Repair memory leaks in plpython.

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

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

<chapter id="typeconv">
<title>Type Conversion</title>

<indexterm zone="typeconv">
 <primary>data type</primary>
 <secondary>conversion</secondary>
</indexterm>

<para>
<acronym>SQL</acronym> statements can, intentionally or not, require
the mixing of different data types in the same expression.
<productname>PostgreSQL</productname> has extensive facilities for
evaluating mixed-type expressions.
</para>

<para>
In many cases a user does not need
to understand the details of the type conversion mechanism.
However, implicit conversions done by <productname>PostgreSQL</productname>
can affect the results of a query.  When necessary, these results
can be tailored by using <emphasis>explicit</emphasis> type conversion.
</para>

<para>
This chapter introduces the <productname>PostgreSQL</productname>
type conversion mechanisms and conventions.
Refer to the relevant sections in <xref linkend="datatype"/> and <xref linkend="functions"/>
for more information on specific data types and allowed functions and
operators.
</para>

<sect1 id="typeconv-overview">
<title>Overview</title>

<para>
<acronym>SQL</acronym> is a strongly typed language. That is, every data item
has an associated data type which determines its behavior and allowed usage.
<productname>PostgreSQL</productname> has an extensible type system that is
more general and flexible than other <acronym>SQL</acronym> implementations.
Hence, most type conversion behavior in <productname>PostgreSQL</productname>
is governed by general rules rather than by ad hoc
heuristics.  This allows the use of mixed-type expressions even with
user-defined types.
</para>

<para>
The <productname>PostgreSQL</productname> scanner/parser divides lexical
elements into five fundamental categories: integers, non-integer numbers,
strings, identifiers, and key words.  Constants of most non-numeric types are
first classified as strings. The <acronym>SQL</acronym> language definition
allows specifying type names with strings, and this mechanism can be used in
<productname>PostgreSQL</productname> to start the parser down the correct
path. For example, the query:

<screen>
SELECT text 'Origin' AS "label", point '(0,0)' AS "value";

 label  | value
--------+-------
 Origin | (0,0)
(1 row)
</screen>

has two literal constants, of type <type>text</type> and <type>point</type>.
If a type is not specified for a string literal, then the placeholder type
<type>unknown</type> is assigned initially, to be resolved in later
stages as described below.
</para>

<para>
There are four fundamental <acronym>SQL</acronym> constructs requiring
distinct type conversion rules in the <productname>PostgreSQL</productname>
parser:

<variablelist>
<varlistentry>
<term>
Function calls
</term>
<listitem>
<para>
Much of the <productname>PostgreSQL</productname> type system is built around a
rich set of functions. Functions can have one or more arguments.
Since <productname>PostgreSQL</productname> permits function
overloading, the function name alone does not uniquely identify the function
to be called; the parser must select the right function based on the data
types of the supplied arguments.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>
Operators
</term>
<listitem>
<para>
<productname>PostgreSQL</productname> allows expressions with
prefix (one-argument) operators,
as well as infix (two-argument) operators.  Like functions, operators can
be overloaded, so the same problem of selecting the right operator
exists.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>
Value Storage
</term>
<listitem>
<para>
<acronym>SQL</acronym> <command>INSERT</command> and <command>UPDATE</command> statements place the results of
expressions into a table. The expressions in the statement must be matched up
with, and perhaps converted to, the types of the target columns.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>
<literal>UNION</literal>, <literal>CASE</literal>, and related constructs
</term>
<listitem>
<para>
Since all query results from a unionized <command>SELECT</command> statement
must appear in a single set of columns, the types of the results of each
<command>SELECT</command> clause must be matched up and converted to a uniform set.
Similarly, the result expressions of a <literal>CASE</literal> construct must be
converted to a common type so that the <literal>CASE</literal> expression as a whole
has a known output type.  Some other constructs, such
as <literal>ARRAY[]</literal> and the <function>GREATEST</function>
and <function>LEAST</function> functions, likewise require determination of a
common type for several subexpressions.
</para>
</listitem>
</varlistentry>
</variablelist>
</para>

<para>
The system catalogs store information about which conversions, or
<firstterm>casts</firstterm>, exist between which data types, and how to
perform those conversions.  Additional casts can be added by the user
with the <xref linkend="sql-createcast"/>
command.  (This is usually
done in conjunction with defining new data types.  The set of casts
between built-in types has been carefully crafted and is best not
altered.)
</para>

<indexterm>
 <primary>data type</primary>
 <secondary>category</secondary>
</indexterm>

