FLEXIBLE_ARRAY_MEMBER];
} text;
</programlisting>

     The <literal>[FLEXIBLE_ARRAY_MEMBER]</literal> notation means that the actual
     length of the data part is not specified by this declaration.
    </para>

    <para>
     When manipulating
     variable-length types, we must  be  careful  to  allocate
     the  correct amount  of memory and set the length field correctly.
     For example, if we wanted to  store  40  bytes  in  a <structname>text</structname>
     structure, we might use a code fragment like this:

<programlisting><![CDATA[
#include "postgres.h"
...
char buffer[40]; /* our source data */
...
text *destination = (text *) palloc(VARHDRSZ + 40);
SET_VARSIZE(destination, VARHDRSZ + 40);
memcpy(destination->data, buffer, 40);
...
]]>
</programlisting>

     <literal>VARHDRSZ</literal> is the same as <literal>sizeof(int32)</literal>, but
     it's considered good style to use the macro <literal>VARHDRSZ</literal>
     to refer to the size of the overhead for a variable-length type.
     Also, the length field <emphasis>must</emphasis> be set using the
     <literal>SET_VARSIZE</literal> macro, not by simple assignment.
    </para>

    <para>
     <xref linkend="xfunc-c-type-table"/> shows the C types
     corresponding to many of the built-in SQL data types
     of <productname>PostgreSQL</productname>.
     The <quote>Defined In</quote> column gives the header file that
     needs to be included to get the type definition.  (The actual
     definition might be in a different file that is included by the
     listed file.  It is recommended that users stick to the defined
     interface.)  Note that you should always include
     <filename>postgres.h</filename> first in any source file of server
     code, because it declares a number of things that you will need
     anyway, and because including other headers first can cause
     portability issues.
    </para>

     <table tocentry="1" id="xfunc-c-type-table">
      <title>Equivalent C Types for Built-in SQL Types</title>
      <tgroup cols="3">
       <colspec colname="col1" colwidth="1*"/>
       <colspec colname="col2" colwidth="1*"/>
       <colspec colname="col3" colwidth="2*"/>
       <thead>
        <row>
         <entry>
          SQL Type
         </entry>
         <entry>
          C Type
         </entry>
         <entry>
          Defined In
         </entry>
        </row>
       </thead>
       <tbody>
        <row>
         <entry><type>boolean</type></entry>
         <entry><type>bool</type></entry>
         <entry><filename>postgres.h</filename> (maybe compiler built-in)</entry>
        </row>
        <row>
         <entry><type>box</type></entry>
         <entry><type>BOX*</type></entry>
         <entry><filename>utils/geo_decls.h</filename></entry>
        </row>
        <row>
         <entry><type>bytea</type></entry>
         <entry><type>bytea*</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>"char"</type></entry>
         <entry><type>char</type></entry>
         <entry>(compiler built-in)</entry>
        </row>
        <row>
         <entry><type>character</type></entry>
         <entry><type>BpChar*</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>cid</type></entry>
         <entry><type>CommandId</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>date</type></entry>
         <entry><type>DateADT</type></entry>
         <entry><filename>utils/date.h</filename></entry>
        </row>
        <row>
         <entry><type>float4</type> (<type>real</type>)</entry>
         <entry><type>float4</type></entry>
        <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>float8</type> (<type>double precision</type>)</entry>
         <entry><type>float8</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>int2</type> (<type>smallint</type>)</entry>
         <entry><type>int16</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>int4</type> (<type>integer</type>)</entry>
         <entry><type>int32</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>int8</type> (<type>bigint</type>)</entry>
         <entry><type>int64</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>interval</type></entry>
         <entry><type>Interval*</type></entry>
         <entry><filename>datatype/timestamp.h</filename></entry>
        </row>
        <row>
         <entry><type>lseg</type></entry>
         <entry><type>LSEG*</type></entry>
         <entry><filename>utils/geo_decls.h</filename></entry>
        </row>
        <row>
         <entry><type>name</type></entry>
         <entry><type>Name</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>numeric</type></entry>
         <entry><type>Numeric</type></entry>
         <entry><filename>utils/numeric.h</filename></entry>
        </row>
        <row>
         <entry><type>oid</type></entry>
         <entry><type>Oid</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>oidvector</type></entry>
         <entry><type>oidvector*</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>path</type></entry>
         <entry><type>PATH*</type></entry>
         <entry><filename>utils/geo_decls.h</filename></entry>
        </row>
        <row>
         <entry><type>point</type></entry>
         <entry><type>POINT*</type></entry>
         <entry><filename>utils/geo_decls.h</filename></entry>
        </row>
        <row>
         <entry><type>regproc</type></entry>
         <entry><type>RegProcedure</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>text</type></entry>
         <entry><type>text*</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>tid</type></entry>
         <entry><type>ItemPointer</type></entry>
         <entry><filename>storage/itemptr.h</filename></entry>
        </row>
        <row>
         <entry><type>time</type></entry>
         <entry><type>TimeADT</type></entry>
         <entry><filename>utils/date.h</filename></entry>
        </row>
        <row>
         <entry><type>time with time zone</type></entry>
         <entry><type>TimeTzADT</type></entry>
         <entry><filename>utils/date.h</filename></entry>
        </row>
        <row>
         <entry><type>timestamp</type></entry>
         <entry><type>Timestamp</type></entry>
         <entry><filename>datatype/timestamp.h</filename></entry>
        </row>
        <row>
         <entry><type>timestamp with time zone</type></entry>
         <entry><type>TimestampTz</type></entry>
         <entry><filename>datatype/timestamp.h</filename></entry>
        </row>
        <row>
         <entry><type>varchar</type></entry>
         <entry><type>VarChar*</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
        <row>
         <entry><type>xid</type></entry>
         <entry><type>TransactionId</type></entry>
         <entry><filename>postgres.h</filename></entry>
        </row>
       </tbody>
      </tgroup>
     </table>

    <para>
     Now that we've gone over all of the possible structures
     for base types, we can show some examples of real functions.
    </para>
   </sect2>

   <sect2 id="xfunc-c-v1-call-conv">
    <title>Version 1 Calling Conventions</title>

    <para>
     The version-1 calling convention relies on macros to suppress most
     of the complexity of passing arguments and results.  The C declaration
     of a version-1 function is always:
<programlisting>
Datum funcname(PG_FUNCTION_ARGS)
</programlisting>
     In addition, the macro call:
<programlisting>
PG_FUNCTION_INFO_V1(funcname);
</programlisting>
     must appear in the same source file.  (Conventionally, it's
     written just before the function itself.)  This macro call is not
     needed for <literal>internal</literal>-language functions, since
     <productname>PostgreSQL</productname> assumes that all internal functions
     use the version-1 convention.  It is, however, required for
     dynamically-loaded functions.
    </para>

    <para>
     In a version-1 function, each actual argument is fetched using a
     <function>PG_GETARG_<replaceable>xxx</replaceable>()</function>
     macro that corresponds to the argument's data type.  (In non-strict
     functions there needs to be a previous check about argument null-ness
     using <function>PG_ARGISNULL()</function>; see below.)
     The result is returned using a
     <function>PG_RETURN_<replaceable>xxx</replaceable>()</function>
     macro for the return type.
     <function>PG_GETARG_<replaceable>xxx</replaceable>()</function>
     takes as its argument the number of the function argument to
     fetch, where the count starts at 0.
     <function>PG_RETURN_<replaceable>xxx</replaceable>()</function>
     takes as its argument the actual value to return.
    </para>

    <para>
     Here are some examples using the version-1 calling convention:
    </para>

<programlisting><![CDATA[
#include "postgres.h"
#include <string.h>
#include "fmgr.h"
#include "utils/geo_decls.h"
#include "varatt.h"

PG_MODULE_MAGIC;

/* by value */

PG_FUNCTION_INFO_V1(add_one);

Datum
add_one(PG_FUNCTION_ARGS)
{
    int32   arg = PG_GETARG_INT32(0);

    PG_RETURN_INT32(arg + 1);
}

/* by reference, fixed length */

PG_FUNCTION_INFO_V1(add_one_float8);

Datum
add_one_float8(PG_FUNCTION_ARGS)
{
    /* The macros for FLOAT8 hide its pass-by-reference nature. */
    float8   arg = PG_GETARG_FLOAT8(0);

    PG_RETURN_FLOAT8(arg + 1.0);
}

PG_FUNCTION_INFO_V1(makepoint);

