<?xml version="1.0" encoding='UTF-8'?>
<!DOCTYPE sect1 PUBLIC "-//OASIS//DTD DocBook V4.4//EN"
    "http://www.oasis-open.org/docbook/xml/4.4/docbookx.dtd">

<sect1 id="tutorial" xreflabel="Tutorial">
  <title>Tutorial</title>

  <para>The <link linkend="overview">previous chapter</link> introduced
  the major top-level mechanisms in MySQL++. Now we&#x2019;ll dig down a
  little deeper and get into real examples. We start off with the basics
  that every MySQL++ program will have to deal with, then work up to
  more complex topics that are still widely interesting. You can stop
  reading the manual after this chapter and still get a lot out of
  MySQL++, ignoring the more advanced parts we present in later
  chapters.</para>


  <sect2 id="examples">
    <title>Running the Examples</title>

    <para>All of the examples are complete running programs. If you
    built the library from source, the examples should have been built
    as well. If you use RPMs instead, the example programs&#x2019; source
    code and a simplified <filename>Makefile</filename> are in the
    <filename>mysql++-devel</filename> package. They are typically
    installed in
    <filename>/usr/share/doc/mysql++-devel-*/examples</filename>, but it
    can vary on different Linuxes.</para>

    <para>Before you get started, please read through any of the
    <filename>README*.txt</filename> files included with the MySQL++
    distribution that are relevant to your platform. We won&#x2019;t
    repeat all of that here.</para>

    <para>Most of the examples require a test database, created by
    <filename>resetdb</filename>. You can run it like so:</para>

    <screen>resetdb [-s server_addr] [-u user] [-p password]</screen>

    <para>Actually, there&#x2019;s a problem with that. It assumes that
    the MySQL++ library is already installed in a directory that the
    operating system&#x2019;s dynamic linker can find. (MySQL++ is almost
    never built statically.) Unless you&#x2019;re installing from RPMs,
    you&#x2019;ve had to build the library from source, and you should
    run at least a few of the examples before installing the library to
    be sure it&#x2019;s working correctly. Since your operating
    system&#x2019;s dynamic linkage system can&#x2019;t find the MySQL++
    libraries without help until they&#x2019;re installed, we&#x2019;ve
    created a few helper scripts to help run the examples.</para>

    <para>MySQL++ comes with the <filename>exrun</filename> shell script
    for Unixy systems, and the <filename>exrun.bat</filename> batch file
    for Windows. You pass the example program and its arguments to the
    <filename>exrun</filename> helper, which sets up the library search
    path so that it will use the as-yet uninstalled version of the
    MySQL++ library in preference to any other on your system:</para>

    <screen>./exrun resetdb [-s server_addr] [-u user] [-p password]</screen>

    <para>That&#x2019;s the typical form for a Unixy system. You leave
    off the <command>./</command> bit on Windows. You can leave it
    off on a Unixy system, too, if you have <filename>.</filename>
    in your <varname>PATH</varname>. (Not a recommendation, just
    an observation.)</para>

    <para>All of the program arguments are optional.</para>

    <para>If you don&#x2019;t give <option>-s</option>,
    the underlying MySQL C API (a.k.a. <ulink
    url="https://dev.mysql.com/downloads/connector/c/">Connector/C</ulink>)
    assumes the server is on the local machine. It chooses
    one of several different IPC options based on the platform
    configuration. There are many different forms you can give
    as <varname>server_addr</varname> with <option>-s</option> to
    override this default behavior:</para>

    <itemizedlist>
      <listitem>
        <para><emphasis>localhost</emphasis> &mdash; this is the
        default; it doesn&#x2019;t buy you anything</para>
      </listitem>

      <listitem>
        <para>On Windows, a simple period tells the underlying MySQL C
        API to use named pipes, if it&#x2019;s available.</para>
      </listitem>

      <listitem>
        <para><emphasis>172.20.0.252:12345</emphasis> &mdash; this would
        connect to IP address
        <computeroutput>172.20.0.252</computeroutput> on TCP port
        <computeroutput>12345</computeroutput>.</para>
      </listitem>

      <listitem>
        <para><emphasis>my.server.name:svc_name</emphasis> &mdash; this
        would first look up TCP service name
        <computeroutput>svc_name</computeroutput> in your system&#x2019;s
        network services database (<filename>/etc/services</filename> on
        Unixy systems, and something like
        <filename>c:\windows\system32\drivers\etc\services</filename> on
        modern Windows variants). If it finds an entry for the service,
        it then tries to connect to that port on the domain name
        given.</para>
      </listitem>
    </itemizedlist>

    <para>For the TCP forms, you can mix names and numbers for the host
    and port/service parts in any combination. If the server name
    doesn&#x2019;t contain a colon, it uses the default port,
    3306.</para>

