Baoding-FenShuaJiang
/
Green_GoodERP18_FSJ_Standard
Sledovat 1
Oblíbit 0
Wiki 捐赠
gooderp18绿色标准版
Nelze vybrat více než 25 témat Téma musí začínat písmenem nebo číslem, může obsahovat pomlčky („-“) a může být dlouhé až 35 znaků.
Strom: 0f75e31513
Větve Značky
${ item.name }
Vytvořit větev ${ searchTerm }
z „0f75e31513“
${ noResults }
Green_GoodERP18_FSJ_Standard/PostgreSQL/doc/postgresql/html/xfunc-c.html

1387 lines
77KB
Surový Blame Historie 

  1. <?xml version="1.0" encoding="UTF-8" standalone="no"?>

  2. <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>37.10. C-Language Functions</title><link rel="stylesheet" type="text/css" href="stylesheet.css" /><link rev="made" href="pgsql-docs@lists.postgresql.org" /><meta name="generator" content="DocBook XSL Stylesheets V1.79.1" /><link rel="prev" href="xfunc-internal.html" title="37.9. Internal Functions" /><link rel="next" href="xfunc-optimization.html" title="37.11. Function Optimization Information" /></head><body><div xmlns="http://www.w3.org/TR/xhtml1/transitional" class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="5" align="center">37.10. C-Language Functions</th></tr><tr><td width="10%" align="left"><a accesskey="p" href="xfunc-internal.html" title="37.9. Internal Functions">Prev</a> </td><td width="10%" align="left"><a accesskey="u" href="extend.html" title="Chapter 37. Extending SQL">Up</a></td><th width="60%" align="center">Chapter 37. Extending <acronym xmlns="http://www.w3.org/1999/xhtml" class="acronym">SQL</acronym></th><td width="10%" align="right"><a accesskey="h" href="index.html" title="PostgreSQL 12.4 Documentation">Home</a></td><td width="10%" align="right"> <a accesskey="n" href="xfunc-optimization.html" title="37.11. Function Optimization Information">Next</a></td></tr></table><hr></hr></div><div class="sect1" id="XFUNC-C"><div class="titlepage"><div><div><h2 class="title" style="clear: both">37.10. C-Language Functions</h2></div></div></div><div class="toc"><dl class="toc"><dt><span class="sect2"><a href="xfunc-c.html#XFUNC-C-DYNLOAD">37.10.1. Dynamic Loading</a></span></dt><dt><span class="sect2"><a href="xfunc-c.html#XFUNC-C-BASETYPE">37.10.2. Base Types in C-Language Functions</a></span></dt><dt><span class="sect2"><a href="xfunc-c.html#id-1.8.3.13.7">37.10.3. Version 1 Calling Conventions</a></span></dt><dt><span class="sect2"><a href="xfunc-c.html#id-1.8.3.13.8">37.10.4. Writing Code</a></span></dt><dt><span class="sect2"><a href="xfunc-c.html#DFUNC">37.10.5. Compiling and Linking Dynamically-Loaded Functions</a></span></dt><dt><span class="sect2"><a href="xfunc-c.html#id-1.8.3.13.10">37.10.6. Composite-Type Arguments</a></span></dt><dt><span class="sect2"><a href="xfunc-c.html#id-1.8.3.13.11">37.10.7. Returning Rows (Composite Types)</a></span></dt><dt><span class="sect2"><a href="xfunc-c.html#XFUNC-C-RETURN-SET">37.10.8. Returning Sets</a></span></dt><dt><span class="sect2"><a href="xfunc-c.html#id-1.8.3.13.13">37.10.9. Polymorphic Arguments and Return Types</a></span></dt><dt><span class="sect2"><a href="xfunc-c.html#id-1.8.3.13.14">37.10.10. Shared Memory and LWLocks</a></span></dt><dt><span class="sect2"><a href="xfunc-c.html#EXTEND-CPP">37.10.11. Using C++ for Extensibility</a></span></dt></dl></div><a id="id-1.8.3.13.2" class="indexterm"></a><p>

  3.     User-defined functions can be written in C (or a language that can

  4.     be made compatible with C, such as C++).  Such functions are

  5.     compiled into dynamically loadable objects (also called shared

  6.     libraries) and are loaded by the server on demand.  The dynamic

  7.     loading feature is what distinguishes <span class="quote">“<span class="quote">C language</span>”</span> functions

  8.     from <span class="quote">“<span class="quote">internal</span>”</span> functions — the actual coding conventions

  9.     are essentially the same for both.  (Hence, the standard internal

  10.     function library is a rich source of coding examples for user-defined

  11.     C functions.)

  12.    </p><p>

  13.     Currently only one calling convention is used for C functions

  14.     (<span class="quote">“<span class="quote">version 1</span>”</span>). Support for that calling convention is

  15.     indicated by writing a <code class="literal">PG_FUNCTION_INFO_V1()</code> macro

  16.     call for the function, as illustrated below.

  17.    </p><div class="sect2" id="XFUNC-C-DYNLOAD"><div class="titlepage"><div><div><h3 class="title">37.10.1. Dynamic Loading</h3></div></div></div><a id="id-1.8.3.13.5.2" class="indexterm"></a><p>

  18.     The first time a user-defined function in a particular

  19.     loadable object file is called in a session,

  20.     the dynamic loader loads that object file into memory so that the

  21.     function can be called.  The <code class="command">CREATE FUNCTION</code>

  22.     for a user-defined C function must therefore specify two pieces of

  23.     information for the function: the name of the loadable

  24.     object file, and the C name (link symbol) of the specific function to call

  25.     within that object file.  If the C name is not explicitly specified then

  26.     it is assumed to be the same as the SQL function name.

  27.    </p><p>

  28.     The following algorithm is used to locate the shared object file

  29.     based on the name given in the <code class="command">CREATE FUNCTION</code>

  30.     command:

  31. 

  32.     </p><div class="orderedlist"><ol class="orderedlist" type="1"><li class="listitem"><p>

  33.        If the name is an absolute path, the given file is loaded.

  34.       </p></li><li class="listitem"><p>

  35.        If the name starts with the string <code class="literal">$libdir</code>,

  36.        that part is replaced by the <span class="productname">PostgreSQL</span> package

  37.         library directory

  38.        name, which is determined at build time.<a id="id-1.8.3.13.5.4.2.2.1.3" class="indexterm"></a>

  39.       </p></li><li class="listitem"><p>

  40.        If the name does not contain a directory part, the file is

  41.        searched for in the path specified by the configuration variable

  42.        <a class="xref" href="runtime-config-client.html#GUC-DYNAMIC-LIBRARY-PATH">dynamic_library_path</a>.<a id="id-1.8.3.13.5.4.2.3.1.2" class="indexterm"></a>

  43.       </p></li><li class="listitem"><p>

  44.        Otherwise (the file was not found in the path, or it contains a

  45.        non-absolute directory part), the dynamic loader will try to

  46.        take the name as given, which will most likely fail.  (It is

  47.        unreliable to depend on the current working directory.)

  48.       </p></li></ol></div><p>

  49. 

  50.     If this sequence does not work, the platform-specific shared

  51.     library file name extension (often <code class="filename">.so</code>) is

  52.     appended to the given name and this sequence is tried again.  If

  53.     that fails as well, the load will fail.

  54.    </p><p>

  55.     It is recommended to locate shared libraries either relative to

  56.     <code class="literal">$libdir</code> or through the dynamic library path.

  57.     This simplifies version upgrades if the new installation is at a

  58.     different location.  The actual directory that

  59.     <code class="literal">$libdir</code> stands for can be found out with the

  60.     command <code class="literal">pg_config --pkglibdir</code>.

  61.    </p><p>

  62.     The user ID the <span class="productname">PostgreSQL</span> server runs

  63.     as must be able to traverse the path to the file you intend to

  64.     load.  Making the file or a higher-level directory not readable

  65.     and/or not executable by the <span class="systemitem">postgres</span>

  66.     user is a common mistake.

  67.    </p><p>

  68.     In any case, the file name that is given in the

  69.     <code class="command">CREATE FUNCTION</code> command is recorded literally

  70.     in the system catalogs, so if the file needs to be loaded again

  71.     the same procedure is applied.

  72.    </p><div class="note"><h3 class="title">Note</h3><p>

  73.      <span class="productname">PostgreSQL</span> will not compile a C function

  74.      automatically.  The object file must be compiled before it is referenced

  75.      in a <code class="command">CREATE

  76.      FUNCTION</code> command.  See <a class="xref" href="xfunc-c.html#DFUNC" title="37.10.5. Compiling and Linking Dynamically-Loaded Functions">Section 37.10.5</a> for additional

  77.      information.

  78.     </p></div><a id="id-1.8.3.13.5.9" class="indexterm"></a><p>

  79.     To ensure that a dynamically loaded object file is not loaded into an

  80.     incompatible server, <span class="productname">PostgreSQL</span> checks that the

  81.     file contains a <span class="quote">“<span class="quote">magic block</span>”</span> with the appropriate contents.

  82.     This allows the server to detect obvious incompatibilities, such as code

  83.     compiled for a different major version of

  84.     <span class="productname">PostgreSQL</span>. To include a magic block,

  85.     write this in one (and only one) of the module source files, after having

  86.     included the header <code class="filename">fmgr.h</code>:

  87. 

  88. </p><pre class="programlisting">

  89. PG_MODULE_MAGIC;

  90. </pre><p>

  91.    </p><p>

  92.     After it is used for the first time, a dynamically loaded object

  93.     file is retained in memory.  Future calls in the same session to

  94.     the function(s) in that file will only incur the small overhead of

  95.     a symbol table lookup.  If you need to force a reload of an object

  96.     file, for example after recompiling it, begin a fresh session.

  97.    </p><a id="id-1.8.3.13.5.12" class="indexterm"></a><a id="id-1.8.3.13.5.13" class="indexterm"></a><a id="id-1.8.3.13.5.14" class="indexterm"></a><a id="id-1.8.3.13.5.15" class="indexterm"></a><p>

  98.     Optionally, a dynamically loaded file can contain initialization and

  99.     finalization functions.  If the file includes a function named

  100.     <code class="function">_PG_init</code>, that function will be called immediately after

  101.     loading the file.  The function receives no parameters and should

  102.     return void.  If the file includes a function named

  103.     <code class="function">_PG_fini</code>, that function will be called immediately before

  104.     unloading the file.  Likewise, the function receives no parameters and

  105.     should return void.  Note that <code class="function">_PG_fini</code> will only be called

  106.     during an unload of the file, not during process termination.

