<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
<HTML
><HEAD
><TITLE
>PL/pgSQL Under the Hood</TITLE
><META
NAME="GENERATOR"
CONTENT="Modular DocBook HTML Stylesheet Version 1.79"><LINK
REV="MADE"
HREF="mailto:pgsql-docs@postgresql.org"><LINK
REL="HOME"
TITLE="PostgreSQL 9.3.9 Documentation"
HREF="index.html"><LINK
REL="UP"
TITLE="PL/pgSQL - SQL Procedural Language"
HREF="plpgsql.html"><LINK
REL="PREVIOUS"
TITLE="Trigger Procedures"
HREF="plpgsql-trigger.html"><LINK
REL="NEXT"
TITLE="Tips for Developing in PL/pgSQL"
HREF="plpgsql-development-tips.html"><LINK
REL="STYLESHEET"
TYPE="text/css"
HREF="stylesheet.css"><META
HTTP-EQUIV="Content-Type"
CONTENT="text/html; charset=ISO-8859-1"><META
NAME="creation"
CONTENT="2015-06-13T20:07:22"></HEAD
><BODY
CLASS="SECT1"
><DIV
CLASS="NAVHEADER"
><TABLE
SUMMARY="Header navigation table"
WIDTH="100%"
BORDER="0"
CELLPADDING="0"
CELLSPACING="0"
><TR
><TH
COLSPAN="5"
ALIGN="center"
VALIGN="bottom"
><A
HREF="index.html"
>PostgreSQL 9.3.9 Documentation</A
></TH
></TR
><TR
><TD
WIDTH="10%"
ALIGN="left"
VALIGN="top"
><A
TITLE="Trigger Procedures"
HREF="plpgsql-trigger.html"
ACCESSKEY="P"
>Prev</A
></TD
><TD
WIDTH="10%"
ALIGN="left"
VALIGN="top"
><A
HREF="plpgsql.html"
ACCESSKEY="U"
>Up</A
></TD
><TD
WIDTH="60%"
ALIGN="center"
VALIGN="bottom"
>Chapter 40. <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> - <ACRONYM
CLASS="ACRONYM"
>SQL</ACRONYM
> Procedural Language</TD
><TD
WIDTH="20%"
ALIGN="right"
VALIGN="top"
><A
TITLE="Tips for Developing in PL/pgSQL"
HREF="plpgsql-development-tips.html"
ACCESSKEY="N"
>Next</A
></TD
></TR
></TABLE
><HR
ALIGN="LEFT"
WIDTH="100%"></DIV
><DIV
CLASS="SECT1"
><H1
CLASS="SECT1"
><A
NAME="PLPGSQL-IMPLEMENTATION"
>40.10. <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> Under the Hood</A
></H1
><P
> This section discusses some implementation details that are
frequently important for <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> users to know.
</P
><DIV
CLASS="SECT2"
><H2
CLASS="SECT2"
><A
NAME="PLPGSQL-VAR-SUBST"
>40.10.1. Variable Substitution</A
></H2
><P
> SQL statements and expressions within a <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> function
can refer to variables and parameters of the function. Behind the scenes,
<SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> substitutes query parameters for such references.
Parameters will only be substituted in places where a parameter or
column reference is syntactically allowed. As an extreme case, consider
this example of poor programming style:
</P><PRE
CLASS="PROGRAMLISTING"
>INSERT INTO foo (foo) VALUES (foo);</PRE
><P>
The first occurrence of <TT
CLASS="LITERAL"
>foo</TT
> must syntactically be a table
name, so it will not be substituted, even if the function has a variable
named <TT
CLASS="LITERAL"
>foo</TT
>. The second occurrence must be the name of a
column of the table, so it will not be substituted either. Only the
third occurrence is a candidate to be a reference to the function's
variable.
</P
><DIV
CLASS="NOTE"
><BLOCKQUOTE
CLASS="NOTE"
><P
><B
>Note: </B
> <SPAN
CLASS="PRODUCTNAME"
>PostgreSQL</SPAN
> versions before 9.0 would try
to substitute the variable in all three cases, leading to syntax errors.
</P
></BLOCKQUOTE
></DIV
><P
> Since the names of variables are syntactically no different from the names
of table columns, there can be ambiguity in statements that also refer to
tables: is a given name meant to refer to a table column, or a variable?
Let's change the previous example to
</P><PRE
CLASS="PROGRAMLISTING"
>INSERT INTO dest (col) SELECT foo + bar FROM src;</PRE
><P>
Here, <TT
CLASS="LITERAL"
>dest</TT
> and <TT
CLASS="LITERAL"
>src</TT
> must be table names, and
<TT
CLASS="LITERAL"
>col</TT
> must be a column of <TT
CLASS="LITERAL"
>dest</TT
>, but <TT
CLASS="LITERAL"
>foo</TT
>
and <TT
CLASS="LITERAL"
>bar</TT
> might reasonably be either variables of the function
or columns of <TT
CLASS="LITERAL"
>src</TT
>.