    <para>If you don&#x2019;t give <option>-u</option>, it assumes your
    user name on the database server is the same as your login name on
    the local machine.</para>

    <para>If you don&#x2019;t give <option>-p</option>, it will assume
    the MySQL user doesn&#x2019;t have a password. (One hopes this
    isn&#x2019;t the case...)</para>

    <para>When running <filename>resetdb</filename>, the user name needs
    to be for an account with permission to create the test database.
    Once the database is created, you can use any account when running
    the other examples that has DELETE, INSERT, SELECT and UPDATE
    permissions for the test database. The MySQL root user can do all
    this, of course, but you might want to set up a separate user,
    having only the permissions necessary to work with the test
    database:</para>

    <screen>
CREATE USER mysqlpp_test@'%' IDENTIFIED BY &#x2019;nunyabinness';
GRANT ALL PRIVILEGES ON mysql_cpp_data.* TO mysqlpp_test@'%';</screen>

    <para>You could then create the sample database with the following
    command:</para>

    <screen>./exrun resetdb -u mysqlpp_test -p nunyabinness</screen>

    <para>(Again, leave off the <command>./</command> bit on
    Windows.)</para>

    <para>You may have to re-run <filename>resetdb</filename> after
    running some of the other examples, as they change the
    database.</para>

    <para>See <filename>README-examples.txt</filename> for more
    details on running the examples.</para>
  </sect2>


  <sect2 id="simple">
    <title>A Simple Example</title>

    <para>The following example demonstrates how to open a connection,
    execute a simple query, and display the results. This is
    <filename>examples/simple1.cpp</filename>:</para>

    <programlisting><xi:include href="simple1.txt" parse="text"
    xmlns:xi="http://www.w3.org/2001/XInclude"/></programlisting>

    <para>This example simply gets the entire "item" column from the
    example table, and prints those values out.</para>

    <para>Notice that MySQL++&#x2019;s <ulink type="classref"
    url="StoreQueryResult"/> derives from
    <classname>std::vector</classname>, and <ulink type="classref"
    url="Row"/> provides an interface that makes it a
    <classname>vector</classname> work-alike. This means you can access
    elements with subscript notation, walk through them with iterators,
    run STL algorithms on them, etc.</para>

    <para><classname>Row</classname> provides a little more in this area
    than a plain old <classname>vector</classname>: you can also access
    fields by name using subscript notation.</para>

    <para>The only thing that isn&#x2019;t explicit in the code above is
    that we delegate command line argument parsing to
    <function>parse_command_line()</function> in the
    <filename>excommon</filename> module. This function exists to give
    the examples a consistent interface, not to hide important details.
    You can treat it like a black box: it takes <varname>argc</varname>
    and <varname>argv</varname> as inputs and sends back database
    connection parameters.</para>
  </sect2>


  <sect2 id="simple2">
    <title>A More Complicated Example</title>

    <para>The <filename>simple1</filename> example above was pretty
    trivial. Let&#x2019;s get a little deeper. Here is
    <filename>examples/simple2.cpp</filename>:</para>

    <programlisting><xi:include href="simple2.txt" parse="text"
    xmlns:xi="http://www.w3.org/2001/XInclude"/></programlisting>

    <para>The main point of this example is that we&#x2019;re accessing
    fields in the row objects by name, instead of index. This is slower,
    but obviously clearer. We&#x2019;re also printing out the entire
    table, not just one column.</para>
  </sect2>


  <sect2 id="exceptions" xreflabel="exceptions">
    <title>Exceptions</title>

    <para>By default, MySQL++ uses exceptions to signal errors.
    We&#x2019;ve been suppressing this in all the examples so far by
    passing <symbol>false</symbol> to
    <classname>Connection</classname>&#x2019;s constructor. This kept
    these early examples simple at the cost of some flexibility and
    power in error handling. In a real program, we recommend that you
    leave exceptions enabled. You do this by either using the default
    <classname>Connection</classname> constructor, or by using the
    create-and-connect constructor.</para>

    <para>All of MySQL++&#x2019;s custom exceptions
    derive from a common base class, <ulink type="classref"
    url="Exception"/>. That in turn derives from Standard C++&#x2019;s
    <classname>std::exception</classname> class. Since the library
    can indirectly cause exceptions to come from the Standard
    C++ Library, it&#x2019;s possible to catch all exceptions from
    MySQL++ by just catching <classname>std::exception</classname>.
    However, it&#x2019;s better to have individual catch blocks
    for each of the concrete exception types that you expect, and
    add a handler for either <classname>Exception</classname>
    or <classname>std::exception</classname> to act as a
    &#x201C;catch-all&#x201D; for unexpected exceptions.</para>