Datum
makepoint(PG_FUNCTION_ARGS)
{
    /* Here, the pass-by-reference nature of Point is not hidden. */
    Point     *pointx = PG_GETARG_POINT_P(0);
    Point     *pointy = PG_GETARG_POINT_P(1);
    Point     *new_point = (Point *) palloc(sizeof(Point));

    new_point->x = pointx->x;
    new_point->y = pointy->y;

    PG_RETURN_POINT_P(new_point);
}

/* by reference, variable length */

PG_FUNCTION_INFO_V1(copytext);

Datum
copytext(PG_FUNCTION_ARGS)
{
    text     *t = PG_GETARG_TEXT_PP(0);

    /*
     * VARSIZE_ANY_EXHDR is the size of the struct in bytes, minus the
     * VARHDRSZ or VARHDRSZ_SHORT of its header.  Construct the copy with a
     * full-length header.
     */
    text     *new_t = (text *) palloc(VARSIZE_ANY_EXHDR(t) + VARHDRSZ);
    SET_VARSIZE(new_t, VARSIZE_ANY_EXHDR(t) + VARHDRSZ);

    /*
     * VARDATA is a pointer to the data region of the new struct.  The source
     * could be a short datum, so retrieve its data through VARDATA_ANY.
     */
    memcpy(VARDATA(new_t),          /* destination */
           VARDATA_ANY(t),          /* source */
           VARSIZE_ANY_EXHDR(t));   /* how many bytes */
    PG_RETURN_TEXT_P(new_t);
}

PG_FUNCTION_INFO_V1(concat_text);

Datum
concat_text(PG_FUNCTION_ARGS)
{
    text  *arg1 = PG_GETARG_TEXT_PP(0);
    text  *arg2 = PG_GETARG_TEXT_PP(1);
    int32 arg1_size = VARSIZE_ANY_EXHDR(arg1);
    int32 arg2_size = VARSIZE_ANY_EXHDR(arg2);
    int32 new_text_size = arg1_size + arg2_size + VARHDRSZ;
    text *new_text = (text *) palloc(new_text_size);

    SET_VARSIZE(new_text, new_text_size);
    memcpy(VARDATA(new_text), VARDATA_ANY(arg1), arg1_size);
    memcpy(VARDATA(new_text) + arg1_size, VARDATA_ANY(arg2), arg2_size);
    PG_RETURN_TEXT_P(new_text);
}
]]>
</programlisting>

    <para>
     Supposing that the above code has been prepared in file
     <filename>funcs.c</filename> and compiled into a shared object,
     we could define the functions to <productname>PostgreSQL</productname>
     with commands like this:
    </para>

<programlisting>
CREATE FUNCTION add_one(integer) RETURNS integer
     AS '<replaceable>DIRECTORY</replaceable>/funcs', 'add_one'
     LANGUAGE C STRICT;

-- note overloading of SQL function name "add_one"
CREATE FUNCTION add_one(double precision) RETURNS double precision
     AS '<replaceable>DIRECTORY</replaceable>/funcs', 'add_one_float8'
     LANGUAGE C STRICT;

CREATE FUNCTION makepoint(point, point) RETURNS point
     AS '<replaceable>DIRECTORY</replaceable>/funcs', 'makepoint'
     LANGUAGE C STRICT;

CREATE FUNCTION copytext(text) RETURNS text
     AS '<replaceable>DIRECTORY</replaceable>/funcs', 'copytext'
     LANGUAGE C STRICT;

CREATE FUNCTION concat_text(text, text) RETURNS text
     AS '<replaceable>DIRECTORY</replaceable>/funcs', 'concat_text'
     LANGUAGE C STRICT;
</programlisting>

    <para>
     Here, <replaceable>DIRECTORY</replaceable> stands for the
     directory of the shared library file (for instance the
     <productname>PostgreSQL</productname> tutorial directory, which
     contains the code for the examples used in this section).
     (Better style would be to use just <literal>'funcs'</literal> in the
     <literal>AS</literal> clause, after having added
     <replaceable>DIRECTORY</replaceable> to the search path.  In any
     case, we can omit the system-specific extension for a shared
     library, commonly <literal>.so</literal>.)
    </para>

    <para>
     Notice that we have specified the functions as <quote>strict</quote>,
     meaning that
     the system should automatically assume a null result if any input
     value is null.  By doing this, we avoid having to check for null inputs
     in the function code.  Without this, we'd have to check for null values
     explicitly, using <function>PG_ARGISNULL()</function>.
    </para>

    <para>
     The macro <function>PG_ARGISNULL(<replaceable>n</replaceable>)</function>
     allows a function to test whether each input is null.  (Of course, doing
     this is only necessary in functions not declared <quote>strict</quote>.)
     As with the
     <function>PG_GETARG_<replaceable>xxx</replaceable>()</function> macros,
     the input arguments are counted beginning at zero.  Note that one
     should refrain from executing
     <function>PG_GETARG_<replaceable>xxx</replaceable>()</function> until
     one has verified that the argument isn't null.
     To return a null result, execute <function>PG_RETURN_NULL()</function>;
     this works in both strict and nonstrict functions.
    </para>

    <para>
     At first glance, the version-1 coding conventions might appear
     to be just pointless obscurantism, compared to using
     plain <literal>C</literal> calling conventions.  They do however allow
     us to deal with <literal>NULL</literal>able arguments/return values,
     and <quote>toasted</quote> (compressed or out-of-line) values.
    </para>

    <para>
     Other options provided by the version-1 interface are two
     variants of the
     <function>PG_GETARG_<replaceable>xxx</replaceable>()</function>
     macros. The first of these,
     <function>PG_GETARG_<replaceable>xxx</replaceable>_COPY()</function>,
     guarantees to return a copy of the specified argument that is
     safe for writing into. (The normal macros will sometimes return a
     pointer to a value that is physically stored in a table, which
     must not be written to. Using the
     <function>PG_GETARG_<replaceable>xxx</replaceable>_COPY()</function>
     macros guarantees a writable result.)
    The second variant consists of the
    <function>PG_GETARG_<replaceable>xxx</replaceable>_SLICE()</function>
    macros which take three arguments. The first is the number of the
    function argument (as above). The second and third are the offset and
    length of the segment to be returned. Offsets are counted from
    zero, and a negative length requests that the remainder of the
    value be returned. These macros provide more efficient access to
    parts of large values in the case where they have storage type
    <quote>external</quote>. (The storage type of a column can be specified using
    <literal>ALTER TABLE <replaceable>tablename</replaceable> ALTER
    COLUMN <replaceable>colname</replaceable> SET STORAGE
    <replaceable>storagetype</replaceable></literal>. <replaceable>storagetype</replaceable> is one of
    <literal>plain</literal>, <literal>external</literal>, <literal>extended</literal>,
     or <literal>main</literal>.)
    </para>

    <para>
     Finally, the version-1 function call conventions make it possible
     to return set results (<xref linkend="xfunc-c-return-set"/>) and
     implement trigger functions (<xref linkend="triggers"/>) and
     procedural-language call handlers (<xref
     linkend="plhandler"/>).  For more details
     see <filename>src/backend/utils/fmgr/README</filename> in the
     source distribution.
    </para>
   </sect2>

   <sect2 id="xfunc-c-code">
    <title>Writing Code</title>

    <para>
     Before we turn to the more advanced topics, we should discuss
     some coding rules for <productname>PostgreSQL</productname>
     C-language functions.  While it might be possible to load functions
     written in languages other than C into
     <productname>PostgreSQL</productname>, this is usually difficult
     (when it is possible at all) because other languages, such as
     C++, FORTRAN, or Pascal often do not follow the same calling
     convention as C.  That is, other languages do not pass argument
     and return values between functions in the same way.  For this
     reason, we will assume that your C-language functions are
     actually written in C.
    </para>

    <para>
     The basic rules for writing and building C functions are as follows:

     <itemizedlist>
      <listitem>
       <para>
        Use <literal>pg_config
        --includedir-server</literal><indexterm><primary>pg_config</primary><secondary>with user-defined C functions</secondary></indexterm>
        to find out where the <productname>PostgreSQL</productname> server header
        files are installed on your system (or the system that your
        users will be running on).
       </para>
      </listitem>

      <listitem>
       <para>
        Compiling and linking your code so that it can be dynamically
        loaded into <productname>PostgreSQL</productname> always
        requires special flags.  See <xref linkend="dfunc"/> for a
        detailed explanation of how to do it for your particular
        operating system.
       </para>
      </listitem>

      <listitem>
       <para>
        Remember to define a <quote>magic block</quote> for your shared library,
        as described in <xref linkend="xfunc-c-dynload"/>.
       </para>
      </listitem>