<para>
An additional heuristic provided by the parser allows improved determination
of the proper casting behavior among groups of types that have implicit casts.
Data types are divided into several basic <firstterm>type
categories</firstterm>, including <type>boolean</type>, <type>numeric</type>,
<type>string</type>, <type>bitstring</type>, <type>datetime</type>,
<type>timespan</type>, <type>geometric</type>, <type>network</type>, and
user-defined.  (For a list see <xref linkend="catalog-typcategory-table"/>;
but note it is also possible to create custom type categories.)  Within each
category there can be one or more <firstterm>preferred types</firstterm>, which
are preferred when there is a choice of possible types.  With careful selection
of preferred types and available implicit casts, it is possible to ensure that
ambiguous expressions (those with multiple candidate parsing solutions) can be
resolved in a useful way.
</para>

<para>
All type conversion rules are designed with several principles in mind:

<itemizedlist>
<listitem>
<para>
Implicit conversions should never have surprising or unpredictable outcomes.
</para>
</listitem>

<listitem>
<para>
There should be no extra overhead in the parser or executor
if a query does not need implicit type conversion.
That is, if a query is well-formed and the types already match, then the query should execute
without spending extra time in the parser and without introducing unnecessary implicit conversion
calls in the query.
</para>
</listitem>

<listitem>
<para>
Additionally, if a query usually requires an implicit conversion for a function, and
if then the user defines a new function with the correct argument types, the parser
should use this new function and no longer do implicit conversion to use the old function.
</para>
</listitem>
</itemizedlist>
</para>

</sect1>

<sect1 id="typeconv-oper">
<title>Operators</title>

<indexterm zone="typeconv-oper">
 <primary>operator</primary>
 <secondary>type resolution in an invocation</secondary>
</indexterm>

  <para>
   The specific operator that is referenced by an operator expression
   is determined using the following procedure.
   Note that this procedure is indirectly affected
   by the precedence of the operators involved, since that will determine
   which sub-expressions are taken to be the inputs of which operators.
   See <xref linkend="sql-precedence"/> for more information.
  </para>

<procedure>
<title>Operator Type Resolution</title>

<step id="op-resol-select" performance="required">
<para>
Select the operators to be considered from the
<classname>pg_operator</classname> system catalog.  If a non-schema-qualified
operator name was used (the usual case), the operators
considered are those with the matching name and argument count that are
visible in the current search path (see <xref linkend="ddl-schemas-path"/>).
If a qualified operator name was given, only operators in the specified
schema are considered.
</para>

<substeps>
<step performance="optional">
<para>
If the search path finds multiple operators with identical argument types,
only the one appearing earliest in the path is considered.  Operators with
different argument types are considered on an equal footing regardless of
search path position.
</para>
</step>
</substeps>
</step>

<step id="op-resol-exact-match" performance="required">
<para>
Check for an operator accepting exactly the input argument types.
If one exists (there can be only one exact match in the set of
operators considered), use it.  Lack of an exact match creates a security
hazard when calling, via qualified name
  <footnote id="op-qualified-security">
   <!-- If you edit this, consider editing func-qualified-security. -->
   <para>
    The hazard does not arise with a non-schema-qualified name, because a
    search path containing schemas that permit untrusted users to create
    objects is not a <link linkend="ddl-schemas-patterns">secure schema usage
    pattern</link>.
   </para>
  </footnote>
(not typical), any operator found in a schema that permits untrusted users to
create objects.  In such situations, cast arguments to force an exact match.
</para>

<substeps>
<step id="op-resol-exact-unknown" performance="optional">
<para>
If one argument of a binary operator invocation is of the <type>unknown</type> type,
then assume it is the same type as the other argument for this check.
Invocations involving two <type>unknown</type> inputs, or a prefix operator
with an <type>unknown</type> input, will never find a match at this step.
</para>
</step>
<step id="op-resol-exact-domain" performance="optional">
<para>
If one argument of a binary operator invocation is of the <type>unknown</type>
type and the other is of a domain type, next check to see if there is an
operator accepting exactly the domain's base type on both sides; if so, use it.
</para>
</step>
</substeps>
</step>