  107.     (Presently, unloads are disabled and will never occur, but this may

  108.     change in the future.)

  109.    </p></div><div class="sect2" id="XFUNC-C-BASETYPE"><div class="titlepage"><div><div><h3 class="title">37.10.2. Base Types in C-Language Functions</h3></div></div></div><a id="id-1.8.3.13.6.2" class="indexterm"></a><p>

  110.      To know how to write C-language functions, you need to know how

  111.      <span class="productname">PostgreSQL</span> internally represents base

  112.      data types and how they can be passed to and from functions.

  113.      Internally, <span class="productname">PostgreSQL</span> regards a base

  114.      type as a <span class="quote">“<span class="quote">blob of memory</span>”</span>.  The user-defined

  115.      functions that you define over a type in turn define the way that

  116.      <span class="productname">PostgreSQL</span> can operate on it.  That

  117.      is, <span class="productname">PostgreSQL</span> will only store and

  118.      retrieve the data from disk and use your user-defined functions

  119.      to input, process, and output the data.

  120.     </p><p>

  121.      Base types can have one of three internal formats:

  122. 

  123.      </p><div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; "><li class="listitem"><p>

  124.         pass by value, fixed-length

  125.        </p></li><li class="listitem"><p>

  126.         pass by reference, fixed-length

  127.        </p></li><li class="listitem"><p>

  128.         pass by reference, variable-length

  129.        </p></li></ul></div><p>

  130.     </p><p>

  131.      By-value  types  can  only be 1, 2, or 4 bytes in length

  132.      (also 8 bytes, if <code class="literal">sizeof(Datum)</code> is 8 on your machine).

  133.      You should be careful to define your types such that they will be the

  134.      same size (in bytes) on all architectures.  For example, the

  135.      <code class="literal">long</code> type is dangerous because it is 4 bytes on some

  136.      machines and 8 bytes on others, whereas <code class="type">int</code> type is 4 bytes

  137.      on most Unix machines.  A reasonable implementation of the

  138.      <code class="type">int4</code> type on Unix machines might be:

  139. 

  140. </p><pre class="programlisting">

  141. /* 4-byte integer, passed by value */

  142. typedef int int4;

  143. </pre><p>

  144. 

  145.      (The actual PostgreSQL C code calls this type <code class="type">int32</code>, because

  146.      it is a convention in C that <code class="type">int<em class="replaceable"><code>XX</code></em></code>

  147.      means <em class="replaceable"><code>XX</code></em> <span class="emphasis"><em>bits</em></span>.  Note

  148.      therefore also that the C type <code class="type">int8</code> is 1 byte in size.  The

  149.      SQL type <code class="type">int8</code> is called <code class="type">int64</code> in C.  See also

  150.      <a class="xref" href="xfunc-c.html#XFUNC-C-TYPE-TABLE" title="Table 37.1. Equivalent C Types for Built-in SQL Types">Table 37.1</a>.)

  151.     </p><p>

  152.      On  the  other hand, fixed-length types of any size can

  153.      be passed by-reference.  For example, here is a  sample

  154.      implementation of a <span class="productname">PostgreSQL</span> type:

  155. 

  156. </p><pre class="programlisting">

  157. /* 16-byte structure, passed by reference */

  158. typedef struct

  159. {

  160.     double  x, y;

  161. } Point;

  162. </pre><p>

  163. 

  164.      Only  pointers  to  such types can be used when passing

  165.      them in and out of <span class="productname">PostgreSQL</span> functions.

  166.      To return a value of such a type, allocate the right amount of

  167.      memory with <code class="literal">palloc</code>, fill in the allocated memory,

  168.      and return a pointer to it.  (Also, if you just want to return the

  169.      same value as one of your input arguments that's of the same data type,

  170.      you can skip the extra <code class="literal">palloc</code> and just return the

  171.      pointer to the input value.)

  172.     </p><p>

  173.      Finally, all variable-length types must also be  passed

  174.      by  reference.   All  variable-length  types must begin

  175.      with an opaque length field of exactly 4 bytes, which will be set

  176.      by <code class="symbol">SET_VARSIZE</code>; never set this field directly! All data to

  177.      be  stored within that type must be located in the memory

  178.      immediately  following  that  length  field.   The

  179.      length field contains the total length of the structure,

  180.      that is,  it  includes  the  size  of  the  length  field

  181.      itself.

  182.     </p><p>

  183.      Another important point is to avoid leaving any uninitialized bits

  184.      within data type values; for example, take care to zero out any

  185.      alignment padding bytes that might be present in structs.  Without

  186.      this, logically-equivalent constants of your data type might be

  187.      seen as unequal by the planner, leading to inefficient (though not

  188.      incorrect) plans.

  189.     </p><div class="warning"><h3 class="title">Warning</h3><p>

  190.       <span class="emphasis"><em>Never</em></span> modify the contents of a pass-by-reference input

  191.       value.  If you do so you are likely to corrupt on-disk data, since

  192.       the pointer you are given might point directly into a disk buffer.

  193.       The sole exception to this rule is explained in

  194.       <a class="xref" href="xaggr.html" title="37.12. User-Defined Aggregates">Section 37.12</a>.

  195.      </p></div><p>

  196.      As an example, we can define the type <code class="type">text</code> as

  197.      follows:

  198. 

  199. </p><pre class="programlisting">

  200. typedef struct {

  201.     int32 length;

  202.     char data[FLEXIBLE_ARRAY_MEMBER];

  203. } text;

  204. </pre><p>

  205. 

  206.      The <code class="literal">[FLEXIBLE_ARRAY_MEMBER]</code> notation means that the actual

  207.      length of the data part is not specified by this declaration.

  208.     </p><p>

  209.      When manipulating

  210.      variable-length types, we must  be  careful  to  allocate

  211.      the  correct amount  of memory and set the length field correctly.

  212.      For example, if we wanted to  store  40  bytes  in  a <code class="structname">text</code>

  213.      structure, we might use a code fragment like this:

  214. 

  215. </p><pre class="programlisting">

  216. #include "postgres.h"

  217. ...

  218. char buffer[40]; /* our source data */

  219. ...

  220. text *destination = (text *) palloc(VARHDRSZ + 40);

  221. SET_VARSIZE(destination, VARHDRSZ + 40);

  222. memcpy(destination-&gt;data, buffer, 40);

  223. ...

  224. 

  225. </pre><p>

  226. 

  227.      <code class="literal">VARHDRSZ</code> is the same as <code class="literal">sizeof(int32)</code>, but

  228.      it's considered good style to use the macro <code class="literal">VARHDRSZ</code>

  229.      to refer to the size of the overhead for a variable-length type.

  230.      Also, the length field <span class="emphasis"><em>must</em></span> be set using the

  231.      <code class="literal">SET_VARSIZE</code> macro, not by simple assignment.

  232.     </p><p>

  233.      <a class="xref" href="xfunc-c.html#XFUNC-C-TYPE-TABLE" title="Table 37.1. Equivalent C Types for Built-in SQL Types">Table 37.1</a> specifies which C type

  234.      corresponds to which SQL type when writing a C-language function

  235.      that uses a built-in type of <span class="productname">PostgreSQL</span>.

  236.      The <span class="quote">“<span class="quote">Defined In</span>”</span> column gives the header file that

  237.      needs to be included to get the type definition.  (The actual

  238.      definition might be in a different file that is included by the

  239.      listed file.  It is recommended that users stick to the defined

  240.      interface.)  Note that you should always include

  241.      <code class="filename">postgres.h</code> first in any source file, because

  242.      it declares a number of things that you will need anyway.

  243.     </p><div class="table" id="XFUNC-C-TYPE-TABLE"><p class="title"><strong>Table 37.1. Equivalent C Types for Built-in SQL Types</strong></p><div class="table-contents"><table class="table" summary="Equivalent C Types for Built-in SQL Types" border="1"><colgroup><col /><col /><col /></colgroup><thead><tr><th>

  244.           SQL Type

  245.          </th><th>

  246.           C Type

  247.          </th><th>

  248.           Defined In

  249.          </th></tr></thead><tbody><tr><td><code class="type">boolean</code></td><td><code class="type">bool</code></td><td><code class="filename">postgres.h</code> (maybe compiler built-in)</td></tr><tr><td><code class="type">box</code></td><td><code class="type">BOX*</code></td><td><code class="filename">utils/geo_decls.h</code></td></tr><tr><td><code class="type">bytea</code></td><td><code class="type">bytea*</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">"char"</code></td><td><code class="type">char</code></td><td>(compiler built-in)</td></tr><tr><td><code class="type">character</code></td><td><code class="type">BpChar*</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">cid</code></td><td><code class="type">CommandId</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">date</code></td><td><code class="type">DateADT</code></td><td><code class="filename">utils/date.h</code></td></tr><tr><td><code class="type">smallint</code> (<code class="type">int2</code>)</td><td><code class="type">int16</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">int2vector</code></td><td><code class="type">int2vector*</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">integer</code> (<code class="type">int4</code>)</td><td><code class="type">int32</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">real</code> (<code class="type">float4</code>)</td><td><code class="type">float4*</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">double precision</code> (<code class="type">float8</code>)</td><td><code class="type">float8*</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">interval</code></td><td><code class="type">Interval*</code></td><td><code class="filename">datatype/timestamp.h</code></td></tr><tr><td><code class="type">lseg</code></td><td><code class="type">LSEG*</code></td><td><code class="filename">utils/geo_decls.h</code></td></tr><tr><td><code class="type">name</code></td><td><code class="type">Name</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">oid</code></td><td><code class="type">Oid</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">oidvector</code></td><td><code class="type">oidvector*</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">path</code></td><td><code class="type">PATH*</code></td><td><code class="filename">utils/geo_decls.h</code></td></tr><tr><td><code class="type">point</code></td><td><code class="type">POINT*</code></td><td><code class="filename">utils/geo_decls.h</code></td></tr><tr><td><code class="type">regproc</code></td><td><code class="type">regproc</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">text</code></td><td><code class="type">text*</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">tid</code></td><td><code class="type">ItemPointer</code></td><td><code class="filename">storage/itemptr.h</code></td></tr><tr><td><code class="type">time</code></td><td><code class="type">TimeADT</code></td><td><code class="filename">utils/date.h</code></td></tr><tr><td><code class="type">time with time zone</code></td><td><code class="type">TimeTzADT</code></td><td><code class="filename">utils/date.h</code></td></tr><tr><td><code class="type">timestamp</code></td><td><code class="type">Timestamp*</code></td><td><code class="filename">datatype/timestamp.h</code></td></tr><tr><td><code class="type">varchar</code></td><td><code class="type">VarChar*</code></td><td><code class="filename">postgres.h</code></td></tr><tr><td><code class="type">xid</code></td><td><code class="type">TransactionId</code></td><td><code class="filename">postgres.h</code></td></tr></tbody></table></div></div><br class="table-break" /><p>

  250.      Now that we've gone over all of the possible structures

  251.      for base types, we can show some examples of real functions.