      <listitem>
       <para>
        When allocating memory, use the
        <productname>PostgreSQL</productname> functions
        <function>palloc</function><indexterm><primary>palloc</primary></indexterm> and <function>pfree</function><indexterm><primary>pfree</primary></indexterm>
        instead of the corresponding C library functions
        <function>malloc</function> and <function>free</function>.
        The memory allocated by <function>palloc</function> will be
        freed automatically at the end of each transaction, preventing
        memory leaks.
       </para>
      </listitem>

      <listitem>
       <para>
        Always zero the bytes of your structures using <function>memset</function>
        (or allocate them with <function>palloc0</function> in the first place).
        Even if you assign to each field of your structure, there might be
        alignment padding (holes in the structure) that contain
        garbage values.  Without this, it's difficult to
        support hash indexes or hash joins, as you must pick out only
        the significant bits of your data structure to compute a hash.
        The planner also sometimes relies on comparing constants via
        bitwise equality, so you can get undesirable planning results if
        logically-equivalent values aren't bitwise equal.
       </para>
      </listitem>

      <listitem>
       <para>
        Most of the internal <productname>PostgreSQL</productname>
        types are declared in <filename>postgres.h</filename>, while
        the function manager interfaces
        (<symbol>PG_FUNCTION_ARGS</symbol>, etc.)  are in
        <filename>fmgr.h</filename>, so you will need to include at
        least these two files.  For portability reasons it's best to
        include <filename>postgres.h</filename> <emphasis>first</emphasis>,
        before any other system or user header files.  Including
        <filename>postgres.h</filename> will also include
        <filename>elog.h</filename> and <filename>palloc.h</filename>
        for you.
       </para>
      </listitem>

      <listitem>
       <para>
        Symbol names defined within object files must not conflict
        with each other or with symbols defined in the
        <productname>PostgreSQL</productname> server executable.  You
        will have to rename your functions or variables if you get
        error messages to this effect.
       </para>
      </listitem>
     </itemizedlist>
    </para>
   </sect2>

&dfunc;

   <sect2 id="xfunc-api-abi-stability-guidance">
    <title>Server API and ABI Stability Guidance</title>

    <para>
     This section contains guidance to authors of extensions and other server
     plugins about API and ABI stability in the
     <productname>PostgreSQL</productname> server.
    </para>

    <sect3 id="xfunc-guidance-general">
     <title>General</title>

     <para>
      The <productname>PostgreSQL</productname> server contains several
      well-demarcated APIs for server plugins, such as the function manager
      (<acronym>fmgr</acronym>, described in this chapter),
      <acronym>SPI</acronym> (<xref linkend="spi"/>), and various hooks
      specifically designed for extensions. These interfaces are carefully
      managed for long-term stability and compatibility. However, the entire
      set of global functions and variables in the server effectively
      constitutes the publicly usable API, and most of it was not designed
      with extensibility and long-term stability in mind.
     </para>

     <para>
      Therefore, while taking advantage of these interfaces is valid, the
      further one strays from the well-trodden path, the likelier it will be
      that one might encounter API or ABI compatibility issues at some point.
      Extension authors are encouraged to provide feedback about their
      requirements, so that over time, as new use patterns arise, certain
      interfaces can be considered more stabilized or new, better-designed
      interfaces can be added.
     </para>
    </sect3>

    <sect3 id="xfunc-guidance-api-compatibility">
     <title>API Compatibility</title>
     <para>
      The <acronym>API</acronym>, or application programming interface, is the
      interface used at compile time.
     </para>

     <sect4 id="xfunc-guidance-api-major-versions">
      <title>Major Versions</title>
      <para>
       There is <emphasis>no</emphasis> promise of API compatibility between
       <productname>PostgreSQL</productname> major versions. Extension code
       therefore might require source code changes to work with multiple major
       versions. These can usually be managed with preprocessor conditions
       such as <literal>#if PG_VERSION_NUM &gt;= 160000</literal>.
       Sophisticated extensions that use interfaces beyond the well-demarcated
       ones usually require a few such changes for each major server version.
      </para>
     </sect4>

     <sect4 id="xfunc-guidance-api-mninor-versions">
      <title>Minor Versions</title>
      <para>
       <productname>PostgreSQL</productname> makes an effort to avoid server
       API breaks in minor releases. In general, extension code that compiles
       and works with a minor release should also compile and work with any
       other minor release of the same major version, past or future.
      </para>

      <para>
       When a change <emphasis>is</emphasis> required, it will be carefully
       managed, taking the requirements of extensions into account. Such
       changes will be communicated in the release notes (<xref
       linkend="release"/>).
      </para>
     </sect4>
    </sect3>

    <sect3 id="xfunc-guidance-abi-compatibility">
     <title>ABI Compatibility</title>
      <para>
       The <acronym>ABI</acronym>, or application binary interface, is the
       interface used at run time.
      </para>

     <sect4 id="xfunc-guidance-abi-major-versions">
      <title>Major Versions</title>
      <para>
       Servers of different major versions have intentionally incompatible
       ABIs. Extensions that use server APIs must therefore be re-compiled for
       each major release. The inclusion of <literal>PG_MODULE_MAGIC</literal>
       (see <xref linkend="xfunc-c-dynload"/>) ensures that code compiled for
       one major version will be rejected by other major versions.
      </para>
     </sect4>

     <sect4 id="xfunc-guidance-abi-mninor-versions">
      <title>Minor Versions</title>
      <para>
       <productname>PostgreSQL</productname> makes an effort to avoid server
       ABI breaks in minor releases. In general, an extension compiled against
       any minor release should work with any other minor release of the same
       major version, past or future.
      </para>

      <para>
       When a change <emphasis>is</emphasis> required,
       <productname>PostgreSQL</productname> will choose the least invasive
       change possible, for example by squeezing a new field into padding
       space or appending it to the end of a struct. These sorts of changes
       should not impact extensions unless they use very unusual code
       patterns.
      </para>

      <para>
       In rare cases, however, even such non-invasive changes may be
       impractical or impossible. In such an event, the change will be
       carefully managed, taking the requirements of extensions into account.
       Such changes will also be documented in the release notes (<xref
       linkend="release"/>).
      </para>

      <para>
       Note, however, that many parts of the server are not designed or
       maintained as publicly-consumable APIs (and that, in most cases, the
       actual boundary is also not well-defined). If urgent needs arise,
       changes in those parts will naturally be made with less consideration
       for extension code than changes in well-defined and widely used
       interfaces.
      </para>

      <para>
       Also, in the absence of automated detection of such changes, this is
       not a guarantee, but historically such breaking changes have been
       extremely rare.
      </para>

     </sect4>
    </sect3>
  </sect2>

   <sect2 id="xfunc-c-composite-type-args">
    <title>Composite-Type Arguments</title>

    <para>
     Composite types do not have a fixed layout like C structures.
     Instances of a composite type can contain null fields.  In
     addition, composite types that are part of an inheritance
     hierarchy can have different fields than other members of the
     same inheritance hierarchy.  Therefore,
     <productname>PostgreSQL</productname> provides a function
     interface for accessing fields of composite types from C.
    </para>

    <para>
     Suppose we want to write a function to answer the query:

<programlisting>
SELECT name, c_overpaid(emp, 1500) AS overpaid
    FROM emp
    WHERE name = 'Bill' OR name = 'Sam';
</programlisting>

     Using the version-1 calling conventions, we can define
     <function>c_overpaid</function> as:

<programlisting><![CDATA[
#include "postgres.h"
#include "executor/executor.h"  /* for GetAttributeByName() */

PG_MODULE_MAGIC;

PG_FUNCTION_INFO_V1(c_overpaid);

Datum
c_overpaid(PG_FUNCTION_ARGS)
{
    HeapTupleHeader  t = PG_GETARG_HEAPTUPLEHEADER(0);
    int32            limit = PG_GETARG_INT32(1);
    bool isnull;
    Datum salary;

    salary = GetAttributeByName(t, "salary", &isnull);
    if (isnull)
        PG_RETURN_BOOL(false);
    /* Alternatively, we might prefer to do PG_RETURN_NULL() for null salary. */

    PG_RETURN_BOOL(DatumGetInt32(salary) > limit);
}
]]>
</programlisting>
    </para>

    <para>
     <function>GetAttributeByName</function> is the
     <productname>PostgreSQL</productname> system function that
     returns attributes out of the specified row.  It has
     three arguments: the argument of type <type>HeapTupleHeader</type> passed
     into
     the  function, the name of the desired attribute, and a
     return parameter that tells whether  the  attribute
     is  null.   <function>GetAttributeByName</function> returns a <type>Datum</type>
     value that you can convert to the proper data type by using the
     appropriate <function>DatumGet<replaceable>XXX</replaceable>()</function>
     function.  Note that the return value is meaningless if the null flag is
     set; always check the null flag before trying to do anything with the
     result.
    </para>