<step id="op-resol-best-match" performance="required">
<para>
Look for the best match.
</para>
<substeps>
<step performance="required">
<para>
Discard candidate operators for which the input types do not match
and cannot be converted (using an implicit conversion) to match.
<type>unknown</type> literals are
assumed to be convertible to anything for this purpose.  If only one
candidate remains, use it; else continue to the next step.
</para>
</step>
<step performance="required">
<para>
If any input argument is of a domain type, treat it as being of the
domain's base type for all subsequent steps.  This ensures that domains
act like their base types for purposes of ambiguous-operator resolution.
</para>
</step>
<step performance="required">
<para>
Run through all candidates and keep those with the most exact matches
on input types.  Keep all candidates if none have exact matches.
If only one candidate remains, use it; else continue to the next step.
</para>
</step>
<step performance="required">
<para>
Run through all candidates and keep those that accept preferred types (of the
input data type's type category) at the most positions where type conversion
will be required.
Keep all candidates if none accept preferred types.
If only one candidate remains, use it; else continue to the next step.
</para>
</step>
<step performance="required">
<para>
If any input arguments are <type>unknown</type>, check the type
categories accepted at those argument positions by the remaining
candidates.  At each position, select the <type>string</type> category
if any
candidate accepts that category.  (This bias towards string is appropriate
since an unknown-type literal looks like a string.) Otherwise, if
all the remaining candidates accept the same type category, select that
category; otherwise fail because the correct choice cannot be deduced
without more clues.  Now discard
candidates that do not accept the selected type category.  Furthermore,
if any candidate accepts a preferred type in that category,
discard candidates that accept non-preferred types for that argument.
Keep all candidates if none survive these tests.
If only one candidate remains, use it; else continue to the next step.
</para>
</step>
<step id="op-resol-last-unknown" performance="required">
<para>
If there are both <type>unknown</type> and known-type arguments, and all
the known-type arguments have the same type, assume that the
<type>unknown</type> arguments are also of that type, and check which
candidates can accept that type at the <type>unknown</type>-argument
positions.  If exactly one candidate passes this test, use it.
Otherwise, fail.
</para>
</step>
</substeps>
</step>
</procedure>

<para>
Some examples follow.
</para>

<example>
<title>Square Root Operator Type Resolution</title>

<para>
There is only one square root operator (prefix <literal>|/</literal>)
defined in the standard catalog, and it takes an argument of type
<type>double precision</type>.
The scanner assigns an initial type of <type>integer</type> to the argument
in this query expression:
<screen>
SELECT |/ 40 AS "square root of 40";
 square root of 40
-------------------
 6.324555320336759
(1 row)
</screen>

So the parser does a type conversion on the operand and the query
is equivalent to:

<screen>
SELECT |/ CAST(40 AS double precision) AS "square root of 40";
</screen>
</para>
</example>

<example>
<title>String Concatenation Operator Type Resolution</title>

<para>
A string-like syntax is used for working with string types and for
working with complex extension types.
Strings with unspecified type are matched with likely operator candidates.
</para>

<para>
An example with one unspecified argument:
<screen>
SELECT text 'abc' || 'def' AS "text and unknown";

 text and unknown
------------------
 abcdef
(1 row)
</screen>
</para>

<para>
In this case the parser looks to see if there is an operator taking <type>text</type>
for both arguments. Since there is, it assumes that the second argument should
be interpreted as type <type>text</type>.
</para>

<para>
Here is a concatenation of two values of unspecified types:
<screen>
SELECT 'abc' || 'def' AS "unspecified";

 unspecified
-------------
 abcdef
(1 row)
</screen>
</para>

<para>
In this case there is no initial hint for which type to use, since no types
are specified in the query. So, the parser looks for all candidate operators
and finds that there are candidates accepting both string-category and
bit-string-category inputs.  Since string category is preferred when available,
that category is selected, and then the
preferred type for strings, <type>text</type>, is used as the specific
type to resolve the unknown-type literals as.
</para>
</example>

<example>
<title>Absolute-Value and Negation Operator Type Resolution</title>

<para>
The <productname>PostgreSQL</productname> operator catalog has several
entries for the prefix operator <literal>@</literal>, all of which implement
absolute-value operations for various numeric data types.  One of these
entries is for type <type>float8</type>, which is the preferred type in
the numeric category.  Therefore, <productname>PostgreSQL</productname>
will use that entry when faced with an <type>unknown</type> input:
<screen>
SELECT @ '-4.5' AS "abs";
 abs
-----
 4.5
(1 row)
</screen>
Here the system has implicitly resolved the unknown-type literal as type
<type>float8</type> before applying the chosen operator.  We can verify that
<type>float8</type> and not some other type was used:
<screen>
SELECT @ '-4.5e500' AS "abs";

ERROR:  "-4.5e500" is out of range for type double precision
</screen>
</para>

<para>
On the other hand, the prefix operator <literal>~</literal> (bitwise negation)
is defined only for integer data types, not for <type>float8</type>.  So, if we
try a similar case with <literal>~</literal>, we get:
<screen>
SELECT ~ '20' AS "negation";