  252.     </p></div><div class="sect2" id="id-1.8.3.13.7"><div class="titlepage"><div><div><h3 class="title">37.10.3. Version 1 Calling Conventions</h3></div></div></div><p>

  253.      The version-1 calling convention relies on macros to suppress most

  254.      of the complexity of passing arguments and results.  The C declaration

  255.      of a version-1 function is always:

  256. </p><pre class="programlisting">

  257. Datum funcname(PG_FUNCTION_ARGS)

  258. </pre><p>

  259.      In addition, the macro call:

  260. </p><pre class="programlisting">

  261. PG_FUNCTION_INFO_V1(funcname);

  262. </pre><p>

  263.      must appear in the same source file.  (Conventionally, it's

  264.      written just before the function itself.)  This macro call is not

  265.      needed for <code class="literal">internal</code>-language functions, since

  266.      <span class="productname">PostgreSQL</span> assumes that all internal functions

  267.      use the version-1 convention.  It is, however, required for

  268.      dynamically-loaded functions.

  269.     </p><p>

  270.      In a version-1 function, each actual argument is fetched using a

  271.      <code class="function">PG_GETARG_<em class="replaceable"><code>xxx</code></em>()</code>

  272.      macro that corresponds to the argument's data type.  (In non-strict

  273.      functions there needs to be a previous check about argument null-ness

  274.      using <code class="function">PG_ARGISNULL()</code>; see below.)

  275.      The result is returned using a

  276.      <code class="function">PG_RETURN_<em class="replaceable"><code>xxx</code></em>()</code>

  277.      macro for the return type.

  278.      <code class="function">PG_GETARG_<em class="replaceable"><code>xxx</code></em>()</code>

  279.      takes as its argument the number of the function argument to

  280.      fetch, where the count starts at 0.

  281.      <code class="function">PG_RETURN_<em class="replaceable"><code>xxx</code></em>()</code>

  282.      takes as its argument the actual value to return.

  283.     </p><p>

  284.      Here are some examples using the version-1 calling convention:

  285.     </p><pre class="programlisting">

  286. #include "postgres.h"

  287. #include &lt;string.h&gt;

  288. #include "fmgr.h"

  289. #include "utils/geo_decls.h"

  290. 

  291. PG_MODULE_MAGIC;

  292. 

  293. /* by value */

  294. 

  295. PG_FUNCTION_INFO_V1(add_one);

  296. 

  297. Datum

  298. add_one(PG_FUNCTION_ARGS)

  299. {

  300.     int32   arg = PG_GETARG_INT32(0);

  301. 

  302.     PG_RETURN_INT32(arg + 1);

  303. }

  304. 

  305. /* by reference, fixed length */

  306. 

  307. PG_FUNCTION_INFO_V1(add_one_float8);

  308. 

  309. Datum

  310. add_one_float8(PG_FUNCTION_ARGS)

  311. {

  312.     /* The macros for FLOAT8 hide its pass-by-reference nature. */

  313.     float8   arg = PG_GETARG_FLOAT8(0);

  314. 

  315.     PG_RETURN_FLOAT8(arg + 1.0);

  316. }

  317. 

  318. PG_FUNCTION_INFO_V1(makepoint);

  319. 

  320. Datum

  321. makepoint(PG_FUNCTION_ARGS)

  322. {

  323.     /* Here, the pass-by-reference nature of Point is not hidden. */

  324.     Point     *pointx = PG_GETARG_POINT_P(0);

  325.     Point     *pointy = PG_GETARG_POINT_P(1);

  326.     Point     *new_point = (Point *) palloc(sizeof(Point));

  327. 

  328.     new_point-&gt;x = pointx-&gt;x;

  329.     new_point-&gt;y = pointy-&gt;y;

  330. 

  331.     PG_RETURN_POINT_P(new_point);

  332. }

  333. 

  334. /* by reference, variable length */

  335. 

  336. PG_FUNCTION_INFO_V1(copytext);

  337. 

  338. Datum

  339. copytext(PG_FUNCTION_ARGS)

  340. {

  341.     text     *t = PG_GETARG_TEXT_PP(0);

  342. 

  343.     /*

  344.      * VARSIZE_ANY_EXHDR is the size of the struct in bytes, minus the

  345.      * VARHDRSZ or VARHDRSZ_SHORT of its header.  Construct the copy with a

  346.      * full-length header.

  347.      */

  348.     text     *new_t = (text *) palloc(VARSIZE_ANY_EXHDR(t) + VARHDRSZ);

  349.     SET_VARSIZE(new_t, VARSIZE_ANY_EXHDR(t) + VARHDRSZ);

  350. 

  351.     /*

  352.      * VARDATA is a pointer to the data region of the new struct.  The source

  353.      * could be a short datum, so retrieve its data through VARDATA_ANY.

  354.      */

  355.     memcpy((void *) VARDATA(new_t), /* destination */

  356.            (void *) VARDATA_ANY(t), /* source */

  357.            VARSIZE_ANY_EXHDR(t));   /* how many bytes */

  358.     PG_RETURN_TEXT_P(new_t);

  359. }

  360. 

  361. PG_FUNCTION_INFO_V1(concat_text);

  362. 

  363. Datum

  364. concat_text(PG_FUNCTION_ARGS)

  365. {

  366.     text  *arg1 = PG_GETARG_TEXT_PP(0);

  367.     text  *arg2 = PG_GETARG_TEXT_PP(1);

  368.     int32 arg1_size = VARSIZE_ANY_EXHDR(arg1);

  369.     int32 arg2_size = VARSIZE_ANY_EXHDR(arg2);

  370.     int32 new_text_size = arg1_size + arg2_size + VARHDRSZ;

  371.     text *new_text = (text *) palloc(new_text_size);

  372. 

  373.     SET_VARSIZE(new_text, new_text_size);

  374.     memcpy(VARDATA(new_text), VARDATA_ANY(arg1), arg1_size);

  375.     memcpy(VARDATA(new_text) + arg1_size, VARDATA_ANY(arg2), arg2_size);

  376.     PG_RETURN_TEXT_P(new_text);

  377. }

  378. 

  379. </pre><p>

  380.      Supposing that the above code has been prepared in file

  381.      <code class="filename">funcs.c</code> and compiled into a shared object,

  382.      we could define the functions to <span class="productname">PostgreSQL</span>

  383.      with commands like this:

  384.     </p><pre class="programlisting">

  385. CREATE FUNCTION add_one(integer) RETURNS integer

  386.      AS '<em class="replaceable"><code>DIRECTORY</code></em>/funcs', 'add_one'

  387.      LANGUAGE C STRICT;

  388. 

  389. -- note overloading of SQL function name "add_one"

  390. CREATE FUNCTION add_one(double precision) RETURNS double precision

  391.      AS '<em class="replaceable"><code>DIRECTORY</code></em>/funcs', 'add_one_float8'

  392.      LANGUAGE C STRICT;

  393. 

  394. CREATE FUNCTION makepoint(point, point) RETURNS point

  395.      AS '<em class="replaceable"><code>DIRECTORY</code></em>/funcs', 'makepoint'

  396.      LANGUAGE C STRICT;

  397. 

  398. CREATE FUNCTION copytext(text) RETURNS text

  399.      AS '<em class="replaceable"><code>DIRECTORY</code></em>/funcs', 'copytext'

  400.      LANGUAGE C STRICT;

  401. 

  402. CREATE FUNCTION concat_text(text, text) RETURNS text

  403.      AS '<em class="replaceable"><code>DIRECTORY</code></em>/funcs', 'concat_text'

  404.      LANGUAGE C STRICT;

  405. </pre><p>

  406.      Here, <em class="replaceable"><code>DIRECTORY</code></em> stands for the

  407.      directory of the shared library file (for instance the

  408.      <span class="productname">PostgreSQL</span> tutorial directory, which

  409.      contains the code for the examples used in this section).

  410.      (Better style would be to use just <code class="literal">'funcs'</code> in the

  411.      <code class="literal">AS</code> clause, after having added

  412.      <em class="replaceable"><code>DIRECTORY</code></em> to the search path.  In any

  413.      case, we can omit the system-specific extension for a shared

  414.      library, commonly <code class="literal">.so</code>.)

  415.     </p><p>

  416.      Notice that we have specified the functions as <span class="quote">“<span class="quote">strict</span>”</span>,

  417.      meaning that

  418.      the system should automatically assume a null result if any input

  419.      value is null.  By doing this, we avoid having to check for null inputs

  420.      in the function code.  Without this, we'd have to check for null values

  421.      explicitly, using <code class="function">PG_ARGISNULL()</code>.

  422.     </p><p>

  423.      The macro <code class="function">PG_ARGISNULL(<em class="replaceable"><code>n</code></em>)</code>

  424.      allows a function to test whether each input is null.  (Of course, doing

  425.      this is only necessary in functions not declared <span class="quote">“<span class="quote">strict</span>”</span>.)

  426.      As with the

  427.      <code class="function">PG_GETARG_<em class="replaceable"><code>xxx</code></em>()</code> macros,

  428.      the input arguments are counted beginning at zero.  Note that one

  429.      should refrain from executing

  430.      <code class="function">PG_GETARG_<em class="replaceable"><code>xxx</code></em>()</code> until

  431.      one has verified that the argument isn't null.

  432.      To return a null result, execute <code class="function">PG_RETURN_NULL()</code>;

  433.      this works in both strict and nonstrict functions.

  434.     </p><p>

  435.      At first glance, the version-1 coding conventions might appear

  436.      to be just pointless obscurantism, compared to using

  437.      plain <code class="literal">C</code> calling conventions.  They do however allow

  438.      us to deal with <code class="literal">NULL</code>able arguments/return values,

  439.      and <span class="quote">“<span class="quote">toasted</span>”</span> (compressed or out-of-line) values.