    <para>When exceptions are suppressed, MySQL++ signals errors
    by returning either an error code or an object that tests
    as false, or by setting an error flag on the object. Classes
    that allow you to suppress exceptions derive from the <ulink
    type="classref" url="OptionalExceptions"/> interface. When
    an <classname>OptionalExceptions</classname> derivative
    creates another object that also derives from this interface,
    it passes on its exception flag. Since everything flows from
    the <ulink type="classref" url="Connection"/> object, disabling
    exceptions on it at the start of the program disables all optional
    exceptions. This is why passing <symbol>false</symbol> for the
    <classname>Connection</classname> constructor&#x2019;s &#x201C;throw
    exceptions&#x201D; parameter suppresses all optional exceptions
    in the <filename>simple[1-3]</filename> examples. It keeps them,
    well, simple.</para>

    <para>This exception suppression mechanism is quite granular.
    It&#x2019;s possible to leave exceptions enabled most of the time,
    but suppress them in sections of the code where they aren&#x2019;t
    helpful. To do this, put the section of code that you want to not
    throw exceptions inside a block, and create a <ulink type="classref"
    url="NoExceptions"/> object at the top of that block. When created,
    it saves the exception flag of the
    <classname>OptionalExceptions</classname> derivative you pass to it,
    and then disables exceptions on it. When the
    <classname>NoExceptions</classname> object goes out of scope at the
    end of the block, it restores the exceptions flag to its previous
    state:</para>

    <programlisting>mysqlpp::Connection con; // default ctor, so exceptions enabled

{
  mysqlpp::NoExceptions ne(con);
  if (!con.select_db("a_db_that_might_not_exist_yet")) {
    // Our DB doesn&#x2019;t exist yet, so create and select it here; no need
    // to push handling of this case way off in an exception handler.
  }
}</programlisting>

    <para>When one <classname>OptionalExceptions</classname> derivative
    passes its exceptions flag to another such object, it is only
    passing a copy; the two objects&#x2019; flags operate independently.
    There&#x2019;s no way to globally enable or disable this flag on
    existing objects in a single call. If you&#x2019;re using the
    <classname>NoExceptions</classname> feature and you&#x2019;re
    still seeing optional exceptions thrown, you disabled exceptions
    on the wrong object. The exception thrower could be unrelated to
    the object you disabled exceptions on, it could be its parent,
    or it could be a child created before you disabled optional
    exceptions.</para>

    <para>MySQL++ throws some exceptions unconditionally:</para>

    <itemizedlist>
      <listitem><para>MySQL++ checks array indices,
      always.  For instance, if your code said
      &#x201C;<varname>row[21]</varname>&#x201D; on a
      row containing only 5 fields, you&#x2019;d get a
      <classname>BadIndex</classname> exception. If you
      say &#x201C;<varname>row["fred"]</varname>&#x201D;
      on a row without a &#x201C;fred&#x201D; field, you get
      a <classname>BadFieldName</classname> exception. In
      the past, MySQL++ delegated some of its index checking
      to the STL containers underpinning it, so you could get
      <classname>std::range_error</classname> instead. As of MySQL++
      v3.0.7, this should no longer happen, but there may be instances
      where it still does.</para></listitem>

      <listitem><para><ulink type="classref" url="String"/> will always
      throw <ulink type="classref" url="BadConversion"/> when you ask it
      to do an improper type conversion. For example, you&#x2019;ll get
      an exception if you try to convert &#x201C;1.25&#x201D; to
      <type>int</type>, but not when you convert &#x201C;1.00&#x201D; to
      <type>int</type>. In the latter case, MySQL++ knows that it can
      safely throw away the fractional part.</para></listitem>

      <listitem><para>If you use template queries and don&#x2019;t pass
      enough parameters when instantiating the template,
      <classname>Query</classname> will throw a <ulink type="classref"
      url="BadParamCount"/> exception.</para></listitem>

      <listitem><para>If you use a C++ data type in a query
      that MySQL++ doesn&#x2019;t know to convert to SQL, MySQL++
      will throw a <ulink type="classref" url="TypeLookupFailed"/>
      exception. It typically happens with <xref linkend="ssqls"/>,
      especially when using data types other than the ones defined
      in <filename>lib/sql_types.h</filename>.</para></listitem>
    </itemizedlist>

    <para>It&#x2019;s educational to modify the examples to force
    exceptions. For instance, misspell a field name, use an out-of-range
    index, or change a type to force a <classname>String</classname>
    conversion error.</para>
  </sect2>


  <sect2 id="qescape" xreflabel="quoting and escaping">
    <title>Quoting and Escaping</title>

    <para>SQL syntax often requires certain data to be quoted. Consider
    this query:</para>

    <programlisting>
SELECT * FROM stock WHERE item = 'Hotdog Buns' </programlisting>

    <para>Because the string &#x201C;Hotdog Buns&#x201D; contains a space,
    it must be quoted. With MySQL++, you don&#x2019;t have to add these
    quote marks manually:</para>