ERROR:  operator is not unique: ~ "unknown"
HINT:  Could not choose a best candidate operator. You might need to add
explicit type casts.
</screen>
This happens because the system cannot decide which of the several
possible <literal>~</literal> operators should be preferred.  We can help
it out with an explicit cast:
<screen>
SELECT ~ CAST('20' AS int8) AS "negation";

 negation
----------
      -21
(1 row)
</screen>
</para>
</example>

<example>
<title>Array Inclusion Operator Type Resolution</title>

<para>
Here is another example of resolving an operator with one known and one
unknown input:
<screen>
SELECT array[1,2] &lt;@ '{1,2,3}' as "is subset";

 is subset
-----------
 t
(1 row)
</screen>
The <productname>PostgreSQL</productname> operator catalog has several
entries for the infix operator <literal>&lt;@</literal>, but the only two that
could possibly accept an integer array on the left-hand side are
array inclusion (<type>anyarray</type> <literal>&lt;@</literal> <type>anyarray</type>)
and range inclusion (<type>anyelement</type> <literal>&lt;@</literal> <type>anyrange</type>).
Since none of these polymorphic pseudo-types (see <xref
linkend="datatype-pseudo"/>) are considered preferred, the parser cannot
resolve the ambiguity on that basis.
However, <xref linkend="op-resol-last-unknown"/> tells
it to assume that the unknown-type literal is of the same type as the other
input, that is, integer array.  Now only one of the two operators can match,
so array inclusion is selected.  (Had range inclusion been selected, we would
have gotten an error, because the string does not have the right format to be
a range literal.)
</para>
</example>

<example>
<title>Custom Operator on a Domain Type</title>

<para>
Users sometimes try to declare operators applying just to a domain type.
This is possible but is not nearly as useful as it might seem, because the
operator resolution rules are designed to select operators applying to the
domain's base type.  As an example consider
<screen>
CREATE DOMAIN mytext AS text CHECK(...);
CREATE FUNCTION mytext_eq_text (mytext, text) RETURNS boolean AS ...;
CREATE OPERATOR = (procedure=mytext_eq_text, leftarg=mytext, rightarg=text);
CREATE TABLE mytable (val mytext);

SELECT * FROM mytable WHERE val = 'foo';
</screen>
This query will not use the custom operator.  The parser will first see if
there is a <type>mytext</type> <literal>=</literal> <type>mytext</type> operator
(<xref linkend="op-resol-exact-unknown"/>), which there is not;
then it will consider the domain's base type <type>text</type>, and see if
there is a <type>text</type> <literal>=</literal> <type>text</type> operator
(<xref linkend="op-resol-exact-domain"/>), which there is;
so it resolves the <type>unknown</type>-type literal as <type>text</type> and
uses the <type>text</type> <literal>=</literal> <type>text</type> operator.
The only way to get the custom operator to be used is to explicitly cast
the literal:
<screen>
SELECT * FROM mytable WHERE val = text 'foo';
</screen>
so that the <type>mytext</type> <literal>=</literal> <type>text</type> operator is found
immediately according to the exact-match rule.  If the best-match rules
are reached, they actively discriminate against operators on domain types.
If they did not, such an operator would create too many ambiguous-operator
failures, because the casting rules always consider a domain as castable
to or from its base type, and so the domain operator would be considered
usable in all the same cases as a similarly-named operator on the base type.
</para>
</example>

</sect1>

<sect1 id="typeconv-func">
<title>Functions</title>

<indexterm zone="typeconv-func">
 <primary>function</primary>
 <secondary>type resolution in an invocation</secondary>
</indexterm>

  <para>
   The specific function that is referenced by a function call
   is determined using the following procedure.
  </para>

<procedure>
<title>Function Type Resolution</title>

<step performance="required">
<para>
Select the functions to be considered from the
<classname>pg_proc</classname> system catalog.  If a non-schema-qualified
function name was used, the functions
considered are those with the matching name and argument count that are
visible in the current search path (see <xref linkend="ddl-schemas-path"/>).
If a qualified function name was given, only functions in the specified
schema are considered.
</para>