    <para>
     There is also <function>GetAttributeByNum</function>, which selects
     the target attribute by column number instead of name.
    </para>

    <para>
     The following command declares the function
     <function>c_overpaid</function> in SQL:

<programlisting>
CREATE FUNCTION c_overpaid(emp, integer) RETURNS boolean
    AS '<replaceable>DIRECTORY</replaceable>/funcs', 'c_overpaid'
    LANGUAGE C STRICT;
</programlisting>

     Notice we have used <literal>STRICT</literal> so that we did not have to
     check whether the input arguments were NULL.
    </para>
   </sect2>

   <sect2 id="xfunc-c-returning-rows">
    <title>Returning Rows (Composite Types)</title>

    <para>
     To return a row or composite-type value from a C-language
     function, you can use a special API that provides macros and
     functions to hide most of the complexity of building composite
     data types.  To use this API, the source file must include:
<programlisting>
#include "funcapi.h"
</programlisting>
    </para>

    <para>
     There are two ways you can build a composite data value (henceforth
     a <quote>tuple</quote>): you can build it from an array of Datum values,
     or from an array of C strings that can be passed to the input
     conversion functions of the tuple's column data types.  In either
     case, you first need to obtain or construct a <structname>TupleDesc</structname>
     descriptor for the tuple structure.  When working with Datums, you
     pass the <structname>TupleDesc</structname> to <function>BlessTupleDesc</function>,
     and then call <function>heap_form_tuple</function> for each row.  When working
     with C strings, you pass the <structname>TupleDesc</structname> to
     <function>TupleDescGetAttInMetadata</function>, and then call
     <function>BuildTupleFromCStrings</function> for each row.  In the case of a
     function returning a set of tuples, the setup steps can all be done
     once during the first call of the function.
    </para>

    <para>
     Several helper functions are available for setting up the needed
     <structname>TupleDesc</structname>.  The recommended way to do this in most
     functions returning composite values is to call:
<programlisting>
TypeFuncClass get_call_result_type(FunctionCallInfo fcinfo,
                                   Oid *resultTypeId,
                                   TupleDesc *resultTupleDesc)
</programlisting>
     passing the same <literal>fcinfo</literal> struct passed to the calling function
     itself.  (This of course requires that you use the version-1
     calling conventions.)  <varname>resultTypeId</varname> can be specified
     as <literal>NULL</literal> or as the address of a local variable to receive the
     function's result type OID.  <varname>resultTupleDesc</varname> should be the
     address of a local <structname>TupleDesc</structname> variable.  Check that the
     result is <literal>TYPEFUNC_COMPOSITE</literal>; if so,
     <varname>resultTupleDesc</varname> has been filled with the needed
     <structname>TupleDesc</structname>.  (If it is not, you can report an error along
     the lines of <quote>function returning record called in context that
     cannot accept type record</quote>.)
    </para>

    <tip>
     <para>
      <function>get_call_result_type</function> can resolve the actual type of a
      polymorphic function result; so it is useful in functions that return
      scalar polymorphic results, not only functions that return composites.
      The <varname>resultTypeId</varname> output is primarily useful for functions
      returning polymorphic scalars.
     </para>
    </tip>

    <note>
     <para>
      <function>get_call_result_type</function> has a sibling
      <function>get_expr_result_type</function>, which can be used to resolve the
      expected output type for a function call represented by an expression
      tree.  This can be used when trying to determine the result type from
      outside the function itself.  There is also
      <function>get_func_result_type</function>, which can be used when only the
      function's OID is available.  However these functions are not able
      to deal with functions declared to return <structname>record</structname>, and
      <function>get_func_result_type</function> cannot resolve polymorphic types,
      so you should preferentially use <function>get_call_result_type</function>.
     </para>
    </note>

    <para>
     Older, now-deprecated functions for obtaining
     <structname>TupleDesc</structname>s are:
<programlisting>
TupleDesc RelationNameGetTupleDesc(const char *relname)
</programlisting>
     to get a <structname>TupleDesc</structname> for the row type of a named relation,
     and:
<programlisting>
TupleDesc TypeGetTupleDesc(Oid typeoid, List *colaliases)
</programlisting>
     to get a <structname>TupleDesc</structname> based on a type OID. This can
     be used to get a <structname>TupleDesc</structname> for a base or
     composite type.  It will not work for a function that returns
     <structname>record</structname>, however, and it cannot resolve polymorphic
     types.
    </para>

    <para>
     Once you have a <structname>TupleDesc</structname>, call:
<programlisting>
TupleDesc BlessTupleDesc(TupleDesc tupdesc)
</programlisting>
     if you plan to work with Datums, or:
<programlisting>
AttInMetadata *TupleDescGetAttInMetadata(TupleDesc tupdesc)
</programlisting>
     if you plan to work with C strings.  If you are writing a function
     returning set, you can save the results of these functions in the
     <structname>FuncCallContext</structname> structure &mdash; use the
     <structfield>tuple_desc</structfield> or <structfield>attinmeta</structfield> field
     respectively.
    </para>

    <para>
     When working with Datums, use:
<programlisting>
HeapTuple heap_form_tuple(TupleDesc tupdesc, Datum *values, bool *isnull)
</programlisting>
     to build a <structname>HeapTuple</structname> given user data in Datum form.
    </para>

    <para>
     When working with C strings, use:
<programlisting>
HeapTuple BuildTupleFromCStrings(AttInMetadata *attinmeta, char **values)
</programlisting>
     to build a <structname>HeapTuple</structname> given user data
     in C string form.  <parameter>values</parameter> is an array of C strings,
     one for each attribute of the return row. Each C string should be in
     the form expected by the input function of the attribute data
     type. In order to return a null value for one of the attributes,
     the corresponding pointer in the <parameter>values</parameter> array
     should be set to <symbol>NULL</symbol>.  This function will need to
     be called again for each row you return.
    </para>

    <para>
     Once you have built a tuple to return from your function, it
     must be converted into a <type>Datum</type>. Use:
<programlisting>
HeapTupleGetDatum(HeapTuple tuple)
</programlisting>
     to convert a <structname>HeapTuple</structname> into a valid Datum.  This
     <type>Datum</type> can be returned directly if you intend to return
     just a single row, or it can be used as the current return value
     in a set-returning function.
    </para>

    <para>
     An example appears in the next section.
    </para>

   </sect2>

   <sect2 id="xfunc-c-return-set">
    <title>Returning Sets</title>

    <para>
     C-language functions have two options for returning sets (multiple
     rows).  In one method, called <firstterm>ValuePerCall</firstterm>
     mode, a set-returning function is called repeatedly (passing the same
     arguments each time) and it returns one new row on each call, until
     it has no more rows to return and signals that by returning NULL.
     The set-returning function (<acronym>SRF</acronym>) must therefore
     save enough state across calls to remember what it was doing and
     return the correct next item on each call.
     In the other method, called <firstterm>Materialize</firstterm> mode,
     an SRF fills and returns a tuplestore object containing its
     entire result; then only one call occurs for the whole result, and
     no inter-call state is needed.
    </para>

    <para>
     When using ValuePerCall mode, it is important to remember that the
     query is not guaranteed to be run to completion; that is, due to
     options such as <literal>LIMIT</literal>, the executor might stop
     making calls to the set-returning function before all rows have been
     fetched.  This means it is not safe to perform cleanup activities in
     the last call, because that might not ever happen.  It's recommended
     to use Materialize mode for functions that need access to external
     resources, such as file descriptors.
    </para>

    <para>
     The remainder of this section documents a set of helper macros that
     are commonly used (though not required to be used) for SRFs using
     ValuePerCall mode.  Additional details about Materialize mode can be
     found in <filename>src/backend/utils/fmgr/README</filename>.  Also,
     the <filename>contrib</filename> modules in
     the <productname>PostgreSQL</productname> source distribution contain
     many examples of SRFs using both ValuePerCall and Materialize mode.
    </para>

    <para>
     To use the ValuePerCall support macros described here,
     include <filename>funcapi.h</filename>.  These macros work with a
     structure <structname>FuncCallContext</structname> that contains the
     state that needs to be saved across calls.  Within the calling
     SRF, <literal>fcinfo-&gt;flinfo-&gt;fn_extra</literal> is used to
     hold a pointer to <structname>FuncCallContext</structname> across
     calls.  The macros automatically fill that field on first use,
     and expect to find the same pointer there on subsequent uses.
<programlisting>
typedef struct FuncCallContext
{
    /*
     * Number of times we've been called before
     *
     * call_cntr is initialized to 0 for you by SRF_FIRSTCALL_INIT(), and
     * incremented for you every time SRF_RETURN_NEXT() is called.
     */
    uint64 call_cntr;