  440.     </p><p>

  441.      Other options provided by the version-1 interface are two

  442.      variants of the

  443.      <code class="function">PG_GETARG_<em class="replaceable"><code>xxx</code></em>()</code>

  444.      macros. The first of these,

  445.      <code class="function">PG_GETARG_<em class="replaceable"><code>xxx</code></em>_COPY()</code>,

  446.      guarantees to return a copy of the specified argument that is

  447.      safe for writing into. (The normal macros will sometimes return a

  448.      pointer to a value that is physically stored in a table, which

  449.      must not be written to. Using the

  450.      <code class="function">PG_GETARG_<em class="replaceable"><code>xxx</code></em>_COPY()</code>

  451.      macros guarantees a writable result.)

  452.     The second variant consists of the

  453.     <code class="function">PG_GETARG_<em class="replaceable"><code>xxx</code></em>_SLICE()</code>

  454.     macros which take three arguments. The first is the number of the

  455.     function argument (as above). The second and third are the offset and

  456.     length of the segment to be returned. Offsets are counted from

  457.     zero, and a negative length requests that the remainder of the

  458.     value be returned. These macros provide more efficient access to

  459.     parts of large values in the case where they have storage type

  460.     <span class="quote">“<span class="quote">external</span>”</span>. (The storage type of a column can be specified using

  461.     <code class="literal">ALTER TABLE <em class="replaceable"><code>tablename</code></em> ALTER

  462.     COLUMN <em class="replaceable"><code>colname</code></em> SET STORAGE

  463.     <em class="replaceable"><code>storagetype</code></em></code>. <em class="replaceable"><code>storagetype</code></em> is one of

  464.     <code class="literal">plain</code>, <code class="literal">external</code>, <code class="literal">extended</code>,

  465.      or <code class="literal">main</code>.)

  466.     </p><p>

  467.      Finally, the version-1 function call conventions make it possible

  468.      to return set results (<a class="xref" href="xfunc-c.html#XFUNC-C-RETURN-SET" title="37.10.8. Returning Sets">Section 37.10.8</a>) and

  469.      implement trigger functions (<a class="xref" href="triggers.html" title="Chapter 38. Triggers">Chapter 38</a>) and

  470.      procedural-language call handlers (<a class="xref" href="plhandler.html" title="Chapter 55. Writing a Procedural Language Handler">Chapter 55</a>).  For more details

  471.      see <code class="filename">src/backend/utils/fmgr/README</code> in the

  472.      source distribution.

  473.     </p></div><div class="sect2" id="id-1.8.3.13.8"><div class="titlepage"><div><div><h3 class="title">37.10.4. Writing Code</h3></div></div></div><p>

  474.      Before we turn to the more advanced topics, we should discuss

  475.      some coding rules for <span class="productname">PostgreSQL</span>

  476.      C-language functions.  While it might be possible to load functions

  477.      written in languages other than C into

  478.      <span class="productname">PostgreSQL</span>, this is usually difficult

  479.      (when it is possible at all) because other languages, such as

  480.      C++, FORTRAN, or Pascal often do not follow the same calling

  481.      convention as C.  That is, other languages do not pass argument

  482.      and return values between functions in the same way.  For this

  483.      reason, we will assume that your C-language functions are

  484.      actually written in C.

  485.     </p><p>

  486.      The basic rules for writing and building C functions are as follows:

  487. 

  488.      </p><div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; "><li class="listitem"><p>

  489.         Use <code class="literal">pg_config

  490.         --includedir-server</code><a id="id-1.8.3.13.8.3.1.1.1.2" class="indexterm"></a>

  491.         to find out where the <span class="productname">PostgreSQL</span> server header

  492.         files are installed on your system (or the system that your

  493.         users will be running on).

  494.        </p></li><li class="listitem"><p>

  495.         Compiling and linking your code so that it can be dynamically

  496.         loaded into <span class="productname">PostgreSQL</span> always

  497.         requires special flags.  See <a class="xref" href="xfunc-c.html#DFUNC" title="37.10.5. Compiling and Linking Dynamically-Loaded Functions">Section 37.10.5</a> for a

  498.         detailed explanation of how to do it for your particular

  499.         operating system.

  500.        </p></li><li class="listitem"><p>

  501.         Remember to define a <span class="quote">“<span class="quote">magic block</span>”</span> for your shared library,

  502.         as described in <a class="xref" href="xfunc-c.html#XFUNC-C-DYNLOAD" title="37.10.1. Dynamic Loading">Section 37.10.1</a>.

  503.        </p></li><li class="listitem"><p>

  504.         When allocating memory, use the

  505.         <span class="productname">PostgreSQL</span> functions

  506.         <code class="function">palloc</code><a id="id-1.8.3.13.8.3.1.4.1.3" class="indexterm"></a> and <code class="function">pfree</code><a id="id-1.8.3.13.8.3.1.4.1.5" class="indexterm"></a>

  507.         instead of the corresponding C library functions

  508.         <code class="function">malloc</code> and <code class="function">free</code>.

  509.         The memory allocated by <code class="function">palloc</code> will be

  510.         freed automatically at the end of each transaction, preventing

  511.         memory leaks.

  512.        </p></li><li class="listitem"><p>

  513.         Always zero the bytes of your structures using <code class="function">memset</code>

  514.         (or allocate them with <code class="function">palloc0</code> in the first place).

  515.         Even if you assign to each field of your structure, there might be

  516.         alignment padding (holes in the structure) that contain

  517.         garbage values.  Without this, it's difficult to

  518.         support hash indexes or hash joins, as you must pick out only

  519.         the significant bits of your data structure to compute a hash.

  520.         The planner also sometimes relies on comparing constants via

  521.         bitwise equality, so you can get undesirable planning results if

  522.         logically-equivalent values aren't bitwise equal.

  523.        </p></li><li class="listitem"><p>

  524.         Most of the internal <span class="productname">PostgreSQL</span>

  525.         types are declared in <code class="filename">postgres.h</code>, while

  526.         the function manager interfaces

  527.         (<code class="symbol">PG_FUNCTION_ARGS</code>, etc.)  are in

  528.         <code class="filename">fmgr.h</code>, so you will need to include at

  529.         least these two files.  For portability reasons it's best to

  530.         include <code class="filename">postgres.h</code> <span class="emphasis"><em>first</em></span>,

  531.         before any other system or user header files.  Including

  532.         <code class="filename">postgres.h</code> will also include

  533.         <code class="filename">elog.h</code> and <code class="filename">palloc.h</code>

  534.         for you.

  535.        </p></li><li class="listitem"><p>

  536.         Symbol names defined within object files must not conflict

  537.         with each other or with symbols defined in the

  538.         <span class="productname">PostgreSQL</span> server executable.  You

  539.         will have to rename your functions or variables if you get

  540.         error messages to this effect.

  541.        </p></li></ul></div><p>

  542.     </p></div><div class="sect2" id="DFUNC"><div class="titlepage"><div><div><h3 class="title">37.10.5. Compiling and Linking Dynamically-Loaded Functions</h3></div></div></div><p>

  543.   Before you are able to use your

  544.   <span class="productname">PostgreSQL</span> extension functions written in

  545.   C, they must be compiled and linked in a special way to produce a

  546.   file that can be dynamically loaded by the server.  To be precise, a

  547.   <em class="firstterm">shared library</em> needs to be

  548.   created.<a id="id-1.8.3.13.9.2.3" class="indexterm"></a>

  549. 

  550.  </p><p>

  551.   For information beyond what is contained in this section

  552.   you should read the documentation of your

  553.   operating system, in particular the manual pages for the C compiler,

  554.   <code class="command">cc</code>, and the link editor, <code class="command">ld</code>.

  555.   In addition, the <span class="productname">PostgreSQL</span> source code

  556.   contains several working examples in the

  557.   <code class="filename">contrib</code> directory.  If you rely on these

  558.   examples you will make your modules dependent on the availability

  559.   of the <span class="productname">PostgreSQL</span> source code, however.

  560.  </p><p>

  561.   Creating shared libraries is generally analogous to linking

  562.   executables: first the source files are compiled into object files,

  563.   then the object files are linked together.  The object files need to

  564.   be created as <em class="firstterm">position-independent code</em>

  565.   (<acronym class="acronym">PIC</acronym>),<a id="id-1.8.3.13.9.4.3" class="indexterm"></a> which

  566.   conceptually means that they can be placed at an arbitrary location

  567.   in memory when they are loaded by the executable.  (Object files

  568.   intended for executables are usually not compiled that way.)  The

  569.   command to link a shared library contains special flags to

  570.   distinguish it from linking an executable (at least in theory

  571.   — on some systems the practice is much uglier).

  572.  </p><p>

  573.   In the following examples we assume that your source code is in a

  574.   file <code class="filename">foo.c</code> and we will create a shared library

  575.   <code class="filename">foo.so</code>.  The intermediate object file will be

  576.   called <code class="filename">foo.o</code> unless otherwise noted.  A shared

  577.   library can contain more than one object file, but we only use one

  578.   here.

  579.  </p><div class="variablelist"><dl class="variablelist"><dt><span class="term">

  580.      <span class="systemitem">FreeBSD</span>

  581.      <a id="id-1.8.3.13.9.6.1.1.2" class="indexterm"></a>

  582.     </span></dt><dd><p>

  583.       The compiler flag to create <acronym class="acronym">PIC</acronym> is

  584.       <code class="option">-fPIC</code>.  To create shared libraries the compiler

  585.       flag is <code class="option">-shared</code>.

  586. </p><pre class="programlisting">

  587. gcc -fPIC -c foo.c

  588. gcc -shared -o foo.so foo.o

  589. </pre><p>

  590.       This is applicable as of version 3.0 of

  591.       <span class="systemitem">FreeBSD</span>.

  592.      </p></dd><dt><span class="term">

  593.      <span class="systemitem">HP-UX</span>

  594.      <a id="id-1.8.3.13.9.6.2.1.2" class="indexterm"></a>

  595.     </span></dt><dd><p>

  596.       The compiler flag of the system compiler to create

  597.       <acronym class="acronym">PIC</acronym> is <code class="option">+z</code>.  When using

  598.       <span class="application">GCC</span> it's <code class="option">-fPIC</code>. The

  599.       linker flag for shared libraries is <code class="option">-b</code>.  So:

  600. </p><pre class="programlisting">

  601. cc +z -c foo.c

  602. </pre><p>

  603.       or:

  604. </p><pre class="programlisting">

  605. gcc -fPIC -c foo.c

  606. </pre><p>

  607.       and then:

  608. </p><pre class="programlisting">

  609. ld -b -o foo.sl foo.o

  610. </pre><p>

  611.       <span class="systemitem">HP-UX</span> uses the extension

  612.       <code class="filename">.sl</code> for shared libraries, unlike most other

  613.       systems.