    <programlisting>
string s = "Hotdog Buns";
query &lt;&lt; "SELECT * FROM stock WHERE item = " &lt;&lt; quote_only &lt;&lt; s; </programlisting>

    <para>That code produces the same query string as in the previous
    example. We used the MySQL++ <type>quote_only</type> manipulator,
    which causes single quotes to be added around the next item inserted
    into the stream. This works for any type of data that can be
    converted to MySQL++&#x2019;s <ulink type="classref"
    url="SQLTypeAdapter">SQLTypeAdapter</ulink> type, plus the <ulink
    type="classref" url="Set"/> template. <link
    linkend="ssqls">SSQLS</link> also uses these manipulators
    internally.</para>

    <para>Quoting is pretty simple, but SQL syntax also often requires
    that certain characters be &#x201C;escaped&#x201D;. Imagine if the
    string in the previous example was &#x201C;Frank&#x2019;s Brand Hotdog
    Buns&#x201D; instead. The resulting query would be:</para>

    <programlisting>
SELECT * FROM stock WHERE item = 'Frank's Brand Hotdog Buns' </programlisting>

    <para>That&#x2019;s not valid SQL syntax. The correct syntax is:</para>

    <programlisting>
SELECT * FROM stock WHERE item = 'Frank''s Brand Hotdog Buns' </programlisting>

    <para>As you might expect, MySQL++ provides that feature, too,
    through its <type>escape</type> manipulator. But here, we want both
    quoting and escaping. That brings us to the most widely useful
    manipulator:</para>

    <programlisting>
string s = "Frank&#x2019;s Brand Hotdog Buns";
query &lt;&lt; "SELECT * FROM stock WHERE item = " &lt;&lt; quote &lt;&lt; s; </programlisting>

    <para>The <type>quote</type> manipulator both quotes strings and
    escapes any characters that are special in SQL.</para>

    <para>MySQL++ provides other manipulators as well. See the <ulink
    url="../refman/manip_8h.html">manip.h</ulink> page in the <ulink
    url="../refman/index.html">reference manual</ulink>.</para>

    <para>It&#x2019;s important to realize that MySQL++&#x2019;s quoting
    and escaping mechanism is type-aware. Manipulators have no effect
    unless you insert the manipulator into a
    <classname>Query</classname> or <ulink type="classref"
    url="SQLQueryParms">SQLQueryParms</ulink> stream.
    <footnote><para><classname>SQLQueryParms</classname> is used as a
    stream only as an implementation detail within the library. End user
    code simply sees it as a <classname>std::vector</classname>
    derivative.</para></footnote> Also, values are only quoted and/or
    escaped if they are of a data type that may need it. For example,
    <ulink type="structref" url="Date">Date</ulink> must be quoted but
    never needs to be escaped, and integer types need neither quoting
    nor escaping. Manipulators are suggestions to the library, not
    commands: MySQL++ will ignore these suggestions if it knows it
    won&#x2019;t result in syntactically-incorrect SQL.</para>

    <para>It&#x2019;s also important to realize that quoting and escaping
    in <classname>Query</classname> streams and template queries is
    never implicit.<footnote><para>By contrast, the
    <classname>Query</classname> methods that take an <link
    linkend="ssqls">SSQLS</link> <emphasis>do</emphasis> add quotes and
    escape strings implicitly. It can do this because SSQLS knows all
    the SQL code and data types, so it never has to guess whether
    quoting or escaping is appropriate.</para></footnote> You must use
    manipulators and template query flags as necessary to tell MySQL++
    where quoting and escaping is necessary. It would be nice if MySQL++
    could do quoting and escaping implicitly based on data type, but
    this isn&#x2019;t possible in all cases.<footnote
    id="whyexpmanip"><para>Unless you&#x2019;re smarter than I am, you
    don&#x2019;t immediately see why explicit manipulators are necessary.
    We can tell when quoting and escaping is <emphasis>not</emphasis>
    appropriate based on type, so doesn&#x2019;t that mean we know when
    it <emphasis>is</emphasis> appropriate?  Alas, no.  For most data
    types, it is possible to know, or at least make an awfully good
    guess, but it&#x2019;s a complete toss-up for C strings, <type>const
    char*</type>. A C string could be either a literal string of SQL
    code, or it can be a value used in a query. Since there&#x2019;s no
    easy way to know and it would damage the library&#x2019;s usability
    to mandate that C strings only be used for one purpose or the other,
    the library requires you to be explicit.</para></footnote> Since
    MySQL++ can&#x2019;t reliably guess when quoting and escaping is
    appropriate, and the programmer doesn&#x2019;t need
    to<footnote><para>One hopes the programmer
    <emphasis>knows</emphasis>.</para></footnote>, MySQL++ makes you
    tell it.</para>
  </sect2>


  <sect2 id="sql-types">
    <title>C++ vs. SQL Data Types</title>

    <para>The C++ and SQL data type systems have several differences
    that can cause problems when using MySQL++, or any other SQL
    based system, for that matter.</para>