<substeps>
<step performance="optional">
<para>
If the search path finds multiple functions of identical argument types,
only the one appearing earliest in the path is considered.  Functions of
different argument types are considered on an equal footing regardless of
search path position.
</para>
</step>
<step performance="optional">
<para>
If a function is declared with a <literal>VARIADIC</literal> array parameter, and
the call does not use the <literal>VARIADIC</literal> keyword, then the function
is treated as if the array parameter were replaced by one or more occurrences
of its element type, as needed to match the call.  After such expansion the
function might have effective argument types identical to some non-variadic
function.  In that case 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>
This creates a security hazard when calling, via qualified name
  <footnote id="func-qualified-security">
   <!-- If you edit this, consider editing op-qualified-security. -->
   <para>
    The hazard does not arise with a non-schema-qualified name, because a
    search path containing schemas that permit untrusted users to create
    objects is not a <link linkend="ddl-schemas-patterns">secure schema usage
    pattern</link>.
   </para>
  </footnote>,
a variadic function found in a schema that permits untrusted users to create
objects.  A malicious user can take control and execute arbitrary SQL
functions as though you executed them.  Substitute a call bearing
the <literal>VARIADIC</literal> keyword, which bypasses this hazard.  Calls
populating <literal>VARIADIC "any"</literal> parameters often have no
equivalent formulation containing the <literal>VARIADIC</literal> keyword.  To
issue those calls safely, the function's schema must permit only trusted users
to create objects.
</para>
</step>
<step performance="optional">
<para>
Functions that have default values for parameters are considered to match any
call that omits zero or more of the defaultable parameter positions.  If more
than one such function matches a call, the one appearing earliest in the
search path is used.  If there are two or more such functions in the same
schema with identical parameter types in the non-defaulted positions (which is
possible if they have different sets of defaultable parameters), the system
will not be able to determine which to prefer, and so an <quote>ambiguous
function call</quote> error will result if no better match to the call can be
found.
</para>
<para>
This creates an availability hazard when calling, via qualified
name<footnoteref linkend="func-qualified-security"/>, any function found in a
schema that permits untrusted users to create objects.  A malicious user can
create a function with the name of an existing function, replicating that
function's parameters and appending novel parameters having default values.
This precludes new calls to the original function.  To forestall this hazard,
place functions in schemas that permit only trusted users to create objects.
</para>
</step>
</substeps>
</step>

<step performance="required">
<para>
Check for a function accepting exactly the input argument types.
If one exists (there can be only one exact match in the set of
functions considered), use it.  Lack of an exact match creates a security
hazard when calling, via qualified
name<footnoteref linkend="func-qualified-security"/>, a function found in a
schema that permits untrusted users to create objects.  In such situations,
cast arguments to force an exact match.  (Cases involving <type>unknown</type>
will never find a match at this step.)
</para>
</step>

<step performance="required">
<para>
If no exact match is found, see if the function call appears
to be a special type conversion request.  This happens if the function call
has just one argument and the function name is the same as the (internal)
name of some data type.  Furthermore, the function argument must be either
an unknown-type literal, or a type that is binary-coercible to the named
data type, or a type that could be converted to the named data type by
applying that type's I/O functions (that is, the conversion is either to or
from one of the standard string types).  When these conditions are met,
the function call is treated as a form of <literal>CAST</literal> specification.
  <footnote>
   <para>
    The reason for this step is to support function-style cast specifications
    in cases where there is not an actual cast function.  If there is a cast
    function, it is conventionally named after its output type, and so there
    is no need to have a special case.  See
    <xref linkend="sql-createcast"/>
    for additional commentary.
   </para>
  </footnote>
</para>
</step>
<step performance="required">
<para>
Look for the best match.
</para>
<substeps>
<step performance="required">
<para>
Discard candidate functions for which the input types do not match
and cannot be converted (using an implicit conversion) to match.
<type>unknown</type> literals are
assumed to be convertible to anything for this purpose.  If only one
candidate remains, use it; else continue to the next step.
</para>
</step>
<step performance="required">
<para>
If any input argument is of a domain type, treat it as being of the
domain's base type for all subsequent steps.  This ensures that domains
act like their base types for purposes of ambiguous-function resolution.
</para>
</step>
<step performance="required">
<para>
Run through all candidates and keep those with the most exact matches
on input types.  Keep all candidates if none have exact matches.
If only one candidate remains, use it; else continue to the next step.
</para>
</step>
<step performance="required">
<para>
Run through all candidates and keep those that accept preferred types (of the
input data type's type category) at the most positions where type conversion
will be required.
Keep all candidates if none accept preferred types.
If only one candidate remains, use it; else continue to the next step.
</para>
</step>
<step performance="required">
<para>
If any input arguments are <type>unknown</type>, check the type categories
accepted
at those argument positions by the remaining candidates.  At each position,
select the <type>string</type> category if any candidate accepts that category.
(This bias towards string
is appropriate since an unknown-type literal looks like a string.)
Otherwise, if all the remaining candidates accept the same type category,
select that category; otherwise fail because
the correct choice cannot be deduced without more clues.
Now discard candidates that do not accept the selected type category.
Furthermore, if any candidate accepts a preferred type in that category,
discard candidates that accept non-preferred types for that argument.
Keep all candidates if none survive these tests.
If only one candidate remains, use it; else continue to the next step.
</para>
</step>
<step performance="required">
<para>
If there are both <type>unknown</type> and known-type arguments, and all
the known-type arguments have the same type, assume that the
<type>unknown</type> arguments are also of that type, and check which
candidates can accept that type at the <type>unknown</type>-argument
positions.  If exactly one candidate passes this test, use it.
Otherwise, fail.
</para>
</step>
</substeps>
</step>
</procedure>