</P
><P
> By default, <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> will report an error if a name
in a SQL statement could refer to either a variable or a table column.
You can fix such a problem by renaming the variable or column,
or by qualifying the ambiguous reference, or by telling
<SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> which interpretation to prefer.
</P
><P
> The simplest solution is to rename the variable or column.
A common coding rule is to use a
different naming convention for <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
>
variables than you use for column names. For example,
if you consistently name function variables
<TT
CLASS="LITERAL"
>v_<TT
CLASS="REPLACEABLE"
><I
>something</I
></TT
></TT
> while none of your
column names start with <TT
CLASS="LITERAL"
>v_</TT
>, no conflicts will occur.
</P
><P
> Alternatively you can qualify ambiguous references to make them clear.
In the above example, <TT
CLASS="LITERAL"
>src.foo</TT
> would be an unambiguous reference
to the table column. To create an unambiguous reference to a variable,
declare it in a labeled block and use the block's label
(see <A
HREF="plpgsql-structure.html"
>Section 40.2</A
>). For example,
</P><PRE
CLASS="PROGRAMLISTING"
><<block>>
DECLARE
foo int;
BEGIN
foo := ...;
INSERT INTO dest (col) SELECT block.foo + bar FROM src;</PRE
><P>
Here <TT
CLASS="LITERAL"
>block.foo</TT
> means the variable even if there is a column
<TT
CLASS="LITERAL"
>foo</TT
> in <TT
CLASS="LITERAL"
>src</TT
>. Function parameters, as well as
special variables such as <TT
CLASS="LITERAL"
>FOUND</TT
>, can be qualified by the
function's name, because they are implicitly declared in an outer block
labeled with the function's name.
</P
><P
> Sometimes it is impractical to fix all the ambiguous references in a
large body of <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> code. In such cases you can
specify that <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> should resolve ambiguous references
as the variable (which is compatible with <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
>'s
behavior before <SPAN
CLASS="PRODUCTNAME"
>PostgreSQL</SPAN
> 9.0), or as the
table column (which is compatible with some other systems such as
<SPAN
CLASS="PRODUCTNAME"
>Oracle</SPAN
>).
</P
><P
> To change this behavior on a system-wide basis, set the configuration
parameter <TT
CLASS="LITERAL"
>plpgsql.variable_conflict</TT
> to one of
<TT
CLASS="LITERAL"
>error</TT
>, <TT
CLASS="LITERAL"
>use_variable</TT
>, or
<TT
CLASS="LITERAL"
>use_column</TT
> (where <TT
CLASS="LITERAL"
>error</TT
> is the factory default).
This parameter affects subsequent compilations
of statements in <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> functions, but not statements
already compiled in the current session.
Because changing this setting
can cause unexpected changes in the behavior of <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
>
functions, it can only be changed by a superuser.
</P
><P
> You can also set the behavior on a function-by-function basis, by
inserting one of these special commands at the start of the function
text:
</P><PRE
CLASS="PROGRAMLISTING"
>#variable_conflict error
#variable_conflict use_variable
#variable_conflict use_column</PRE
><P>
These commands affect only the function they are written in, and override
the setting of <TT
CLASS="LITERAL"
>plpgsql.variable_conflict</TT
>. An example is
</P><PRE
CLASS="PROGRAMLISTING"
>CREATE FUNCTION stamp_user(id int, comment text) RETURNS void AS $$
#variable_conflict use_variable
DECLARE
curtime timestamp := now();
BEGIN
UPDATE users SET last_modified = curtime, comment = comment
WHERE users.id = id;
END;
$$ LANGUAGE plpgsql;</PRE
><P>
In the <TT
CLASS="LITERAL"
>UPDATE</TT
> command, <TT
CLASS="LITERAL"
>curtime</TT
>, <TT
CLASS="LITERAL"
>comment</TT
>,
and <TT
CLASS="LITERAL"
>id</TT
> will refer to the function's variable and parameters
whether or not <TT
CLASS="LITERAL"
>users</TT
> has columns of those names. Notice
that we had to qualify the reference to <TT
CLASS="LITERAL"
>users.id</TT
> in the
<TT
CLASS="LITERAL"
>WHERE</TT
> clause to make it refer to the table column.