    <para>Most of the data types you can store in a SQL database are
    either numbers or text strings. If you&#x2019;re only looking at
    the data going between the database server and your application,
    there aren&#x2019;t even numbers: SQL is a textual language, so
    numbers and everything else gets transferred between the client
    and the database server in text string form.<footnote><para>Yes,
    we&#x2019;re aware that there is a feature in MySQL that lets you
    transfer row data in a binary form, but we don&#x2019;t support
    this yet. We may, someday, probably as an extension to <link
    linkend="ssqls">SSQLS</link>. The only real reason to do so
    is to shave off some of the data translation overhead, which
    is typically neglibible in practice, swamped by the far greater
    disk and network I/O overheads inherent in use of a client-server
    database system like MySQL.</para></footnote> Consequently, MySQL++
    has a lot of <link linkend="string-types">special support</link>
    for text strings, and can translate to several C++ numeric data
    types transparently.</para>

    <para>Some people worry that this translation via an intermediate
    string form will cause data loss. Obviously the text string data
    types are immune from problems in this regard. We&#x2019;re also
    confident that MySQL++ translates <link linkend="blob">BLOB</link>
    and integer data types losslessly.</para>

    <para>The biggest worry is with floating-point numbers. (The FLOAT
    and DOUBLE SQL data types.) We did have a problem with this in
    older versions of MySQL++, but we believe we fixed it completely
    in v3.0.2. No one has since proven data loss via this path. There
    is still a known problem

    <footnote>
      <para>SQL&#x2019;s DECIMAL data type is a configurable-precision
      fixed-point number format. MySQL++ currently translates these to
      <type>double</type>, a floating-point data format, the closest
      thing available in the C++ type system. Since the main reason
      to use DECIMAL is to get away from the weird roundoff behavior
      of floating-point numbers, this could be viewed as a serious
      problem. The thing is, though, in all the years MySQL++ has
      been around, I don&#x2019;t remember anyone actually complaining
      about it. Apparently there&#x2019;s either no one using DECIMAL
      with MySQL++, or they&#x2019;re ignoring any roundoff errors
      they get as a result. Until this wheel squeaks, it&#x2019;s not
      likely to be greased. To fix this, we&#x2019;ll have to create
      a new custom data type to hold such column values, which will
      be a lot of work for apparently little return.</para>
    </footnote>

    with the SQL DECIMAL type, which is somewhat related to the
    floating-point issue, but it&#x2019;s apparently rarely encountered,
    which is why it hasn&#x2019;t been fixed yet.</para>

    <para>The best way to avoid problems with data translation
    is to always use the special MySQL++ data types defined in
    <filename>lib/sql_types.h</filename> corresponding to your
    SQL schema. These typedefs begin with <type>sql_</type> and
    end with a lowercase version of the standard SQL type name,
    with spaces replaced by underscores. There are variants
    ending in <type>_null</type> that wrap these base types
    so they&#x2019;re <link linkend="sql-null">compatible with
    SQL null</link>. For instance, the SQL type <type>TINYINT
    UNSIGNED NOT NULL</type> is represented in MySQL++ by
    <classname>mysqlpp::sql_tinyint_unsigned</classname>. If you drop
    the <type>NOT NULL</type> part, the corresponding C++ type is
    <classname>mysqlpp::sql_tinyint_unsigned_null</classname>.</para>

    <para>MySQL++ doesn&#x2019;t force you to use these typedefs. It
    tries to be flexible with regard to data conversions,
    so you could probably use <type>int</type> anywhere you
    use <classname>mysqlpp::sql_tinyint_unsigned</classname>,
    for example. That said, the MySQL++ typedefs give several
    advantages:</para>

    <itemizedlist>
      <listitem><para>Space efficiency: the MySQL++ types are no
      larger than necessary to hold the MySQL data.</para></listitem>

      <listitem><para>Portability: if your program has to run on
      multiple different system types (even just 32- and 64-bit
      versions of the same operating system and processor type)
      using the MySQL++ typedefs insulates your code from platform
      changes.</para></listitem>

      <listitem><para>Clarity: using C++ types named similarly to the
      SQL types reduces the risk of confusion when working with code in
      both languages at the same time.</para></listitem>

      <listitem><para>Compatibility: using the MySQL++ types ensures
      that data conversions between SQL and C++ forms are compatible.
      Na&iuml;ve use of plain old C++ types can result in data
      truncation, <ulink type="classref" url="TypeLookupFailed"/>
      exceptions, and worse.</para>

      <para>Type compatibility is important not just at the time
      you write your program, it also helps forward compatibility:
      we occasionally change the definitions of the MySQL++
      typedefs to reduce the differences between the C++
      and SQL type systems. We&#x2019;ll be fixing the DECIMAL issue
      brought up above this way, for instance; if your program
      uses <classname>sql_decimal</classname> instead of the
      current underlying type, <type>double</type>, your program
      will pick up this improvement automatically with just a
      recompile.</para></listitem>
    </itemizedlist>