<para>
Note that the <quote>best match</quote> rules are identical for operator and
function type resolution.
Some examples follow.
</para>

<example>
<title>Rounding Function Argument Type Resolution</title>

<para>
There is only one <function>round</function> function that takes two
arguments; it takes a first argument of type <type>numeric</type> and
a second argument of type <type>integer</type>.
So the following query automatically converts
the first argument of type <type>integer</type> to
<type>numeric</type>:

<screen>
SELECT round(4, 4);

 round
--------
 4.0000
(1 row)
</screen>

That query is actually transformed by the parser to:
<screen>
SELECT round(CAST (4 AS numeric), 4);
</screen>
</para>

<para>
Since numeric constants with decimal points are initially assigned the
type <type>numeric</type>, the following query will require no type
conversion and therefore might be slightly more efficient:
<screen>
SELECT round(4.0, 4);
</screen>
</para>
</example>

<example>
<title>Variadic Function Resolution</title>

<para>
<screen>
CREATE FUNCTION public.variadic_example(VARIADIC numeric[]) RETURNS int
  LANGUAGE sql AS 'SELECT 1';
CREATE FUNCTION
</screen>

This function accepts, but does not require, the VARIADIC keyword.  It
tolerates both integer and numeric arguments:

<screen>
SELECT public.variadic_example(0),
       public.variadic_example(0.0),
       public.variadic_example(VARIADIC array[0.0]);
 variadic_example | variadic_example | variadic_example
------------------+------------------+------------------
                1 |                1 |                1
(1 row)
</screen>

However, the first and second calls will prefer more-specific functions, if
available:

<screen>
CREATE FUNCTION public.variadic_example(numeric) RETURNS int
  LANGUAGE sql AS 'SELECT 2';
CREATE FUNCTION

CREATE FUNCTION public.variadic_example(int) RETURNS int
  LANGUAGE sql AS 'SELECT 3';
CREATE FUNCTION

SELECT public.variadic_example(0),
       public.variadic_example(0.0),
       public.variadic_example(VARIADIC array[0.0]);
 variadic_example | variadic_example | variadic_example
------------------+------------------+------------------
                3 |                2 |                1
(1 row)
</screen>

Given the default configuration and only the first function existing, the
first and second calls are insecure.  Any user could intercept them by
creating the second or third function.  By matching the argument type exactly
and using the <literal>VARIADIC</literal> keyword, the third call is secure.
</para>
</example>

<example>
<title>Substring Function Type Resolution</title>

<para>
There are several <function>substr</function> functions, one of which
takes types <type>text</type> and <type>integer</type>.  If called
with a string constant of unspecified type, the system chooses the
candidate function that accepts an argument of the preferred category
<literal>string</literal> (namely of type <type>text</type>).

<screen>
SELECT substr('1234', 3);

 substr
--------
     34
(1 row)
</screen>
</para>

<para>
If the string is declared to be of type <type>varchar</type>, as might be the case
if it comes from a table, then the parser will try to convert it to become <type>text</type>:
<screen>
SELECT substr(varchar '1234', 3);

 substr
--------
     34
(1 row)
</screen>

This is transformed by the parser to effectively become:
<screen>
SELECT substr(CAST (varchar '1234' AS text), 3);
</screen>
</para>
<para>
<note>
<para>
The parser learns from the <structname>pg_cast</structname> catalog that
<type>text</type> and <type>varchar</type>
are binary-compatible, meaning that one can be passed to a function that
accepts the other without doing any physical conversion.  Therefore, no
type conversion call is really inserted in this case.
</para>
</note>
</para>

<para>
And, if the function is called with an argument of type <type>integer</type>,
the parser will try to convert that to <type>text</type>:
<screen>
SELECT substr(1234, 3);
ERROR:  function substr(integer, integer) does not exist
HINT:  No function matches the given name and argument types. You might need
to add explicit type casts.
</screen>

This does not work because <type>integer</type> does not have an implicit cast
to <type>text</type>.  An explicit cast will work, however:
<screen>
SELECT substr(CAST (1234 AS text), 3);

 substr
--------
     34
(1 row)
</screen>
</para>
</example>

</sect1>

<sect1 id="typeconv-query">
<title>Value Storage</title>

  <para>
   Values to be inserted into a table are converted to the destination
   column's data type according to the
   following steps.
  </para>

<procedure>
<title>Value Storage Type Conversion</title>

<step performance="required">
<para>
Check for an exact match with the target.
</para>
</step>

<step performance="required">
<para>
Otherwise, try to convert the expression to the target type.  This is possible
if an <firstterm>assignment cast</firstterm> between the two types is registered in the
<structname>pg_cast</structname> catalog (see <xref linkend="sql-createcast"/>).
Alternatively, if the expression is an unknown-type literal, the contents of
the literal string will be fed to the input conversion routine for the target
type.
</para>
</step>