    /*
     * OPTIONAL maximum number of calls
     *
     * max_calls is here for convenience only and setting it is optional.
     * If not set, you must provide alternative means to know when the
     * function is done.
     */
    uint64 max_calls;

    /*
     * OPTIONAL pointer to miscellaneous user-provided context information
     *
     * user_fctx is for use as a pointer to your own data to retain
     * arbitrary context information between calls of your function.
     */
    void *user_fctx;

    /*
     * OPTIONAL pointer to struct containing attribute type input metadata
     *
     * attinmeta is for use when returning tuples (i.e., composite data types)
     * and is not used when returning base data types. It is only needed
     * if you intend to use BuildTupleFromCStrings() to create the return
     * tuple.
     */
    AttInMetadata *attinmeta;

    /*
     * memory context used for structures that must live for multiple calls
     *
     * multi_call_memory_ctx is set by SRF_FIRSTCALL_INIT() for you, and used
     * by SRF_RETURN_DONE() for cleanup. It is the most appropriate memory
     * context for any memory that is to be reused across multiple calls
     * of the SRF.
     */
    MemoryContext multi_call_memory_ctx;

    /*
     * OPTIONAL pointer to struct containing tuple description
     *
     * tuple_desc is for use when returning tuples (i.e., composite data types)
     * and is only needed if you are going to build the tuples with
     * heap_form_tuple() rather than with BuildTupleFromCStrings().  Note that
     * the TupleDesc pointer stored here should usually have been run through
     * BlessTupleDesc() first.
     */
    TupleDesc tuple_desc;

} FuncCallContext;
</programlisting>
    </para>

    <para>
     The macros to be used by an <acronym>SRF</acronym> using this
     infrastructure are:
<programlisting>
SRF_IS_FIRSTCALL()
</programlisting>
     Use this to determine if your function is being called for the first or a
     subsequent time. On the first call (only), call:
<programlisting>
SRF_FIRSTCALL_INIT()
</programlisting>
     to initialize the <structname>FuncCallContext</structname>. On every function call,
     including the first, call:
<programlisting>
SRF_PERCALL_SETUP()
</programlisting>
     to set up for using the <structname>FuncCallContext</structname>.
    </para>

    <para>
     If your function has data to return in the current call, use:
<programlisting>
SRF_RETURN_NEXT(funcctx, result)
</programlisting>
     to return it to the caller.  (<literal>result</literal> must be of type
     <type>Datum</type>, either a single value or a tuple prepared as
     described above.)  Finally, when your function is finished
     returning data, use:
<programlisting>
SRF_RETURN_DONE(funcctx)
</programlisting>
     to clean up and end the <acronym>SRF</acronym>.
    </para>

    <para>
     The memory context that is current when the <acronym>SRF</acronym> is called is
     a transient context that will be cleared between calls.  This means
     that you do not need to call <function>pfree</function> on everything
     you allocated using <function>palloc</function>; it will go away anyway.  However, if you want to allocate
     any data structures to live across calls, you need to put them somewhere
     else.  The memory context referenced by
     <structfield>multi_call_memory_ctx</structfield> is a suitable location for any
     data that needs to survive until the <acronym>SRF</acronym> is finished running.  In most
     cases, this means that you should switch into
     <structfield>multi_call_memory_ctx</structfield> while doing the
     first-call setup.
     Use <literal>funcctx-&gt;user_fctx</literal> to hold a pointer to
     any such cross-call data structures.
     (Data you allocate
     in <structfield>multi_call_memory_ctx</structfield> will go away
     automatically when the query ends, so it is not necessary to free
     that data manually, either.)
    </para>

    <warning>
     <para>
      While the actual arguments to the function remain unchanged between
      calls, if you detoast the argument values (which is normally done
      transparently by the
      <function>PG_GETARG_<replaceable>xxx</replaceable></function> macro)
      in the transient context then the detoasted copies will be freed on
      each cycle. Accordingly, if you keep references to such values in
      your <structfield>user_fctx</structfield>, you must either copy them into the
      <structfield>multi_call_memory_ctx</structfield> after detoasting, or ensure
      that you detoast the values only in that context.
     </para>
    </warning>

    <para>
     A complete pseudo-code example looks like the following:
<programlisting>
Datum
my_set_returning_function(PG_FUNCTION_ARGS)
{
    FuncCallContext  *funcctx;
    Datum             result;
    <replaceable>further declarations as needed</replaceable>

    if (SRF_IS_FIRSTCALL())
    {
        MemoryContext oldcontext;

        funcctx = SRF_FIRSTCALL_INIT();
        oldcontext = MemoryContextSwitchTo(funcctx-&gt;multi_call_memory_ctx);
        /* One-time setup code appears here: */
        <replaceable>user code</replaceable>
        <replaceable>if returning composite</replaceable>
            <replaceable>build TupleDesc, and perhaps AttInMetadata</replaceable>
        <replaceable>endif returning composite</replaceable>
        <replaceable>user code</replaceable>
        MemoryContextSwitchTo(oldcontext);
    }

    /* Each-time setup code appears here: */
    <replaceable>user code</replaceable>
    funcctx = SRF_PERCALL_SETUP();
    <replaceable>user code</replaceable>

    /* this is just one way we might test whether we are done: */
    if (funcctx-&gt;call_cntr &lt; funcctx-&gt;max_calls)
    {
        /* Here we want to return another item: */
        <replaceable>user code</replaceable>
        <replaceable>obtain result Datum</replaceable>
        SRF_RETURN_NEXT(funcctx, result);
    }
    else
    {
        /* Here we are done returning items, so just report that fact. */
        /* (Resist the temptation to put cleanup code here.) */
        SRF_RETURN_DONE(funcctx);
    }
}
</programlisting>
    </para>

    <para>
     A complete example of a simple <acronym>SRF</acronym> returning a composite type
     looks like:
<programlisting><![CDATA[
PG_FUNCTION_INFO_V1(retcomposite);

Datum
retcomposite(PG_FUNCTION_ARGS)
{
    FuncCallContext     *funcctx;
    int                  call_cntr;
    int                  max_calls;
    TupleDesc            tupdesc;
    AttInMetadata       *attinmeta;

    /* stuff done only on the first call of the function */
    if (SRF_IS_FIRSTCALL())
    {
        MemoryContext   oldcontext;

        /* create a function context for cross-call persistence */
        funcctx = SRF_FIRSTCALL_INIT();

        /* switch to memory context appropriate for multiple function calls */
        oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);

        /* total number of tuples to be returned */
        funcctx->max_calls = PG_GETARG_INT32(0);

        /* Build a tuple descriptor for our result type */
        if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
            ereport(ERROR,
                    (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
                     errmsg("function returning record called in context "
                            "that cannot accept type record")));

        /*
         * generate attribute metadata needed later to produce tuples from raw
         * C strings
         */
        attinmeta = TupleDescGetAttInMetadata(tupdesc);
        funcctx->attinmeta = attinmeta;

        MemoryContextSwitchTo(oldcontext);
    }

    /* stuff done on every call of the function */
    funcctx = SRF_PERCALL_SETUP();

    call_cntr = funcctx->call_cntr;
    max_calls = funcctx->max_calls;
    attinmeta = funcctx->attinmeta;

    if (call_cntr < max_calls)    /* do when there is more left to send */
    {
        char       **values;
        HeapTuple    tuple;
        Datum        result;

        /*
         * Prepare a values array for building the returned tuple.
         * This should be an array of C strings which will
         * be processed later by the type input functions.
         */
        values = (char **) palloc(3 * sizeof(char *));
        values[0] = (char *) palloc(16 * sizeof(char));
        values[1] = (char *) palloc(16 * sizeof(char));
        values[2] = (char *) palloc(16 * sizeof(char));

        snprintf(values[0], 16, "%d", 1 * PG_GETARG_INT32(1));
        snprintf(values[1], 16, "%d", 2 * PG_GETARG_INT32(1));
        snprintf(values[2], 16, "%d", 3 * PG_GETARG_INT32(1));

        /* build a tuple */
        tuple = BuildTupleFromCStrings(attinmeta, values);

        /* make the tuple into a datum */
        result = HeapTupleGetDatum(tuple);

        /* clean up (this is not really necessary) */
        pfree(values[0]);
        pfree(values[1]);
        pfree(values[2]);
        pfree(values);