    <para>Most of these typedefs use standard C++ data types, but
    a few are aliases for a MySQL++ specific type. For instance,
    the SQL type <classname>DATETIME</classname> is mirrored in
    MySQL++ by <classname>mysqlpp::DateTime</classname>. For
    consistency, <filename>sql_types.h</filename> includes a
    typedef alias for <classname>DateTime</classname> called
    <classname>mysqlpp::sql_datetime</classname>.</para>

    <para>MySQL++ doesn&#x2019;t have typedefs for the most exotic data
    types, like those for the geospatial types. Patches to correct
    this will be thoughtfully considered.</para>
  </sect2>


  <sect2 id="sql-null">
    <title>Handling SQL Nulls</title>

    <para>Both C++ and SQL have things in them called NULL, but they
    differ in several ways. Consequently, MySQL++ has to provide
    special support for this, rather than just wrap native C++
    facilities as it can with most data type issues.</para>

    <sect3 id="sql-null-type">
      <title>SQL NULL is a type modifier</title>

      <para>The primary distinction is one of type. In SQL,
      &#x201C;NULL&#x201D; is a type modifier, which affects whether
      you can legally store a null value in that column. There&#x2019;s
      simply nothing like it in C++.</para>

      <para>To emulate SQL NULL, MySQL++ provides the <ulink
      type="classref" url="null">Null</ulink> template to allow
      the creation of distinct &#x201C;nullable&#x201D; versions of
      existing C++ types. So for example, if you have a <type>TINYINT
      UNSIGNED</type> column that can have nulls, the proper
      declaration for MySQL++ would be:</para>

      <programlisting>
mysqlpp::Null&lt;mysqlpp::sql_tinyint_unsigned&gt; myfield;</programlisting>

      <para>As of MySQL++ 3.1, we also provide shorter aliases for
      such types:</para>

      <programlisting>
mysqlpp::sql_tinyint_unsigned_null myfield;</programlisting>

      <para>These types are declared in
      <filename>lib/sql_types.h</filename>. You might want to scan
      through that to see what all is available.</para>

      <para>Template instantiations are first-class types in the C++
      language, so there&#x2019;s no possible confusion between this
      feature of MySQL++ and C++&#x2019;s native NULL concept.</para>
    </sect3>

    <sect3 id="sql-null-value">
      <title>SQL NULL is a unique value</title>

      <para>There&#x2019;s a secondary distinction between SQL null and
      anything available in the standard C++ type system: SQL null
      is a distinct value, equal to nothing else. We can&#x2019;t
      use C++&#x2019;s <symbol>NULL</symbol> for this because it
      is ambiguous, being equal to 0 in integer context. MySQL++
      provides the global <varname>null</varname> object, which you
      can assign to a <classname>Null</classname> template instance
      to make it equal to SQL null:</para>

      <programlisting>
myfield = mysqlpp::null;</programlisting>

      <para>If you insert a MySQL++ field holding a SQL null into a
      C++ IOstream, you get &#x201C;(NULL)&#x201D;, something fairly
      unlikely to be in a normal output string, thus reasonably
      preserving the uniqueness of the SQL null value.</para>

      <para>MySQL++ also tries to enforce the uniqueness of the
      SQL null value at compile time in assignments and data
      conversions. If you try to store a SQL null in a field type
      that isn&#x2019;t wrapped by <classname>Null</classname>
      or try to assign a <classname>Null</classname>-wrapped
      field value to a variable of the inner non-wrapped type,
      the compiler will emit some ugly error message, yelling about
      <type>CannotConvertNullToAnyOtherDataType</type>.  (The exact
      message is compiler-dependent.)</para>

      <para>If you don&#x2019;t like these behaviors, you can change
      them by passing a different value for the second parameter
      to template <classname>Null</classname>. By default, this
      parameter is <ulink type="structref" url="NullIsNull"/>,
      meaning that we should enforce the uniqueness of SQL
      null. To relax the distinctions, you can instantiate the
      <classname>Null</classname> template with a different behavior
      type: <ulink type="structref" url="NullIsZero"/> or <ulink
      type="structref" url="NullIsBlank"/>. Consider this code:</para>

      <programlisting>
mysqlpp::Null&lt;unsigned char, mysqlpp::NullIsZero&gt; myfield(mysqlpp::null);
cout &lt;&lt; myfield &lt;&lt; endl;
cout &lt;&lt; int(myfield) &lt;&lt; endl;</programlisting>

      <para>This will print &#x201C;0&#x201D; twice. If you had used the
      default for the second <classname>Null</classname> template
      parameter, the first output statement would have printed
      &#x201C;(NULL)&#x201D;, and the second wouldn&#x2019;t even
      compile.</para>
    </sect3>
  </sect2>