<step performance="required">
<para>
Check to see if there is a sizing cast for the target type.  A sizing
cast is a cast from that type to itself.  If one is found in the
<structname>pg_cast</structname> catalog, apply it to the expression before storing
into the destination column.  The implementation function for such a cast
always takes an extra parameter of type <type>integer</type>, which receives
the destination column's <structfield>atttypmod</structfield> value (typically its
declared length, although the interpretation of <structfield>atttypmod</structfield>
varies for different data types), and it may take a third <type>boolean</type>
parameter that says whether the cast is explicit or implicit.  The cast
function
is responsible for applying any length-dependent semantics such as size
checking or truncation.
</para>
</step>

</procedure>

<example>
<title><type>character</type> Storage Type Conversion</title>

<para>
For a target column declared as <type>character(20)</type> the following
statement shows that the stored value is sized correctly:

<screen>
CREATE TABLE vv (v character(20));
INSERT INTO vv SELECT 'abc' || 'def';
SELECT v, octet_length(v) FROM vv;

          v           | octet_length
----------------------+--------------
 abcdef               |           20
(1 row)
</screen>
</para>

<para>
What has really happened here is that the two unknown literals are resolved
to <type>text</type> by default, allowing the <literal>||</literal> operator
to be resolved as <type>text</type> concatenation.  Then the <type>text</type>
result of the operator is converted to <type>bpchar</type> (<quote>blank-padded
char</quote>, the internal name of the <type>character</type> data type) to match the target
column type.  (Since the conversion from <type>text</type> to
<type>bpchar</type> is binary-coercible, this conversion does
not insert any real function call.)  Finally, the sizing function
<literal>bpchar(bpchar, integer, boolean)</literal> is found in the system catalog
and applied to the operator's result and the stored column length.  This
type-specific function performs the required length check and addition of
padding spaces.
</para>
</example>
</sect1>

<sect1 id="typeconv-union-case">
<title><literal>UNION</literal>, <literal>CASE</literal>, and Related Constructs</title>

<indexterm zone="typeconv-union-case">
 <primary>UNION</primary>
 <secondary>determination of result type</secondary>
</indexterm>

<indexterm zone="typeconv-union-case">
 <primary>CASE</primary>
 <secondary>determination of result type</secondary>
</indexterm>

<indexterm zone="typeconv-union-case">
 <primary>ARRAY</primary>
 <secondary>determination of result type</secondary>
</indexterm>

<indexterm zone="typeconv-union-case">
 <primary>VALUES</primary>
 <secondary>determination of result type</secondary>
</indexterm>

<indexterm zone="typeconv-union-case">
 <primary>GREATEST</primary>
 <secondary>determination of result type</secondary>
</indexterm>

<indexterm zone="typeconv-union-case">
 <primary>LEAST</primary>
 <secondary>determination of result type</secondary>
</indexterm>

<para>
SQL <literal>UNION</literal> constructs must match up possibly dissimilar
types to become a single result set.  The resolution algorithm is
applied separately to each output column of a union query.  The
<literal>INTERSECT</literal> and <literal>EXCEPT</literal> constructs resolve
dissimilar types in the same way as <literal>UNION</literal>.
Some other constructs, including
<literal>CASE</literal>, <literal>ARRAY</literal>, <literal>VALUES</literal>,
and the <function>GREATEST</function> and <function>LEAST</function>
functions, use the identical
algorithm to match up their component expressions and select a result
data type.
</para>

<procedure>
<title>Type Resolution for <literal>UNION</literal>, <literal>CASE</literal>,
and Related Constructs</title>

<step performance="required">
<para>
If all inputs are of the same type, and it is not <type>unknown</type>,
resolve as that type.
</para>
</step>

<step performance="required">
<para>
If any input is of a domain type, treat it as being of the
domain's base type for all subsequent steps.
  <footnote>
   <para>
    Somewhat like the treatment of domain inputs for operators and
    functions, this behavior allows a domain type to be preserved through
    a <literal>UNION</literal> or similar construct, so long as the user is
    careful to ensure that all inputs are implicitly or explicitly of that
    exact type.  Otherwise the domain's base type will be used.
   </para>
  </footnote>
</para>
</step>

<step performance="required">
<para>
If all inputs are of type <type>unknown</type>, resolve as type
<type>text</type> (the preferred type of the string category).
Otherwise, <type>unknown</type> inputs are ignored for the purposes
of the remaining rules.
</para>
</step>

<step performance="required">
<para>
If the non-unknown inputs are not all of the same type category, fail.
</para>
</step>