        SRF_RETURN_NEXT(funcctx, result);
    }
    else    /* do when there is no more left */
    {
        SRF_RETURN_DONE(funcctx);
    }
}
]]>
</programlisting>

     One way to declare this function in SQL is:
<programlisting>
CREATE TYPE __retcomposite AS (f1 integer, f2 integer, f3 integer);

CREATE OR REPLACE FUNCTION retcomposite(integer, integer)
    RETURNS SETOF __retcomposite
    AS '<replaceable>filename</replaceable>', 'retcomposite'
    LANGUAGE C IMMUTABLE STRICT;
</programlisting>
     A different way is to use OUT parameters:
<programlisting>
CREATE OR REPLACE FUNCTION retcomposite(IN integer, IN integer,
    OUT f1 integer, OUT f2 integer, OUT f3 integer)
    RETURNS SETOF record
    AS '<replaceable>filename</replaceable>', 'retcomposite'
    LANGUAGE C IMMUTABLE STRICT;
</programlisting>
     Notice that in this method the output type of the function is formally
     an anonymous <structname>record</structname> type.
    </para>
   </sect2>

   <sect2 id="xfunc-c-polymorphic">
    <title>Polymorphic Arguments and Return Types</title>

    <para>
     C-language functions can be declared to accept and
     return the polymorphic types described in <xref
     linkend="extend-types-polymorphic"/>.
     When a function's arguments or return types
     are defined as polymorphic types, the function author cannot know
     in advance what data type it will be called with, or
     need to return. There are two routines provided in <filename>fmgr.h</filename>
     to allow a version-1 C function to discover the actual data types
     of its arguments and the type it is expected to return. The routines are
     called <literal>get_fn_expr_rettype(FmgrInfo *flinfo)</literal> and
     <literal>get_fn_expr_argtype(FmgrInfo *flinfo, int argnum)</literal>.
     They return the result or argument type OID, or <symbol>InvalidOid</symbol> if the
     information is not available.
     The structure <literal>flinfo</literal> is normally accessed as
     <literal>fcinfo-&gt;flinfo</literal>. The parameter <literal>argnum</literal>
     is zero based.  <function>get_call_result_type</function> can also be used
     as an alternative to <function>get_fn_expr_rettype</function>.
     There is also <function>get_fn_expr_variadic</function>, which can be used to
     find out whether variadic arguments have been merged into an array.
     This is primarily useful for <literal>VARIADIC "any"</literal> functions,
     since such merging will always have occurred for variadic functions
     taking ordinary array types.
    </para>

    <para>
     For example, suppose we want to write a function to accept a single
     element of any type, and return a one-dimensional array of that type:

<programlisting>
PG_FUNCTION_INFO_V1(make_array);
Datum
make_array(PG_FUNCTION_ARGS)
{
    ArrayType  *result;
    Oid         element_type = get_fn_expr_argtype(fcinfo-&gt;flinfo, 0);
    Datum       element;
    bool        isnull;
    int16       typlen;
    bool        typbyval;
    char        typalign;
    int         ndims;
    int         dims[MAXDIM];
    int         lbs[MAXDIM];

    if (!OidIsValid(element_type))
        elog(ERROR, "could not determine data type of input");

    /* get the provided element, being careful in case it's NULL */
    isnull = PG_ARGISNULL(0);
    if (isnull)
        element = (Datum) 0;
    else
        element = PG_GETARG_DATUM(0);

    /* we have one dimension */
    ndims = 1;
    /* and one element */
    dims[0] = 1;
    /* and lower bound is 1 */
    lbs[0] = 1;

    /* get required info about the element type */
    get_typlenbyvalalign(element_type, &amp;typlen, &amp;typbyval, &amp;typalign);

    /* now build the array */
    result = construct_md_array(&amp;element, &amp;isnull, ndims, dims, lbs,
                                element_type, typlen, typbyval, typalign);

    PG_RETURN_ARRAYTYPE_P(result);
}
</programlisting>
    </para>

    <para>
     The following command declares the function
     <function>make_array</function> in SQL:

<programlisting>
CREATE FUNCTION make_array(anyelement) RETURNS anyarray
    AS '<replaceable>DIRECTORY</replaceable>/funcs', 'make_array'
    LANGUAGE C IMMUTABLE;
</programlisting>
    </para>

    <para>
     There is a variant of polymorphism that is only available to C-language
     functions: they can be declared to take parameters of type
     <literal>"any"</literal>.  (Note that this type name must be double-quoted,
     since it's also an SQL reserved word.)  This works like
     <type>anyelement</type> except that it does not constrain different
     <literal>"any"</literal> arguments to be the same type, nor do they help
     determine the function's result type.  A C-language function can also
     declare its final parameter to be <literal>VARIADIC "any"</literal>.  This will
     match one or more actual arguments of any type (not necessarily the same
     type).  These arguments will <emphasis>not</emphasis> be gathered into an array
     as happens with normal variadic functions; they will just be passed to
     the function separately.  The <function>PG_NARGS()</function> macro and the
     methods described above must be used to determine the number of actual
     arguments and their types when using this feature.  Also, users of such
     a function might wish to use the <literal>VARIADIC</literal> keyword in their
     function call, with the expectation that the function would treat the
     array elements as separate arguments.  The function itself must implement
     that behavior if wanted, after using <function>get_fn_expr_variadic</function> to
     detect that the actual argument was marked with <literal>VARIADIC</literal>.
    </para>
   </sect2>

   <sect2 id="xfunc-shared-addin">
    <title>Shared Memory</title>

    <sect3 id="xfunc-shared-addin-at-startup">
     <title>Requesting Shared Memory at Startup</title>

     <para>
      Add-ins can reserve shared memory on server startup.  To do so, the
      add-in's shared library must be preloaded by specifying it in
      <xref linkend="guc-shared-preload-libraries"/><indexterm><primary>shared_preload_libraries</primary></indexterm>.
      The shared library should also register a
      <literal>shmem_request_hook</literal> in its
      <function>_PG_init</function> function.  This
      <literal>shmem_request_hook</literal> can reserve shared memory by
      calling:
<programlisting>
void RequestAddinShmemSpace(Size size)
</programlisting>
      Each backend should obtain a pointer to the reserved shared memory by
      calling:
<programlisting>
void *ShmemInitStruct(const char *name, Size size, bool *foundPtr)
</programlisting>
      If this function sets <literal>foundPtr</literal> to
      <literal>false</literal>, the caller should proceed to initialize the
      contents of the reserved shared memory.  If <literal>foundPtr</literal>
      is set to <literal>true</literal>, the shared memory was already
      initialized by another backend, and the caller need not initialize
      further.
     </para>

     <para>
      To avoid race conditions, each backend should use the LWLock
      <function>AddinShmemInitLock</function> when initializing its allocation
      of shared memory, as shown here:
<programlisting>
static mystruct *ptr = NULL;
bool        found;

LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
ptr = ShmemInitStruct("my struct name", size, &amp;found);
if (!found)
{
    ... initialize contents of shared memory ...
    ptr->locks = GetNamedLWLockTranche("my tranche name");
}
LWLockRelease(AddinShmemInitLock);
</programlisting>
      <literal>shmem_startup_hook</literal> provides a convenient place for the
      initialization code, but it is not strictly required that all such code
      be placed in this hook.  Each backend will execute the registered
      <literal>shmem_startup_hook</literal> shortly after it attaches to shared
      memory.  Note that add-ins should still acquire
      <function>AddinShmemInitLock</function> within this hook, as shown in the
      example above.
     </para>

     <para>
      An example of a <literal>shmem_request_hook</literal> and
      <literal>shmem_startup_hook</literal> can be found in
      <filename>contrib/pg_stat_statements/pg_stat_statements.c</filename> in
      the <productname>PostgreSQL</productname> source tree.
     </para>
    </sect3>

    <sect3 id="xfunc-shared-addin-after-startup">
     <title>Requesting Shared Memory After Startup</title>

     <para>
      There is another, more flexible method of reserving shared memory that
      can be done after server startup and outside a
      <literal>shmem_request_hook</literal>.  To do so, each backend that will
      use the shared memory should obtain a pointer to it by calling:
<programlisting>
void *GetNamedDSMSegment(const char *name, size_t size,
                         void (*init_callback) (void *ptr),
                         bool *found)
</programlisting>
      If a dynamic shared memory segment with the given name does not yet
      exist, this function will allocate it and initialize it with the provided
      <function>init_callback</function> callback function.  If the segment has
      already been allocated and initialized by another backend, this function
      simply attaches the existing dynamic shared memory segment to the current
      backend.
     </para>