  <sect2 id="string-types">
    <title>MySQL++&#x2019;s Special String Types</title>

    <para>MySQL++ has two classes that work like
    <classname>std::string</classname> to some degree: <ulink
    type="classref" url="String"/> and <ulink type="classref"
    url="SQLTypeAdapter"/>. These classes exist to provide functionality
    that <classname>std::string</classname> doesn&#x2019;t provide, but
    they are neither derivatives of nor complete supersets of
    <classname>std::string</classname>.  As a result, end-user code
    generally doesn&#x2019;t deal with these classes directly, because
    <classname>std::string</classname> is a better general-purpose
    string type. In fact, MySQL++ itself uses
    <classname>std::string</classname> most of the time, too. But, the
    places these specialized stringish types do get used are so
    important to the way MySQL++ works that it&#x2019;s well worth taking
    the time to understand them.</para>


    <sect3 id="SQLTypeAdapter">
      <title>SQLTypeAdapter</title>

      <para>The simpler of the two is
      <classname>SQLTypeAdapter</classname>, or
      <classname>STA</classname> for short.<footnote><para>In version 2
      of MySQL++ and earlier, <classname>SQLTypeAdapter</classname> was
      called <classname>SQLString</classname>, but it was confusing
      because its name and the fact that it derived from
      <classname>std::string</classname> suggested that it was a
      general-purpose string type. MySQL++ even used it this way in a
      few places internally. In v3, we made it a simple base class and
      renamed it to reflect its proper limited
      function.</para></footnote></para>

      <para>As its name suggests, its only purpose is to adapt other
      data types to be used with SQL. It has a whole bunch of conversion
      constructors, one for all data types we expect to be used with
      MySQL++ for values in queries. SQL queries are strings, so
      constructors that take stringish types just make a copy of that
      string, and all the others &#x201C;stringize&#x201D; the value in
      the format needed by
      SQL.<footnote><para><classname>SQLTypeAdapter</classname>
      doesn&#x2019;t do <xref linkend="qescape"/> itself. That happens
      elsewhere, right at the point that the <classname>STA</classname>
      gets used to build a query.</para></footnote> The conversion
      constructors preserve type information, so this stringization
      process doesn&#x2019;t throw away any essential information.</para>

      <para><classname>STA</classname> is used anywhere MySQL++ needs to
      be able to accept any of several data types for use in a SQL
      query. Major users are <classname>Query</classname>&#x2019;s
      template query mechanism and the <classname>Query</classname>
      stream quoting and escaping mechanism. You care about
      <classname>STA</classname> because any time you pass a data value
      to MySQL++ to be used in building a SQL query, it goes through
      <classname>STA</classname>. <classname>STA</classname> is one of
      the key pieces in MySQL++ that makes it easy to generate
      syntactically-correct SQL queries.</para>
    </sect3>


    <sect3 id="String">
      <title>String</title>

      <para>If MySQL++ can be said to have its own generic string type,
      it&#x2019;s <classname>String</classname>, but it&#x2019;s not
      really functional enough for general use. It&#x2019;s possible that
      in future versions of MySQL++ we&#x2019;ll expand its interface to
      include everything <classname>std::string</classname> does, so
      that&#x2019;s why it&#x2019;s called that.<footnote><para>If you
      used MySQL++ before v3, <classname>String</classname> used to be
      called <classname>ColData</classname>. It was renamed because
      starting in v2.3, we began using it for holding more than just
      column data. I considered renaming it
      <classname>SQLString</classname> instead, but that would have
      confused old MySQL++ users to no end. Instead, I followed the
      example of <classname>Set</classname>, MySQL++&#x2019;s specialized
      <classname>std::set</classname> variant.</para></footnote></para>

      <para>The key thing <classname>String</classname> provides over
      <classname>std::string</classname> is conversion of strings in SQL
      value formats to their plain old C++ data types. For example, if you
      initialize it with the string &#x201C;2007-11-19&#x201D;, you can
      assign the <classname>String</classname> to a <ulink
      type="structref" url="Date">Date</ulink>, not because
      <classname>Date</classname> knows how to initialize itself from
      <classname>String</classname>, but the reverse:
      <classname>String</classname> has a bunch of implicit conversion
      operators defined for it, so you can use it in any type context
      that makes sense in your application.</para>

      <para>Because <methodname>Row::operator[]</methodname> returns
      <classname>String</classname>, you can say things like
      this:</para>