  614.      </p></dd><dt><span class="term">

  615.      <span class="systemitem">Linux</span>

  616.      <a id="id-1.8.3.13.9.6.3.1.2" class="indexterm"></a>

  617.     </span></dt><dd><p>

  618.       The compiler flag to create <acronym class="acronym">PIC</acronym> is

  619.       <code class="option">-fPIC</code>.

  620.       The compiler flag to create a shared library is

  621.       <code class="option">-shared</code>.  A complete example looks like this:

  622. </p><pre class="programlisting">

  623. cc -fPIC -c foo.c

  624. cc -shared -o foo.so foo.o

  625. </pre><p>

  626.      </p></dd><dt><span class="term">

  627.      <span class="systemitem">macOS</span>

  628.      <a id="id-1.8.3.13.9.6.4.1.2" class="indexterm"></a>

  629.     </span></dt><dd><p>

  630.       Here is an example.  It assumes the developer tools are installed.

  631. </p><pre class="programlisting">

  632. cc -c foo.c

  633. cc -bundle -flat_namespace -undefined suppress -o foo.so foo.o

  634. </pre><p>

  635.      </p></dd><dt><span class="term">

  636.      <span class="systemitem">NetBSD</span>

  637.      <a id="id-1.8.3.13.9.6.5.1.2" class="indexterm"></a>

  638.     </span></dt><dd><p>

  639.       The compiler flag to create <acronym class="acronym">PIC</acronym> is

  640.       <code class="option">-fPIC</code>.  For <acronym class="acronym">ELF</acronym> systems, the

  641.       compiler with the flag <code class="option">-shared</code> is used to link

  642.       shared libraries.  On the older non-ELF systems, <code class="literal">ld

  643.       -Bshareable</code> is used.

  644. </p><pre class="programlisting">

  645. gcc -fPIC -c foo.c

  646. gcc -shared -o foo.so foo.o

  647. </pre><p>

  648.      </p></dd><dt><span class="term">

  649.      <span class="systemitem">OpenBSD</span>

  650.      <a id="id-1.8.3.13.9.6.6.1.2" class="indexterm"></a>

  651.     </span></dt><dd><p>

  652.       The compiler flag to create <acronym class="acronym">PIC</acronym> is

  653.       <code class="option">-fPIC</code>.  <code class="literal">ld -Bshareable</code> is

  654.       used to link shared libraries.

  655. </p><pre class="programlisting">

  656. gcc -fPIC -c foo.c

  657. ld -Bshareable -o foo.so foo.o

  658. </pre><p>

  659.      </p></dd><dt><span class="term">

  660.      <span class="systemitem">Solaris</span>

  661.      <a id="id-1.8.3.13.9.6.7.1.2" class="indexterm"></a>

  662.     </span></dt><dd><p>

  663.       The compiler flag to create <acronym class="acronym">PIC</acronym> is

  664.       <code class="option">-KPIC</code> with the Sun compiler and

  665.       <code class="option">-fPIC</code> with <span class="application">GCC</span>.  To

  666.       link shared libraries, the compiler option is

  667.       <code class="option">-G</code> with either compiler or alternatively

  668.       <code class="option">-shared</code> with <span class="application">GCC</span>.

  669. </p><pre class="programlisting">

  670. cc -KPIC -c foo.c

  671. cc -G -o foo.so foo.o

  672. </pre><p>

  673.       or

  674. </p><pre class="programlisting">

  675. gcc -fPIC -c foo.c

  676. gcc -G -o foo.so foo.o

  677. </pre><p>

  678.      </p></dd></dl></div><div class="tip"><h3 class="title">Tip</h3><p>

  679.    If this is too complicated for you, you should consider using

  680.    <a class="ulink" href="https://www.gnu.org/software/libtool/" target="_top">

  681.    <span class="productname">GNU Libtool</span></a>,

  682.    which hides the platform differences behind a uniform interface.

  683.   </p></div><p>

  684.   The resulting shared library file can then be loaded into

  685.   <span class="productname">PostgreSQL</span>.  When specifying the file name

  686.   to the <code class="command">CREATE FUNCTION</code> command, one must give it

  687.   the name of the shared library file, not the intermediate object file.

  688.   Note that the system's standard shared-library extension (usually

  689.   <code class="literal">.so</code> or <code class="literal">.sl</code>) can be omitted from

  690.   the <code class="command">CREATE FUNCTION</code> command, and normally should

  691.   be omitted for best portability.

  692.  </p><p>

  693.   Refer back to <a class="xref" href="xfunc-c.html#XFUNC-C-DYNLOAD" title="37.10.1. Dynamic Loading">Section 37.10.1</a> about where the

  694.   server expects to find the shared library files.

  695.  </p></div><div class="sect2" id="id-1.8.3.13.10"><div class="titlepage"><div><div><h3 class="title">37.10.6. Composite-Type Arguments</h3></div></div></div><p>

  696.      Composite types do not have a fixed layout like C structures.

  697.      Instances of a composite type can contain null fields.  In

  698.      addition, composite types that are part of an inheritance

  699.      hierarchy can have different fields than other members of the

  700.      same inheritance hierarchy.  Therefore,

  701.      <span class="productname">PostgreSQL</span> provides a function

  702.      interface for accessing fields of composite types from C.

  703.     </p><p>

  704.      Suppose we want to write a function to answer the query:

  705. 

  706. </p><pre class="programlisting">

  707. SELECT name, c_overpaid(emp, 1500) AS overpaid

  708.     FROM emp

  709.     WHERE name = 'Bill' OR name = 'Sam';

  710. </pre><p>

  711. 

  712.      Using the version-1 calling conventions, we can define

  713.      <code class="function">c_overpaid</code> as:

  714. 

  715. </p><pre class="programlisting">

  716. #include "postgres.h"

  717. #include "executor/executor.h"  /* for GetAttributeByName() */

  718. 

  719. PG_MODULE_MAGIC;

  720. 

  721. PG_FUNCTION_INFO_V1(c_overpaid);

  722. 

  723. Datum

  724. c_overpaid(PG_FUNCTION_ARGS)

  725. {

  726.     HeapTupleHeader  t = PG_GETARG_HEAPTUPLEHEADER(0);

  727.     int32            limit = PG_GETARG_INT32(1);

  728.     bool isnull;

  729.     Datum salary;

  730. 

  731.     salary = GetAttributeByName(t, "salary", &amp;isnull);

  732.     if (isnull)

  733.         PG_RETURN_BOOL(false);

  734.     /* Alternatively, we might prefer to do PG_RETURN_NULL() for null salary. */

  735. 

  736.     PG_RETURN_BOOL(DatumGetInt32(salary) &gt; limit);

  737. }

  738. 

  739. </pre><p>

  740.     </p><p>

  741.      <code class="function">GetAttributeByName</code> is the

  742.      <span class="productname">PostgreSQL</span> system function that

  743.      returns attributes out of the specified row.  It has

  744.      three arguments: the argument of type <code class="type">HeapTupleHeader</code> passed

  745.      into

  746.      the  function, the name of the desired attribute, and a

  747.      return parameter that tells whether  the  attribute

  748.      is  null.   <code class="function">GetAttributeByName</code> returns a <code class="type">Datum</code>

  749.      value that you can convert to the proper data type by using the

  750.      appropriate <code class="function">DatumGet<em class="replaceable"><code>XXX</code></em>()</code>

  751.      macro.  Note that the return value is meaningless if the null flag is

  752.      set; always check the null flag before trying to do anything with the

  753.      result.

  754.     </p><p>

  755.      There is also <code class="function">GetAttributeByNum</code>, which selects

  756.      the target attribute by column number instead of name.

  757.     </p><p>

  758.      The following command declares the function

  759.      <code class="function">c_overpaid</code> in SQL:

  760. 

  761. </p><pre class="programlisting">

  762. CREATE FUNCTION c_overpaid(emp, integer) RETURNS boolean

  763.     AS '<em class="replaceable"><code>DIRECTORY</code></em>/funcs', 'c_overpaid'

  764.     LANGUAGE C STRICT;

  765. </pre><p>

  766. 

  767.      Notice we have used <code class="literal">STRICT</code> so that we did not have to

  768.      check whether the input arguments were NULL.

  769.     </p></div><div class="sect2" id="id-1.8.3.13.11"><div class="titlepage"><div><div><h3 class="title">37.10.7. Returning Rows (Composite Types)</h3></div></div></div><p>

  770.      To return a row or composite-type value from a C-language

  771.      function, you can use a special API that provides macros and

  772.      functions to hide most of the complexity of building composite

  773.      data types.  To use this API, the source file must include:

  774. </p><pre class="programlisting">

  775. #include "funcapi.h"

  776. </pre><p>

  777.     </p><p>

  778.      There are two ways you can build a composite data value (henceforth

  779.      a <span class="quote">“<span class="quote">tuple</span>”</span>): you can build it from an array of Datum values,

  780.      or from an array of C strings that can be passed to the input

  781.      conversion functions of the tuple's column data types.  In either

  782.      case, you first need to obtain or construct a <code class="structname">TupleDesc</code>

  783.      descriptor for the tuple structure.  When working with Datums, you

  784.      pass the <code class="structname">TupleDesc</code> to <code class="function">BlessTupleDesc</code>,

  785.      and then call <code class="function">heap_form_tuple</code> for each row.  When working

  786.      with C strings, you pass the <code class="structname">TupleDesc</code> to

  787.      <code class="function">TupleDescGetAttInMetadata</code>, and then call

  788.      <code class="function">BuildTupleFromCStrings</code> for each row.  In the case of a

  789.      function returning a set of tuples, the setup steps can all be done

  790.      once during the first call of the function.

  791.     </p><p>

  792.      Several helper functions are available for setting up the needed

  793.      <code class="structname">TupleDesc</code>.  The recommended way to do this in most

  794.      functions returning composite values is to call:

  795. </p><pre class="programlisting">

  796. TypeFuncClass get_call_result_type(FunctionCallInfo fcinfo,

  797.                                    Oid *resultTypeId,

  798.                                    TupleDesc *resultTupleDesc)

  799. </pre><p>

  800.      passing the same <code class="literal">fcinfo</code> struct passed to the calling function

  801.      itself.  (This of course requires that you use the version-1

  802.      calling conventions.)  <code class="varname">resultTypeId</code> can be specified

  803.      as <code class="literal">NULL</code> or as the address of a local variable to receive the

  804.      function's result type OID.  <code class="varname">resultTupleDesc</code> should be the

  805.      address of a local <code class="structname">TupleDesc</code> variable.  Check that the

  806.      result is <code class="literal">TYPEFUNC_COMPOSITE</code>; if so,

  807.      <code class="varname">resultTupleDesc</code> has been filled with the needed

  808.      <code class="structname">TupleDesc</code>.  (If it is not, you can report an error along

  809.      the lines of <span class="quote">“<span class="quote">function returning record called in context that

  810.      cannot accept type record</span>”</span>.)