     <para>
      Unlike shared memory reserved at server startup, there is no need to
      acquire <function>AddinShmemInitLock</function> or otherwise take action
      to avoid race conditions when reserving shared memory with
      <function>GetNamedDSMSegment</function>.  This function ensures that only
      one backend allocates and initializes the segment and that all other
      backends receive a pointer to the fully allocated and initialized
      segment.
     </para>

     <para>
      A complete usage example of <function>GetNamedDSMSegment</function> can
      be found in
      <filename>src/test/modules/test_dsm_registry/test_dsm_registry.c</filename>
      in the <productname>PostgreSQL</productname> source tree.
     </para>
    </sect3>
   </sect2>

   <sect2 id="xfunc-addin-lwlocks">
    <title>LWLocks</title>

    <sect3 id="xfunc-addin-lwlocks-at-startup">
     <title>Requesting LWLocks at Startup</title>

     <para>
      Add-ins can reserve LWLocks on server startup.  As with shared memory
      reserved at server startup, the add-in's shared library must be preloaded
      by specifying it in
      <xref linkend="guc-shared-preload-libraries"/><indexterm><primary>shared_preload_libraries</primary></indexterm>,
      and the shared library should register a
      <literal>shmem_request_hook</literal> in its
      <function>_PG_init</function> function.  This
      <literal>shmem_request_hook</literal> can reserve LWLocks by calling:
<programlisting>
void RequestNamedLWLockTranche(const char *tranche_name, int num_lwlocks)
</programlisting>
      This ensures that an array of <literal>num_lwlocks</literal> LWLocks is
      available under the name <literal>tranche_name</literal>.  A pointer to
      this array can be obtained by calling:
<programlisting>
LWLockPadded *GetNamedLWLockTranche(const char *tranche_name)
</programlisting>
     </para>
    </sect3>

    <sect3 id="xfunc-addin-lwlocks-after-startup">
     <title>Requesting LWLocks After Startup</title>

     <para>
      There is another, more flexible method of obtaining LWLocks that can be
      done after server startup and outside a
      <literal>shmem_request_hook</literal>.  To do so, first allocate a
      <literal>tranche_id</literal> by calling:
<programlisting>
int LWLockNewTrancheId(void)
</programlisting>
      Next, initialize each LWLock, passing the new
      <literal>tranche_id</literal> as an argument:
<programlisting>
void LWLockInitialize(LWLock *lock, int tranche_id)
</programlisting>
      Similar to shared memory, each backend should ensure that only one
      process allocates a new <literal>tranche_id</literal> and initializes
      each new LWLock.  One way to do this is to only call these functions in
      your shared memory initialization code with the
      <function>AddinShmemInitLock</function> held exclusively.  If using
      <function>GetNamedDSMSegment</function>, calling these functions in the
      <function>init_callback</function> callback function is sufficient to
      avoid race conditions.
     </para>

     <para>
      Finally, each backend using the <literal>tranche_id</literal> should
      associate it with a <literal>tranche_name</literal> by calling:
<programlisting>
void LWLockRegisterTranche(int tranche_id, const char *tranche_name)
</programlisting>
     </para>

     <para>
      A complete usage example of <function>LWLockNewTrancheId</function>,
      <function>LWLockInitialize</function>, and
      <function>LWLockRegisterTranche</function> can be found in
      <filename>contrib/pg_prewarm/autoprewarm.c</filename> in the
      <productname>PostgreSQL</productname> source tree.
     </para>
    </sect3>
   </sect2>

   <sect2 id="xfunc-addin-wait-events">
    <title>Custom Wait Events</title>

    <para>
     Add-ins can define custom wait events under the wait event type
     <literal>Extension</literal> by calling:
<programlisting>
uint32 WaitEventExtensionNew(const char *wait_event_name)
</programlisting>
     The wait event is associated to a user-facing custom string.
     An example can be found in <filename>src/test/modules/worker_spi</filename>
     in the PostgreSQL source tree.
    </para>
    <para>
     Custom wait events can be viewed in
     <link linkend="monitoring-pg-stat-activity-view"><structname>pg_stat_activity</structname></link>:
<screen>
=# SELECT wait_event_type, wait_event FROM pg_stat_activity
     WHERE backend_type ~ 'worker_spi';
 wait_event_type |  wait_event
-----------------+---------------
 Extension       | WorkerSpiMain
(1 row)
</screen>
    </para>
   </sect2>

   <sect2 id="xfunc-addin-injection-points">
    <title>Injection Points</title>

    <para>
     An injection point with a given <literal>name</literal> is declared using
     macro:
<programlisting>
INJECTION_POINT(name);
</programlisting>

     There are a few injection points already declared at strategic points
     within the server code. After adding a new injection point the code needs
     to be compiled in order for that injection point to be available in the
     binary. Add-ins written in C-language can declare injection points in
     their own code using the same macro.
    </para>

    <para>
     Executing an injection point can require allocating a small amount of
     memory, which can fail. If you need to have an injection point in a
     critical section where dynamic allocations are not allowed, you can use
     a two-step approach with the following macros:
<programlisting>
INJECTION_POINT_LOAD(name);
INJECTION_POINT_CACHED(name);
</programlisting>

     Before entering the critical section,
     call <function>INJECTION_POINT_LOAD</function>. It checks the shared
     memory state, and loads the callback into backend-private memory if it is
     active. Inside the critical section, use
     <function>INJECTION_POINT_CACHED</function> to execute the callback.
    </para>

    <para>
     Add-ins can attach callbacks to an already-declared injection point by
     calling:
<programlisting>
extern void InjectionPointAttach(const char *name,
                                 const char *library,
                                 const char *function,
                                 const void *private_data,
                                 int private_data_size);
</programlisting>

     <literal>name</literal> is the name of the injection point, which when
     reached during execution will execute the <literal>function</literal>
     loaded from <literal>library</literal>. <literal>private_data</literal>
     is a private area of data of size <literal>private_data_size</literal>
     given as argument to the callback when executed.
    </para>

    <para>
     Here is an example of callback for
     <literal>InjectionPointCallback</literal>:
<programlisting>
static void
custom_injection_callback(const char *name, const void *private_data)
{
    uint32 wait_event_info = WaitEventInjectionPointNew(name);

    pgstat_report_wait_start(wait_event_info);
    elog(NOTICE, "%s: executed custom callback", name);
    pgstat_report_wait_end();
}
</programlisting>
     This callback prints a message to server error log with severity
     <literal>NOTICE</literal>, but callbacks may implement more complex
     logic.
    </para>

    <para>
     An alternative way to define the action to take when an injection point
     is reached is to add the testing code alongside the normal source
     code. This can be useful if the action e.g. depends on local variables
     that are not accessible to loaded modules. The
     <function>IS_INJECTION_POINT_ATTACHED</function> macro can then be used
     to check if an injection point is attached, for example:
<programlisting>
#ifdef USE_INJECTION_POINTS
if (IS_INJECTION_POINT_ATTACHED("before-foobar"))
{
    /* change a local variable if injection point is attached */
    local_var = 123;

    /* also execute the callback */
    INJECTION_POINT_CACHED("before-foobar");
}
#endif
</programlisting>
     Note that the callback attached to the injection point will not be
     executed by the <function>IS_INJECTION_POINT_ATTACHED</function>
     macro. If you want to execute the callback, you must also call
     <function>INJECTION_POINT_CACHED</function> like in the above example.
    </para>

    <para>
     Optionally, it is possible to detach an injection point by calling:
<programlisting>
extern bool InjectionPointDetach(const char *name);
</programlisting>
     On success, <literal>true</literal> is returned, <literal>false</literal>
     otherwise.
    </para>

    <para>
     A callback attached to an injection point is available across all the
     backends including the backends started after
     <literal>InjectionPointAttach</literal> is called. It remains attached
     while the server is running or until the injection point is detached
     using <literal>InjectionPointDetach</literal>.
    </para>

    <para>
     An example can be found in
     <filename>src/test/modules/injection_points</filename> in the PostgreSQL
     source tree.
    </para>

    <para>
     Enabling injections points requires
     <option>--enable-injection-points</option> with
     <command>configure</command> or <option>-Dinjection_points=true</option>
     with <application>Meson</application>.
    </para>
   </sect2>

   <sect2 id="xfunc-addin-custom-cumulative-statistics">
    <title>Custom Cumulative Statistics</title>

    <para>
     It is possible for add-ins written in C-language to use custom types
     of cumulative statistics registered in the
     <link linkend="monitoring-stats-setup">Cumulative Statistics System</link>.
    </para>