But we did not have to qualify the reference to <TT
CLASS="LITERAL"
>comment</TT
>
as a target in the <TT
CLASS="LITERAL"
>UPDATE</TT
> list, because syntactically
that must be a column of <TT
CLASS="LITERAL"
>users</TT
>. We could write the same
function without depending on the <TT
CLASS="LITERAL"
>variable_conflict</TT
> setting
in this way:
</P><PRE
CLASS="PROGRAMLISTING"
>CREATE FUNCTION stamp_user(id int, comment text) RETURNS void AS $$
<<fn>>
DECLARE
curtime timestamp := now();
BEGIN
UPDATE users SET last_modified = fn.curtime, comment = stamp_user.comment
WHERE users.id = stamp_user.id;
END;
$$ LANGUAGE plpgsql;</PRE
><P>
</P
><P
> Variable substitution does not happen in the command string given
to <TT
CLASS="COMMAND"
>EXECUTE</TT
> or one of its variants. If you need to
insert a varying value into such a command, do so as part of
constructing the string value, or use <TT
CLASS="LITERAL"
>USING</TT
>, as illustrated in
<A
HREF="plpgsql-statements.html#PLPGSQL-STATEMENTS-EXECUTING-DYN"
>Section 40.5.4</A
>.
</P
><P
> Variable substitution currently works only in <TT
CLASS="COMMAND"
>SELECT</TT
>,
<TT
CLASS="COMMAND"
>INSERT</TT
>, <TT
CLASS="COMMAND"
>UPDATE</TT
>, and <TT
CLASS="COMMAND"
>DELETE</TT
> commands,
because the main SQL engine allows query parameters only in these
commands. To use a non-constant name or value in other statement
types (generically called utility statements), you must construct
the utility statement as a string and <TT
CLASS="COMMAND"
>EXECUTE</TT
> it.
</P
></DIV
><DIV
CLASS="SECT2"
><H2
CLASS="SECT2"
><A
NAME="PLPGSQL-PLAN-CACHING"
>40.10.2. Plan Caching</A
></H2
><P
> The <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> interpreter parses the function's source
text and produces an internal binary instruction tree the first time the
function is called (within each session). The instruction tree
fully translates the
<SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> statement structure, but individual
<ACRONYM
CLASS="ACRONYM"
>SQL</ACRONYM
> expressions and <ACRONYM
CLASS="ACRONYM"
>SQL</ACRONYM
> commands
used in the function are not translated immediately.
</P
><P
>
As each expression and <ACRONYM
CLASS="ACRONYM"
>SQL</ACRONYM
> command is first
executed in the function, the <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> interpreter
parses and analyzes the command to create a prepared statement,
using the <ACRONYM
CLASS="ACRONYM"
>SPI</ACRONYM
> manager's
<CODE
CLASS="FUNCTION"
>SPI_prepare</CODE
> function.
Subsequent visits to that expression or command
reuse the prepared statement. Thus, a function with conditional code
paths that are seldom visited will never incur the overhead of
analyzing those commands that are never executed within the current
session. A disadvantage is that errors
in a specific expression or command cannot be detected until that
part of the function is reached in execution. (Trivial syntax
errors will be detected during the initial parsing pass, but
anything deeper will not be detected until execution.)
</P
><P
> <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> (or more precisely, the SPI manager) can
furthermore attempt to cache the execution plan associated with any
particular prepared statement. If a cached plan is not used, then
a fresh execution plan is generated on each visit to the statement,
and the current parameter values (that is, <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
>
variable values) can be used to optimize the selected plan. If the
statement has no parameters, or is executed many times, the SPI manager
will consider creating a <I
CLASS="FIRSTTERM"
>generic</I
> plan that is not dependent
on specific parameter values, and caching that for re-use. Typically
this will happen only if the execution plan is not very sensitive to
the values of the <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> variables referenced in it.
If it is, generating a plan each time is a net win. See <A
HREF="sql-prepare.html"
>PREPARE</A
> for more information about the behavior of
prepared statements.
</P
><P
> Because <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> saves prepared statements
and sometimes execution plans in this way,
SQL commands that appear directly in a
<SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> function must refer to the
same tables and columns on every execution; that is, you cannot use
a parameter as the name of a table or column in an SQL command. To get
around this restriction, you can construct dynamic commands using
the <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> <TT
CLASS="COMMAND"
>EXECUTE</TT
>
statement — at the price of performing new parse analysis and
constructing a new execution plan on every execution.
</P
><P
> The mutable nature of record variables presents another problem in this
connection. When fields of a record variable are used in
expressions or statements, the data types of the fields must not
change from one call of the function to the next, since each
expression will be analyzed using the data type that is present
when the expression is first reached. <TT
CLASS="COMMAND"
>EXECUTE</TT
> can be
used to get around this problem when necessary.
</P
><P
> If the same function is used as a trigger for more than one table,
<SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> prepares and caches statements
independently for each such table — that is, there is a cache
for each trigger function and table combination, not just for each
function. This alleviates some of the problems with varying
data types; for instance, a trigger function will be able to work
successfully with a column named <TT
CLASS="LITERAL"
>key</TT
> even if it happens
to have different types in different tables.