  811.     </p><div class="tip"><h3 class="title">Tip</h3><p>

  812.       <code class="function">get_call_result_type</code> can resolve the actual type of a

  813.       polymorphic function result; so it is useful in functions that return

  814.       scalar polymorphic results, not only functions that return composites.

  815.       The <code class="varname">resultTypeId</code> output is primarily useful for functions

  816.       returning polymorphic scalars.

  817.      </p></div><div class="note"><h3 class="title">Note</h3><p>

  818.       <code class="function">get_call_result_type</code> has a sibling

  819.       <code class="function">get_expr_result_type</code>, which can be used to resolve the

  820.       expected output type for a function call represented by an expression

  821.       tree.  This can be used when trying to determine the result type from

  822.       outside the function itself.  There is also

  823.       <code class="function">get_func_result_type</code>, which can be used when only the

  824.       function's OID is available.  However these functions are not able

  825.       to deal with functions declared to return <code class="structname">record</code>, and

  826.       <code class="function">get_func_result_type</code> cannot resolve polymorphic types,

  827.       so you should preferentially use <code class="function">get_call_result_type</code>.

  828.      </p></div><p>

  829.      Older, now-deprecated functions for obtaining

  830.      <code class="structname">TupleDesc</code>s are:

  831. </p><pre class="programlisting">

  832. TupleDesc RelationNameGetTupleDesc(const char *relname)

  833. </pre><p>

  834.      to get a <code class="structname">TupleDesc</code> for the row type of a named relation,

  835.      and:

  836. </p><pre class="programlisting">

  837. TupleDesc TypeGetTupleDesc(Oid typeoid, List *colaliases)

  838. </pre><p>

  839.      to get a <code class="structname">TupleDesc</code> based on a type OID. This can

  840.      be used to get a <code class="structname">TupleDesc</code> for a base or

  841.      composite type.  It will not work for a function that returns

  842.      <code class="structname">record</code>, however, and it cannot resolve polymorphic

  843.      types.

  844.     </p><p>

  845.      Once you have a <code class="structname">TupleDesc</code>, call:

  846. </p><pre class="programlisting">

  847. TupleDesc BlessTupleDesc(TupleDesc tupdesc)

  848. </pre><p>

  849.      if you plan to work with Datums, or:

  850. </p><pre class="programlisting">

  851. AttInMetadata *TupleDescGetAttInMetadata(TupleDesc tupdesc)

  852. </pre><p>

  853.      if you plan to work with C strings.  If you are writing a function

  854.      returning set, you can save the results of these functions in the

  855.      <code class="structname">FuncCallContext</code> structure — use the

  856.      <code class="structfield">tuple_desc</code> or <code class="structfield">attinmeta</code> field

  857.      respectively.

  858.     </p><p>

  859.      When working with Datums, use:

  860. </p><pre class="programlisting">

  861. HeapTuple heap_form_tuple(TupleDesc tupdesc, Datum *values, bool *isnull)

  862. </pre><p>

  863.      to build a <code class="structname">HeapTuple</code> given user data in Datum form.

  864.     </p><p>

  865.      When working with C strings, use:

  866. </p><pre class="programlisting">

  867. HeapTuple BuildTupleFromCStrings(AttInMetadata *attinmeta, char **values)

  868. </pre><p>

  869.      to build a <code class="structname">HeapTuple</code> given user data

  870.      in C string form.  <em class="parameter"><code>values</code></em> is an array of C strings,

  871.      one for each attribute of the return row. Each C string should be in

  872.      the form expected by the input function of the attribute data

  873.      type. In order to return a null value for one of the attributes,

  874.      the corresponding pointer in the <em class="parameter"><code>values</code></em> array

  875.      should be set to <code class="symbol">NULL</code>.  This function will need to

  876.      be called again for each row you return.

  877.     </p><p>

  878.      Once you have built a tuple to return from your function, it

  879.      must be converted into a <code class="type">Datum</code>. Use:

  880. </p><pre class="programlisting">

  881. HeapTupleGetDatum(HeapTuple tuple)

  882. </pre><p>

  883.      to convert a <code class="structname">HeapTuple</code> into a valid Datum.  This

  884.      <code class="type">Datum</code> can be returned directly if you intend to return

  885.      just a single row, or it can be used as the current return value

  886.      in a set-returning function.

  887.     </p><p>

  888.      An example appears in the next section.

  889.     </p></div><div class="sect2" id="XFUNC-C-RETURN-SET"><div class="titlepage"><div><div><h3 class="title">37.10.8. Returning Sets</h3></div></div></div><p>

  890.      C-language functions have two options for returning sets (multiple

  891.      rows).  In one method, called <em class="firstterm">ValuePerCall</em>

  892.      mode, a set-returning function is called repeatedly (passing the same

  893.      arguments each time) and it returns one new row on each call, until

  894.      it has no more rows to return and signals that by returning NULL.

  895.      The set-returning function (<acronym class="acronym">SRF</acronym>) must therefore

  896.      save enough state across calls to remember what it was doing and

  897.      return the correct next item on each call.

  898.      In the other method, called <em class="firstterm">Materialize</em> mode,

  899.      a SRF fills and returns a tuplestore object containing its

  900.      entire result; then only one call occurs for the whole result, and

  901.      no inter-call state is needed.

  902.     </p><p>

  903.      When using ValuePerCall mode, it is important to remember that the

  904.      query is not guaranteed to be run to completion; that is, due to

  905.      options such as <code class="literal">LIMIT</code>, the executor might stop

  906.      making calls to the set-returning function before all rows have been

  907.      fetched.  This means it is not safe to perform cleanup activities in

  908.      the last call, because that might not ever happen.  It's recommended

  909.      to use Materialize mode for functions that need access to external

  910.      resources, such as file descriptors.

  911.     </p><p>

  912.      The remainder of this section documents a set of helper macros that

  913.      are commonly used (though not required to be used) for SRFs using

  914.      ValuePerCall mode.  Additional details about Materialize mode can be

  915.      found in <code class="filename">src/backend/utils/fmgr/README</code>.  Also,

  916.      the <code class="filename">contrib</code> modules in

  917.      the <span class="productname">PostgreSQL</span> source distribution contain

  918.      many examples of SRFs using both ValuePerCall and Materialize mode.

  919.     </p><p>

  920.      To use the ValuePerCall support macros described here,

  921.      include <code class="filename">funcapi.h</code>.  These macros work with a

  922.      structure <code class="structname">FuncCallContext</code> that contains the

  923.      state that needs to be saved across calls.  Within the calling

  924.      SRF, <code class="literal">fcinfo-&gt;flinfo-&gt;fn_extra</code> is used to

  925.      hold a pointer to <code class="structname">FuncCallContext</code> across

  926.      calls.  The macros automatically fill that field on first use,

  927.      and expect to find the same pointer there on subsequent uses.

  928. </p><pre class="programlisting">

  929. typedef struct FuncCallContext

  930. {

  931.     /*

  932.      * Number of times we've been called before

  933.      *

  934.      * call_cntr is initialized to 0 for you by SRF_FIRSTCALL_INIT(), and

  935.      * incremented for you every time SRF_RETURN_NEXT() is called.

  936.      */

  937.     uint64 call_cntr;

  938. 

  939.     /*

  940.      * OPTIONAL maximum number of calls

  941.      *

  942.      * max_calls is here for convenience only and setting it is optional.

  943.      * If not set, you must provide alternative means to know when the

  944.      * function is done.

  945.      */

  946.     uint64 max_calls;

  947. 

  948.     /*

  949.      * OPTIONAL pointer to miscellaneous user-provided context information

  950.      *

  951.      * user_fctx is for use as a pointer to your own data to retain

  952.      * arbitrary context information between calls of your function.

  953.      */

  954.     void *user_fctx;

  955. 

  956.     /*

  957.      * OPTIONAL pointer to struct containing attribute type input metadata

  958.      *

  959.      * attinmeta is for use when returning tuples (i.e., composite data types)

  960.      * and is not used when returning base data types. It is only needed

  961.      * if you intend to use BuildTupleFromCStrings() to create the return

  962.      * tuple.

  963.      */

  964.     AttInMetadata *attinmeta;

  965. 

  966.     /*

  967.      * memory context used for structures that must live for multiple calls

  968.      *

  969.      * multi_call_memory_ctx is set by SRF_FIRSTCALL_INIT() for you, and used

  970.      * by SRF_RETURN_DONE() for cleanup. It is the most appropriate memory

  971.      * context for any memory that is to be reused across multiple calls

  972.      * of the SRF.

  973.      */

  974.     MemoryContext multi_call_memory_ctx;

  975. 

  976.     /*

  977.      * OPTIONAL pointer to struct containing tuple description

  978.      *

  979.      * tuple_desc is for use when returning tuples (i.e., composite data types)

  980.      * and is only needed if you are going to build the tuples with

  981.      * heap_form_tuple() rather than with BuildTupleFromCStrings().  Note that

  982.      * the TupleDesc pointer stored here should usually have been run through

  983.      * BlessTupleDesc() first.

  984.      */

  985.     TupleDesc tuple_desc;

  986. 

  987. } FuncCallContext;

  988. </pre><p>

  989.     </p><p>

  990.      The macros to be used by an <acronym class="acronym">SRF</acronym> using this

  991.      infrastructure are:

  992. </p><pre class="programlisting">

  993. SRF_IS_FIRSTCALL()

  994. </pre><p>

  995.      Use this to determine if your function is being called for the first or a

  996.      subsequent time. On the first call (only), call:

  997. </p><pre class="programlisting">

  998. SRF_FIRSTCALL_INIT()

  999. </pre><p>

  1000.      to initialize the <code class="structname">FuncCallContext</code>. On every function call,

  1001.      including the first, call:

  1002. </p><pre class="programlisting">

  1003. SRF_PERCALL_SETUP()

  1004. </pre><p>

  1005.      to set up for using the <code class="structname">FuncCallContext</code>.