    <para>
     First, define a <literal>PgStat_KindInfo</literal> that includes all
     the information related to the custom type registered. For example:
<programlisting>
static const PgStat_KindInfo custom_stats = {
    .name = "custom_stats",
    .fixed_amount = false,
    .shared_size = sizeof(PgStatShared_Custom),
    .shared_data_off = offsetof(PgStatShared_Custom, stats),
    .shared_data_len = sizeof(((PgStatShared_Custom *) 0)->stats),
    .pending_size = sizeof(PgStat_StatCustomEntry),
}
</programlisting>

     Then, each backend that needs to use this custom type needs to register
     it with <literal>pgstat_register_kind</literal> and a unique ID used to
     store the entries related to this type of statistics:
<programlisting>
extern PgStat_Kind pgstat_register_kind(PgStat_Kind kind,
                                        const PgStat_KindInfo *kind_info);
</programlisting>
     While developing a new extension, use
     <literal>PGSTAT_KIND_EXPERIMENTAL</literal> for
     <parameter>kind</parameter>. When you are ready to release the extension
     to users, reserve a kind ID at the
     <ulink url="https://wiki.postgresql.org/wiki/CustomCumulativeStats">
     Custom Cumulative Statistics</ulink> page.
    </para>

    <para>
     The details of the API for <literal>PgStat_KindInfo</literal> can
     be found in <filename>src/include/utils/pgstat_internal.h</filename>.
    </para>

    <para>
     The type of statistics registered is associated with a name and a unique
     ID shared across the server in shared memory. Each backend using a
     custom type of statistics maintains a local cache storing the information
     of each custom <literal>PgStat_KindInfo</literal>.
    </para>

    <para>
     Place the extension module implementing the custom cumulative statistics
     type in <xref linkend="guc-shared-preload-libraries"/> so that it will
     be loaded early during <productname>PostgreSQL</productname> startup.
    </para>

    <para>
     An example describing how to register and use custom statistics can be
     found in <filename>src/test/modules/injection_points</filename>.
    </para>
   </sect2>

   <sect2 id="extend-cpp">
    <title>Using C++ for Extensibility</title>

    <indexterm zone="extend-cpp">
     <primary>C++</primary>
    </indexterm>

    <para>
     Although the <productname>PostgreSQL</productname> backend is written in
     C, it is possible to write extensions in C++ if these guidelines are
     followed:

     <itemizedlist>
      <listitem>
       <para>
         All functions accessed by the backend must present a C interface
         to the backend;  these C functions can then call C++ functions.
         For example, <literal>extern C</literal> linkage is required for
         backend-accessed functions.  This is also necessary for any
         functions that are passed as pointers between the backend and
         C++ code.
       </para>
      </listitem>
      <listitem>
       <para>
        Free memory using the appropriate deallocation method.  For example,
        most backend memory is allocated using <function>palloc()</function>, so use
        <function>pfree()</function> to free it.  Using C++
        <function>delete</function> in such cases will fail.
       </para>
      </listitem>
      <listitem>
       <para>
        Prevent exceptions from propagating into the C code (use a catch-all
        block at the top level of all <literal>extern C</literal> functions).  This
        is necessary even if the C++ code does not explicitly throw any
        exceptions, because events like out-of-memory can still throw
        exceptions.  Any exceptions must be caught and appropriate errors
        passed back to the C interface.  If possible, compile C++ with
        <option>-fno-exceptions</option> to eliminate exceptions entirely; in such
        cases, you must check for failures in your C++ code, e.g.,  check for
        NULL returned by <function>new()</function>.
       </para>
      </listitem>
      <listitem>
       <para>
        If calling backend functions from C++ code, be sure that the
        C++ call stack contains only plain old data structures
        (<acronym>POD</acronym>).  This is necessary because backend errors
        generate a distant <function>longjmp()</function> that does not properly
        unroll a C++ call stack with non-POD objects.
       </para>
      </listitem>
     </itemizedlist>
    </para>

    <para>
     In summary, it is best to place C++ code behind a wall of
     <literal>extern C</literal> functions that interface to the backend,
     and avoid exception, memory, and call stack leakage.
    </para>
   </sect2>

  </sect1>

  <sect1 id="xfunc-optimization">
   <title>Function Optimization Information</title>

  <indexterm zone="xfunc-optimization">
   <primary>optimization information</primary>
   <secondary>for functions</secondary>
  </indexterm>

   <para>
    By default, a function is just a <quote>black box</quote> that the
    database system knows very little about the behavior of.  However,
    that means that queries using the function may be executed much less
    efficiently than they could be.  It is possible to supply additional
    knowledge that helps the planner optimize function calls.
   </para>

   <para>
    Some basic facts can be supplied by declarative annotations provided in
    the <link linkend="sql-createfunction"><command>CREATE FUNCTION</command></link> command.  Most important of
    these is the function's <link linkend="xfunc-volatility">volatility
    category</link> (<literal>IMMUTABLE</literal>, <literal>STABLE</literal>,
    or <literal>VOLATILE</literal>); one should always be careful to
    specify this correctly when defining a function.
    The parallel safety property (<literal>PARALLEL
    UNSAFE</literal>, <literal>PARALLEL RESTRICTED</literal>, or
    <literal>PARALLEL SAFE</literal>) must also be specified if you hope
    to use the function in parallelized queries.
    It can also be useful to specify the function's estimated execution
    cost, and/or the number of rows a set-returning function is estimated
    to return.  However, the declarative way of specifying those two
    facts only allows specifying a constant value, which is often
    inadequate.
   </para>

   <para>
    It is also possible to attach a <firstterm>planner support
    function</firstterm> to an SQL-callable function (called
    its <firstterm>target function</firstterm>), and thereby provide
    knowledge about the target function that is too complex to be
    represented declaratively.  Planner support functions have to be
    written in C (although their target functions might not be), so this is
    an advanced feature that relatively few people will use.
   </para>

   <para>
    A planner support function must have the SQL signature
<programlisting>
supportfn(internal) returns internal
</programlisting>
    It is attached to its target function by specifying
    the <literal>SUPPORT</literal> clause when creating the target function.
   </para>

   <para>
    The details of the API for planner support functions can be found in
    file <filename>src/include/nodes/supportnodes.h</filename> in the
    <productname>PostgreSQL</productname> source code.  Here we provide
    just an overview of what planner support functions can do.
    The set of possible requests to a support function is extensible,
    so more things might be possible in future versions.
   </para>

   <para>
    Some function calls can be simplified during planning based on
    properties specific to the function.  For example,
    <literal>int4mul(n, 1)</literal> could be simplified to
    just <literal>n</literal>.  This type of transformation can be
    performed by a planner support function, by having it implement
    the <literal>SupportRequestSimplify</literal> request type.
    The support function will be called for each instance of its target
    function found in a query parse tree.  If it finds that the particular
    call can be simplified into some other form, it can build and return a
    parse tree representing that expression.  This will automatically work
    for operators based on the function, too &mdash; in the example just
    given, <literal>n * 1</literal> would also be simplified to
    <literal>n</literal>.
    (But note that this is just an example; this particular
    optimization is not actually performed by
    standard <productname>PostgreSQL</productname>.)
    We make no guarantee that <productname>PostgreSQL</productname> will
    never call the target function in cases that the support function could
    simplify.  Ensure rigorous equivalence between the simplified
    expression and an actual execution of the target function.
   </para>

   <para>
    For target functions that return <type>boolean</type>, it is often useful to estimate
    the fraction of rows that will be selected by a <literal>WHERE</literal> clause using that
    function.  This can be done by a support function that implements
    the <literal>SupportRequestSelectivity</literal> request type.
   </para>

   <para>
    If the target function's run time is highly dependent on its inputs,
    it may be useful to provide a non-constant cost estimate for it.
    This can be done by a support function that implements
    the <literal>SupportRequestCost</literal> request type.
   </para>

   <para>
    For target functions that return sets, it is often useful to provide
    a non-constant estimate for the number of rows that will be returned.
    This can be done by a support function that implements
    the <literal>SupportRequestRows</literal> request type.
   </para>

   <para>
    For target functions that return <type>boolean</type>, it may be possible to
    convert a function call appearing in <literal>WHERE</literal> into an indexable operator
    clause or clauses.  The converted clauses might be exactly equivalent
    to the function's condition, or they could be somewhat weaker (that is,
    they might accept some values that the function condition does not).
    In the latter case the index condition is said to
    be <firstterm>lossy</firstterm>; it can still be used to scan an index,
    but the function call will have to be executed for each row returned by
    the index to see if it really passes the <literal>WHERE</literal> condition or not.
    To create such conditions, the support function must implement
    the <literal>SupportRequestIndexCondition</literal> request type.
   </para>
  </sect1>