<step performance="required">
<para>
Select the first non-unknown input type as the candidate type,
then consider each other non-unknown input type, left to right.
  <footnote>
   <para>
    For historical reasons, <literal>CASE</literal> treats
    its <literal>ELSE</literal> clause (if any) as the <quote>first</quote>
    input, with the <literal>THEN</literal> clauses(s) considered after
    that.  In all other cases, <quote>left to right</quote> means the order
    in which the expressions appear in the query text.
   </para>
  </footnote>
If the candidate type can be implicitly converted to the other type,
but not vice-versa, select the other type as the new candidate type.
Then continue considering the remaining inputs.  If, at any stage of this
process, a preferred type is selected, stop considering additional
inputs.
</para>
</step>

<step performance="required">
<para>
Convert all inputs to the final candidate type.  Fail if there is not an
implicit conversion from a given input type to the candidate type.
</para>
</step>
</procedure>

<para>
Some examples follow.
</para>

<example>
<title>Type Resolution with Underspecified Types in a Union</title>

<para>
<screen>
SELECT text 'a' AS "text" UNION SELECT 'b';

 text
------
 a
 b
(2 rows)
</screen>
Here, the unknown-type literal <literal>'b'</literal> will be resolved to type <type>text</type>.
</para>
</example>

<example>
<title>Type Resolution in a Simple Union</title>

<para>
<screen>
SELECT 1.2 AS "numeric" UNION SELECT 1;

 numeric
---------
       1
     1.2
(2 rows)
</screen>
The literal <literal>1.2</literal> is of type <type>numeric</type>,
and the <type>integer</type> value <literal>1</literal> can be cast implicitly to
<type>numeric</type>, so that type is used.
</para>
</example>

<example>
<title>Type Resolution in a Transposed Union</title>

<para>
<screen>
SELECT 1 AS "real" UNION SELECT CAST('2.2' AS REAL);

 real
------
    1
  2.2
(2 rows)
</screen>
Here, since type <type>real</type> cannot be implicitly cast to <type>integer</type>,
but <type>integer</type> can be implicitly cast to <type>real</type>, the union
result type is resolved as <type>real</type>.
</para>
</example>

<example>
<title>Type Resolution in a Nested Union</title>

<para>
<screen>
SELECT NULL UNION SELECT NULL UNION SELECT 1;

ERROR:  UNION types text and integer cannot be matched
</screen>
This failure occurs because <productname>PostgreSQL</productname> treats
multiple <literal>UNION</literal>s as a nest of pairwise operations;
that is, this input is the same as
<screen>
(SELECT NULL UNION SELECT NULL) UNION SELECT 1;
</screen>
The inner <literal>UNION</literal> is resolved as emitting
type <type>text</type>, according to the rules given above.  Then the
outer <literal>UNION</literal> has inputs of types <type>text</type>
and <type>integer</type>, leading to the observed error.  The problem
can be fixed by ensuring that the leftmost <literal>UNION</literal>
has at least one input of the desired result type.
</para>

<para>
<literal>INTERSECT</literal> and <literal>EXCEPT</literal> operations are
likewise resolved pairwise.  However, the other constructs described in this
section consider all of their inputs in one resolution step.
</para>
</example>
</sect1>

<sect1 id="typeconv-select">
<title><literal>SELECT</literal> Output Columns</title>

<indexterm zone="typeconv-select">
 <primary>SELECT</primary>
 <secondary>determination of result type</secondary>
</indexterm>

<para>
The rules given in the preceding sections will result in assignment
of non-<type>unknown</type> data types to all expressions in an SQL query,
except for unspecified-type literals that appear as simple output
columns of a <command>SELECT</command> command.  For example, in

<screen>
SELECT 'Hello World';
</screen>

there is nothing to identify what type the string literal should be
taken as.  In this situation <productname>PostgreSQL</productname> will fall back
to resolving the literal's type as <type>text</type>.
</para>

<para>
When the <command>SELECT</command> is one arm of a <literal>UNION</literal>
(or <literal>INTERSECT</literal> or <literal>EXCEPT</literal>) construct, or when it
appears within <command>INSERT ... SELECT</command>, this rule is not applied
since rules given in preceding sections take precedence.  The type of an
unspecified-type literal can be taken from the other <literal>UNION</literal> arm
in the first case, or from the destination column in the second case.
</para>

<para>
<literal>RETURNING</literal> lists are treated the same as <command>SELECT</command>
output lists for this purpose.
</para>

<note>
 <para>
  Prior to <productname>PostgreSQL</productname> 10, this rule did not exist, and
  unspecified-type literals in a <command>SELECT</command> output list were
  left as type <type>unknown</type>.  That had assorted bad consequences,
  so it's been changed.
 </para>
</note>

</sect1>
</chapter>