  1006.     </p><p>

  1007.      If your function has data to return in the current call, use:

  1008. </p><pre class="programlisting">

  1009. SRF_RETURN_NEXT(funcctx, result)

  1010. </pre><p>

  1011.      to return it to the caller.  (<code class="literal">result</code> must be of type

  1012.      <code class="type">Datum</code>, either a single value or a tuple prepared as

  1013.      described above.)  Finally, when your function is finished

  1014.      returning data, use:

  1015. </p><pre class="programlisting">

  1016. SRF_RETURN_DONE(funcctx)

  1017. </pre><p>

  1018.      to clean up and end the <acronym class="acronym">SRF</acronym>.

  1019.     </p><p>

  1020.      The memory context that is current when the <acronym class="acronym">SRF</acronym> is called is

  1021.      a transient context that will be cleared between calls.  This means

  1022.      that you do not need to call <code class="function">pfree</code> on everything

  1023.      you allocated using <code class="function">palloc</code>; it will go away anyway.  However, if you want to allocate

  1024.      any data structures to live across calls, you need to put them somewhere

  1025.      else.  The memory context referenced by

  1026.      <code class="structfield">multi_call_memory_ctx</code> is a suitable location for any

  1027.      data that needs to survive until the <acronym class="acronym">SRF</acronym> is finished running.  In most

  1028.      cases, this means that you should switch into

  1029.      <code class="structfield">multi_call_memory_ctx</code> while doing the

  1030.      first-call setup.

  1031.      Use <code class="literal">funcctx-&gt;user_fctx</code> to hold a pointer to

  1032.      any such cross-call data structures.

  1033.      (Data you allocate

  1034.      in <code class="structfield">multi_call_memory_ctx</code> will go away

  1035.      automatically when the query ends, so it is not necessary to free

  1036.      that data manually, either.)

  1037.     </p><div class="warning"><h3 class="title">Warning</h3><p>

  1038.       While the actual arguments to the function remain unchanged between

  1039.       calls, if you detoast the argument values (which is normally done

  1040.       transparently by the

  1041.       <code class="function">PG_GETARG_<em class="replaceable"><code>xxx</code></em></code> macro)

  1042.       in the transient context then the detoasted copies will be freed on

  1043.       each cycle. Accordingly, if you keep references to such values in

  1044.       your <code class="structfield">user_fctx</code>, you must either copy them into the

  1045.       <code class="structfield">multi_call_memory_ctx</code> after detoasting, or ensure

  1046.       that you detoast the values only in that context.

  1047.      </p></div><p>

  1048.      A complete pseudo-code example looks like the following:

  1049. </p><pre class="programlisting">

  1050. Datum

  1051. my_set_returning_function(PG_FUNCTION_ARGS)

  1052. {

  1053.     FuncCallContext  *funcctx;

  1054.     Datum             result;

  1055.     <em class="replaceable"><code>further declarations as needed</code></em>

  1056. 

  1057.     if (SRF_IS_FIRSTCALL())

  1058.     {

  1059.         MemoryContext oldcontext;

  1060. 

  1061.         funcctx = SRF_FIRSTCALL_INIT();

  1062.         oldcontext = MemoryContextSwitchTo(funcctx-&gt;multi_call_memory_ctx);

  1063.         /* One-time setup code appears here: */

  1064.         <em class="replaceable"><code>user code</code></em>

  1065.         <em class="replaceable"><code>if returning composite</code></em>

  1066.             <em class="replaceable"><code>build TupleDesc, and perhaps AttInMetadata</code></em>

  1067.         <em class="replaceable"><code>endif returning composite</code></em>

  1068.         <em class="replaceable"><code>user code</code></em>

  1069.         MemoryContextSwitchTo(oldcontext);

  1070.     }

  1071. 

  1072.     /* Each-time setup code appears here: */

  1073.     <em class="replaceable"><code>user code</code></em>

  1074.     funcctx = SRF_PERCALL_SETUP();

  1075.     <em class="replaceable"><code>user code</code></em>

  1076. 

  1077.     /* this is just one way we might test whether we are done: */

  1078.     if (funcctx-&gt;call_cntr &lt; funcctx-&gt;max_calls)

  1079.     {

  1080.         /* Here we want to return another item: */

  1081.         <em class="replaceable"><code>user code</code></em>

  1082.         <em class="replaceable"><code>obtain result Datum</code></em>

  1083.         SRF_RETURN_NEXT(funcctx, result);

  1084.     }

  1085.     else

  1086.     {

  1087.         /* Here we are done returning items, so just report that fact. */

  1088.         /* (Resist the temptation to put cleanup code here.) */

  1089.         SRF_RETURN_DONE(funcctx);

  1090.     }

  1091. }

  1092. </pre><p>

  1093.     </p><p>

  1094.      A complete example of a simple <acronym class="acronym">SRF</acronym> returning a composite type

  1095.      looks like:

  1096. </p><pre class="programlisting">

  1097. PG_FUNCTION_INFO_V1(retcomposite);

  1098. 

  1099. Datum

  1100. retcomposite(PG_FUNCTION_ARGS)

  1101. {

  1102.     FuncCallContext     *funcctx;

  1103.     int                  call_cntr;

  1104.     int                  max_calls;

  1105.     TupleDesc            tupdesc;

  1106.     AttInMetadata       *attinmeta;

  1107. 

  1108.     /* stuff done only on the first call of the function */

  1109.     if (SRF_IS_FIRSTCALL())

  1110.     {

  1111.         MemoryContext   oldcontext;

  1112. 

  1113.         /* create a function context for cross-call persistence */

  1114.         funcctx = SRF_FIRSTCALL_INIT();

  1115. 

  1116.         /* switch to memory context appropriate for multiple function calls */

  1117.         oldcontext = MemoryContextSwitchTo(funcctx-&gt;multi_call_memory_ctx);

  1118. 

  1119.         /* total number of tuples to be returned */

  1120.         funcctx-&gt;max_calls = PG_GETARG_UINT32(0);

  1121. 

  1122.         /* Build a tuple descriptor for our result type */

  1123.         if (get_call_result_type(fcinfo, NULL, &amp;tupdesc) != TYPEFUNC_COMPOSITE)

  1124.             ereport(ERROR,

  1125.                     (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),

  1126.                      errmsg("function returning record called in context "

  1127.                             "that cannot accept type record")));

  1128. 

  1129.         /*

  1130.          * generate attribute metadata needed later to produce tuples from raw

  1131.          * C strings

  1132.          */

  1133.         attinmeta = TupleDescGetAttInMetadata(tupdesc);

  1134.         funcctx-&gt;attinmeta = attinmeta;

  1135. 

  1136.         MemoryContextSwitchTo(oldcontext);

  1137.     }

  1138. 

  1139.     /* stuff done on every call of the function */

  1140.     funcctx = SRF_PERCALL_SETUP();

  1141. 

  1142.     call_cntr = funcctx-&gt;call_cntr;

  1143.     max_calls = funcctx-&gt;max_calls;

  1144.     attinmeta = funcctx-&gt;attinmeta;

  1145. 

  1146.     if (call_cntr &lt; max_calls)    /* do when there is more left to send */

  1147.     {

  1148.         char       **values;

  1149.         HeapTuple    tuple;

  1150.         Datum        result;

  1151. 

  1152.         /*

  1153.          * Prepare a values array for building the returned tuple.

  1154.          * This should be an array of C strings which will

  1155.          * be processed later by the type input functions.

  1156.          */

  1157.         values = (char **) palloc(3 * sizeof(char *));

  1158.         values[0] = (char *) palloc(16 * sizeof(char));

  1159.         values[1] = (char *) palloc(16 * sizeof(char));

  1160.         values[2] = (char *) palloc(16 * sizeof(char));

  1161. 

  1162.         snprintf(values[0], 16, "%d", 1 * PG_GETARG_INT32(1));

  1163.         snprintf(values[1], 16, "%d", 2 * PG_GETARG_INT32(1));

  1164.         snprintf(values[2], 16, "%d", 3 * PG_GETARG_INT32(1));

  1165. 

  1166.         /* build a tuple */

  1167.         tuple = BuildTupleFromCStrings(attinmeta, values);

  1168. 

  1169.         /* make the tuple into a datum */

  1170.         result = HeapTupleGetDatum(tuple);

  1171. 

  1172.         /* clean up (this is not really necessary) */

  1173.         pfree(values[0]);

  1174.         pfree(values[1]);

  1175.         pfree(values[2]);

  1176.         pfree(values);

  1177. 

  1178.         SRF_RETURN_NEXT(funcctx, result);

  1179.     }

  1180.     else    /* do when there is no more left */

  1181.     {

  1182.         SRF_RETURN_DONE(funcctx);

  1183.     }

  1184. }

  1185. 

  1186. </pre><p>

  1187. 

  1188.      One way to declare this function in SQL is:

  1189. </p><pre class="programlisting">

  1190. CREATE TYPE __retcomposite AS (f1 integer, f2 integer, f3 integer);

  1191. 

  1192. CREATE OR REPLACE FUNCTION retcomposite(integer, integer)

  1193.     RETURNS SETOF __retcomposite

  1194.     AS '<em class="replaceable"><code>filename</code></em>', 'retcomposite'

  1195.     LANGUAGE C IMMUTABLE STRICT;

  1196. </pre><p>

  1197.      A different way is to use OUT parameters:

  1198. </p><pre class="programlisting">

  1199. CREATE OR REPLACE FUNCTION retcomposite(IN integer, IN integer,

  1200.     OUT f1 integer, OUT f2 integer, OUT f3 integer)

  1201.     RETURNS SETOF record

  1202.     AS '<em class="replaceable"><code>filename</code></em>', 'retcomposite'

  1203.     LANGUAGE C IMMUTABLE STRICT;

  1204. </pre><p>

  1205.      Notice that in this method the output type of the function is formally

  1206.      an anonymous <code class="structname">record</code> type.

  1207.     </p></div><div class="sect2" id="id-1.8.3.13.13"><div class="titlepage"><div><div><h3 class="title">37.10.9. Polymorphic Arguments and Return Types</h3></div></div></div><p>

  1208.      C-language functions can be declared to accept and

  1209.      return the polymorphic types

  1210.      <code class="type">anyelement</code>, <code class="type">anyarray</code>, <code class="type">anynonarray</code>,

  1211.      <code class="type">anyenum</code>, and <code class="type">anyrange</code>.