</P
><P
> Likewise, functions having polymorphic argument types have a separate
statement cache for each combination of actual argument types they have
been invoked for, so that data type differences do not cause unexpected
failures.
</P
><P
> Statement caching can sometimes have surprising effects on the
interpretation of time-sensitive values. For example there
is a difference between what these two functions do:
</P><PRE
CLASS="PROGRAMLISTING"
>CREATE FUNCTION logfunc1(logtxt text) RETURNS void AS $$
BEGIN
INSERT INTO logtable VALUES (logtxt, 'now');
END;
$$ LANGUAGE plpgsql;</PRE
><P>
and:
</P><PRE
CLASS="PROGRAMLISTING"
>CREATE FUNCTION logfunc2(logtxt text) RETURNS void AS $$
DECLARE
curtime timestamp;
BEGIN
curtime := 'now';
INSERT INTO logtable VALUES (logtxt, curtime);
END;
$$ LANGUAGE plpgsql;</PRE
><P>
</P
><P
> In the case of <CODE
CLASS="FUNCTION"
>logfunc1</CODE
>, the
<SPAN
CLASS="PRODUCTNAME"
>PostgreSQL</SPAN
> main parser knows when
analyzing the <TT
CLASS="COMMAND"
>INSERT</TT
> that the
string <TT
CLASS="LITERAL"
>'now'</TT
> should be interpreted as
<TT
CLASS="TYPE"
>timestamp</TT
>, because the target column of
<CODE
CLASS="CLASSNAME"
>logtable</CODE
> is of that type. Thus,
<TT
CLASS="LITERAL"
>'now'</TT
> will be converted to a <TT
CLASS="TYPE"
>timestamp</TT
>
constant when the
<TT
CLASS="COMMAND"
>INSERT</TT
> is analyzed, and then used in all
invocations of <CODE
CLASS="FUNCTION"
>logfunc1</CODE
> during the lifetime
of the session. Needless to say, this isn't what the programmer
wanted. A better idea is to use the <TT
CLASS="LITERAL"
>now()</TT
> or
<TT
CLASS="LITERAL"
>current_timestamp</TT
> function.
</P
><P
> In the case of <CODE
CLASS="FUNCTION"
>logfunc2</CODE
>, the
<SPAN
CLASS="PRODUCTNAME"
>PostgreSQL</SPAN
> main parser does not know
what type <TT
CLASS="LITERAL"
>'now'</TT
> should become and therefore
it returns a data value of type <TT
CLASS="TYPE"
>text</TT
> containing the string
<TT
CLASS="LITERAL"
>now</TT
>. During the ensuing assignment
to the local variable <TT
CLASS="VARNAME"
>curtime</TT
>, the
<SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
> interpreter casts this
string to the <TT
CLASS="TYPE"
>timestamp</TT
> type by calling the
<CODE
CLASS="FUNCTION"
>text_out</CODE
> and <CODE
CLASS="FUNCTION"
>timestamp_in</CODE
>
functions for the conversion. So, the computed time stamp is updated
on each execution as the programmer expects. Even though this
happens to work as expected, it's not terribly efficient, so
use of the <TT
CLASS="LITERAL"
>now()</TT
> function would still be a better idea.
</P
></DIV
></DIV
><DIV
CLASS="NAVFOOTER"
><HR
ALIGN="LEFT"
WIDTH="100%"><TABLE
SUMMARY="Footer navigation table"
WIDTH="100%"
BORDER="0"
CELLPADDING="0"
CELLSPACING="0"
><TR
><TD
WIDTH="33%"
ALIGN="left"
VALIGN="top"
><A
HREF="plpgsql-trigger.html"
ACCESSKEY="P"
>Prev</A
></TD
><TD
WIDTH="34%"
ALIGN="center"
VALIGN="top"
><A
HREF="index.html"
ACCESSKEY="H"
>Home</A
></TD
><TD
WIDTH="33%"
ALIGN="right"
VALIGN="top"
><A
HREF="plpgsql-development-tips.html"
ACCESSKEY="N"
>Next</A
></TD
></TR
><TR
><TD
WIDTH="33%"
ALIGN="left"
VALIGN="top"
>Trigger Procedures</TD
><TD
WIDTH="34%"
ALIGN="center"
VALIGN="top"
><A
HREF="plpgsql.html"
ACCESSKEY="U"
>Up</A
></TD
><TD
WIDTH="33%"
ALIGN="right"
VALIGN="top"
>Tips for Developing in <SPAN
CLASS="APPLICATION"
>PL/pgSQL</SPAN
></TD
></TR
></TABLE
></DIV
></BODY
></HTML
>