  1212.      See <a class="xref" href="extend-type-system.html#EXTEND-TYPES-POLYMORPHIC" title="37.2.5. Polymorphic Types">Section 37.2.5</a> for a more detailed explanation

  1213.      of polymorphic functions. When function arguments or return types

  1214.      are defined as polymorphic types, the function author cannot know

  1215.      in advance what data type it will be called with, or

  1216.      need to return. There are two routines provided in <code class="filename">fmgr.h</code>

  1217.      to allow a version-1 C function to discover the actual data types

  1218.      of its arguments and the type it is expected to return. The routines are

  1219.      called <code class="literal">get_fn_expr_rettype(FmgrInfo *flinfo)</code> and

  1220.      <code class="literal">get_fn_expr_argtype(FmgrInfo *flinfo, int argnum)</code>.

  1221.      They return the result or argument type OID, or <code class="symbol">InvalidOid</code> if the

  1222.      information is not available.

  1223.      The structure <code class="literal">flinfo</code> is normally accessed as

  1224.      <code class="literal">fcinfo-&gt;flinfo</code>. The parameter <code class="literal">argnum</code>

  1225.      is zero based.  <code class="function">get_call_result_type</code> can also be used

  1226.      as an alternative to <code class="function">get_fn_expr_rettype</code>.

  1227.      There is also <code class="function">get_fn_expr_variadic</code>, which can be used to

  1228.      find out whether variadic arguments have been merged into an array.

  1229.      This is primarily useful for <code class="literal">VARIADIC "any"</code> functions,

  1230.      since such merging will always have occurred for variadic functions

  1231.      taking ordinary array types.

  1232.     </p><p>

  1233.      For example, suppose we want to write a function to accept a single

  1234.      element of any type, and return a one-dimensional array of that type:

  1235. 

  1236. </p><pre class="programlisting">

  1237. PG_FUNCTION_INFO_V1(make_array);

  1238. Datum

  1239. make_array(PG_FUNCTION_ARGS)

  1240. {

  1241.     ArrayType  *result;

  1242.     Oid         element_type = get_fn_expr_argtype(fcinfo-&gt;flinfo, 0);

  1243.     Datum       element;

  1244.     bool        isnull;

  1245.     int16       typlen;

  1246.     bool        typbyval;

  1247.     char        typalign;

  1248.     int         ndims;

  1249.     int         dims[MAXDIM];

  1250.     int         lbs[MAXDIM];

  1251. 

  1252.     if (!OidIsValid(element_type))

  1253.         elog(ERROR, "could not determine data type of input");

  1254. 

  1255.     /* get the provided element, being careful in case it's NULL */

  1256.     isnull = PG_ARGISNULL(0);

  1257.     if (isnull)

  1258.         element = (Datum) 0;

  1259.     else

  1260.         element = PG_GETARG_DATUM(0);

  1261. 

  1262.     /* we have one dimension */

  1263.     ndims = 1;

  1264.     /* and one element */

  1265.     dims[0] = 1;

  1266.     /* and lower bound is 1 */

  1267.     lbs[0] = 1;

  1268. 

  1269.     /* get required info about the element type */

  1270.     get_typlenbyvalalign(element_type, &amp;typlen, &amp;typbyval, &amp;typalign);

  1271. 

  1272.     /* now build the array */

  1273.     result = construct_md_array(&amp;element, &amp;isnull, ndims, dims, lbs,

  1274.                                 element_type, typlen, typbyval, typalign);

  1275. 

  1276.     PG_RETURN_ARRAYTYPE_P(result);

  1277. }

  1278. </pre><p>

  1279.     </p><p>

  1280.      The following command declares the function

  1281.      <code class="function">make_array</code> in SQL:

  1282. 

  1283. </p><pre class="programlisting">

  1284. CREATE FUNCTION make_array(anyelement) RETURNS anyarray

  1285.     AS '<em class="replaceable"><code>DIRECTORY</code></em>/funcs', 'make_array'

  1286.     LANGUAGE C IMMUTABLE;

  1287. </pre><p>

  1288.     </p><p>

  1289.      There is a variant of polymorphism that is only available to C-language

  1290.      functions: they can be declared to take parameters of type

  1291.      <code class="literal">"any"</code>.  (Note that this type name must be double-quoted,

  1292.      since it's also a SQL reserved word.)  This works like

  1293.      <code class="type">anyelement</code> except that it does not constrain different

  1294.      <code class="literal">"any"</code> arguments to be the same type, nor do they help

  1295.      determine the function's result type.  A C-language function can also

  1296.      declare its final parameter to be <code class="literal">VARIADIC "any"</code>.  This will

  1297.      match one or more actual arguments of any type (not necessarily the same

  1298.      type).  These arguments will <span class="emphasis"><em>not</em></span> be gathered into an array

  1299.      as happens with normal variadic functions; they will just be passed to

  1300.      the function separately.  The <code class="function">PG_NARGS()</code> macro and the

  1301.      methods described above must be used to determine the number of actual

  1302.      arguments and their types when using this feature.  Also, users of such

  1303.      a function might wish to use the <code class="literal">VARIADIC</code> keyword in their

  1304.      function call, with the expectation that the function would treat the

  1305.      array elements as separate arguments.  The function itself must implement

  1306.      that behavior if wanted, after using <code class="function">get_fn_expr_variadic</code> to

  1307.      detect that the actual argument was marked with <code class="literal">VARIADIC</code>.

  1308.     </p></div><div class="sect2" id="id-1.8.3.13.14"><div class="titlepage"><div><div><h3 class="title">37.10.10. Shared Memory and LWLocks</h3></div></div></div><p>

  1309.      Add-ins can reserve LWLocks and an allocation of shared memory on server

  1310.      startup.  The add-in's shared library must be preloaded by specifying

  1311.      it in

  1312.      <a class="xref" href="runtime-config-client.html#GUC-SHARED-PRELOAD-LIBRARIES">shared_preload_libraries</a><a id="id-1.8.3.13.14.2.2" class="indexterm"></a>.

  1313.      Shared memory is reserved by calling:

  1314. </p><pre class="programlisting">

  1315. void RequestAddinShmemSpace(int size)

  1316. </pre><p>

  1317.      from your <code class="function">_PG_init</code> function.

  1318.     </p><p>

  1319.      LWLocks are reserved by calling:

  1320. </p><pre class="programlisting">

  1321. void RequestNamedLWLockTranche(const char *tranche_name, int num_lwlocks)

  1322. </pre><p>

  1323.      from <code class="function">_PG_init</code>.  This will ensure that an array of

  1324.      <code class="literal">num_lwlocks</code> LWLocks is available under the name

  1325.      <code class="literal">tranche_name</code>.  Use <code class="function">GetNamedLWLockTranche</code>

  1326.      to get a pointer to this array.

  1327.     </p><p>

  1328.      To avoid possible race-conditions, each backend should use the LWLock

  1329.      <code class="function">AddinShmemInitLock</code> when connecting to and initializing

  1330.      its allocation of shared memory, as shown here:

  1331. </p><pre class="programlisting">

  1332. static mystruct *ptr = NULL;

  1333. 

  1334. if (!ptr)

  1335. {

  1336.         bool    found;

  1337. 

  1338.         LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);

  1339.         ptr = ShmemInitStruct("my struct name", size, &amp;found);

  1340.         if (!found)

  1341.         {

  1342.                 initialize contents of shmem area;

  1343.                 acquire any requested LWLocks using:

  1344.                 ptr-&gt;locks = GetNamedLWLockTranche("my tranche name");

  1345.         }

  1346.         LWLockRelease(AddinShmemInitLock);

  1347. }

  1348. </pre><p>

  1349.     </p></div><div class="sect2" id="EXTEND-CPP"><div class="titlepage"><div><div><h3 class="title">37.10.11. Using C++ for Extensibility</h3></div></div></div><a id="id-1.8.3.13.15.2" class="indexterm"></a><p>

  1350.      Although the <span class="productname">PostgreSQL</span> backend is written in

  1351.      C, it is possible to write extensions in C++ if these guidelines are

  1352.      followed:

  1353. 

  1354.      </p><div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; "><li class="listitem"><p>

  1355.          All functions accessed by the backend must present a C interface

  1356.          to the backend;  these C functions can then call C++ functions.

  1357.          For example, <code class="literal">extern C</code> linkage is required for

  1358.          backend-accessed functions.  This is also necessary for any

  1359.          functions that are passed as pointers between the backend and

  1360.          C++ code.

  1361.        </p></li><li class="listitem"><p>

  1362.         Free memory using the appropriate deallocation method.  For example,

  1363.         most backend memory is allocated using <code class="function">palloc()</code>, so use

  1364.         <code class="function">pfree()</code> to free it.  Using C++

  1365.         <code class="function">delete</code> in such cases will fail.

  1366.        </p></li><li class="listitem"><p>

  1367.         Prevent exceptions from propagating into the C code (use a catch-all

  1368.         block at the top level of all <code class="literal">extern C</code> functions).  This

  1369.         is necessary even if the C++ code does not explicitly throw any

  1370.         exceptions, because events like out-of-memory can still throw

  1371.         exceptions.  Any exceptions must be caught and appropriate errors

  1372.         passed back to the C interface.  If possible, compile C++ with

  1373.         <code class="option">-fno-exceptions</code> to eliminate exceptions entirely; in such

  1374.         cases, you must check for failures in your C++ code, e.g.  check for

  1375.         NULL returned by <code class="function">new()</code>.

  1376.        </p></li><li class="listitem"><p>

  1377.         If calling backend functions from C++ code, be sure that the

  1378.         C++ call stack contains only plain old data structures

  1379.         (<acronym class="acronym">POD</acronym>).  This is necessary because backend errors

  1380.         generate a distant <code class="function">longjmp()</code> that does not properly

  1381.         unroll a C++ call stack with non-POD objects.

  1382.        </p></li></ul></div><p>

  1383.     </p><p>

  1384.      In summary, it is best to place C++ code behind a wall of

  1385.      <code class="literal">extern C</code> functions that interface to the backend,

  1386.      and avoid exception, memory, and call stack leakage.

  1387.     </p></div></div><div class="navfooter"><hr /><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="xfunc-internal.html">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="extend.html">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="xfunc-optimization.html">Next</a></td></tr><tr><td width="40%" align="left" valign="top">37.9. Internal Functions </td><td width="20%" align="center"><a accesskey="h" href="index.html">Home</a></td><td width="40%" align="right" valign="top"> 37.11. Function Optimization Information</td></tr></table></div></body></html>
上海开阖软件有限公司 沪ICP备12045867号-1