      <programlisting>int x = row["x"];</programlisting>

      <para>In a very real sense, <classname>String</classname> is the
      inverse of <classname>STA</classname>:
      <classname>String</classname> converts SQL value strings to C++
      data types, and <classname>STA</classname> converts C++ data types
      to SQL value strings.<footnote><para>During the development of
      MySQL++ v3.0, I tried merging
      <classname>SQLTypeAdapter</classname> and
      <classname>String</classname> into a single class to take
      advantage of this. The resulting class gave the C++ compiler the
      freedom to tie itself up in knots, because it was then allowed to
      convert almost any data type to almost any other. You&#x2019;d get
      a tangle of ambiguous data type conversion errors from the most
      innocent code.</para></footnote></para>

      <para><classname>String</classname> has two main uses.</para>

      <para>By far the most common use is as the field value type of
      <classname>Row</classname>, as exemplified above. It&#x2019;s not
      just the return type of <methodname>Row::operator[]</methodname>,
      though: it&#x2019;s actually the value type used within
      <classname>Row</classname>&#x2019;s internal array. As a result,
      any time MySQL++ pulls data from the database, it goes through
      <classname>String</classname> when converting it from the string
      form used in SQL result sets to the C++ data type you actually
      want the data in. It&#x2019;s the core of the structure population
      mechanism in <link linkend="ssqls">the SSQLS feature</link>, for
      example.</para>

      <para>Because <classname>String</classname> is the last pristine
      form of data in a result set before it gets out of MySQL++&#x2019;s
      internals where end-user code can see it, MySQL++&#x2019;s
      <type>sql_blob</type> and related <type>typedef</type>s are
      aliases for <classname>String</classname>. Using anything else
      would require copies; while the whole &#x201C;networked database
      server&#x201D; thing means most of MySQL++ can be quite inefficient
      and still not affect benchmark results meaningfully, BLOBs tend to
      be big, so making unnecessary copies can really make a difference.
      Which brings us to...</para>
    </sect3>


    <sect3 id="string-refcount">
      <title>Reference Counting</title>

      <para>To avoid unnecessary buffer copies, both
      <classname>STA</classname> and <classname>String</classname>
      are implemented in terms of a reference-counted copy-on-write
      buffer scheme. Both classes share the same underlying mechanism,
      and so are interoperable. This means that if you construct
      one of these objects from another, it doesn&#x2019;t actually
      copy the string data, it only copies a pointer to the data
      buffer, and increments its reference count. If the object
      has new data assigned to it or it&#x2019;s otherwise modified,
      it decrements its reference count and creates its own copy of
      the buffer. This has a lot of practical import, such as the
      fact that even though <methodname>Row::operator[]</methodname>
      returns <classname>String</classname>s by value, it&#x2019;s
      still efficient.</para>
    </sect3>
  </sect2>


  <sect2 id="blob">
    <title>Dealing with Binary Data</title>

    <para>Historically, there was no way to hold arbitrary-sized blocks
    of raw binary data in an SQL database. There was resistance to
    adding such a feature to SQL for a long time because it&#x2019;s
    better, where possible, to decompose blocks of raw binary data into
    a series of numbers and text strings that <emphasis>can</emphasis>
    be stored in the database. This lets you query, address and
    manipulate elements of the data block individually.</para>

    <para>A classic SQL newbie mistake is trying to treat the database
    server as a file system. Some embedded platforms use a database
    engine as a file system, but MySQL doesn&#x2019;t typically live
    in that world. When your platform already has a perfectly good
    file system, you should use it for big, nondecomposable blocks
    of binary data in most cases.</para>

    <para>A common example people use when discussing this is images
    in database-backed web applications. If you store the image in the
    database, you have to write code to retrieve the image from the
    database and send it to the client; there&#x2019;s more overhead,
    and less efficient use of the system&#x2019;s I/O caching system. If
    you store the image in the filesystem, all you have to do is
    point the web server to the directory where the images live,
    and put a URL for that image in your generated HTML. Because
    you&#x2019;re giving the web server a direct path to a file on
    disk, operation is far more efficient. Web servers are very
    good at slurping whole files off of disk and sending them out
    to the network, and operating systems are very good at caching
    file accesses. Plus, you avoid the overhead of pushing the data
    through the high-level language your web app is written in, which
    is typically an interpreted language, not C++. Some people still
    hold out on this, claiming that database engines have superior
    security features, but I call bunk on that, too. Operating systems
    and web servers are capable of building access control systems
    every bit as granular and secure as a database system.</para>

    <para>Occasionally you really do need to store a nondecomposable
    block of binary data in the database. For such cases, modern
    SQL database servers support BLOB data types, for Binary Large
    OBject. This is often just called binary data, though of course
    all data in a modern computer is binary at some level.</para>

    <para>The tricky part about dealing with binary data in MySQL++ is
    to ensure that you don&#x2019;t ever treat the data as a C string,
    which is really easy to do accidentally. C strings treat zero bytes
    as special end-of-string characters, but they&#x2019;re not special
    at all in binary data. We&#x2019;ve made a lot of improvements
    to the way MySQL++ handles <link linkend="string-types">string
    data</link> to avoid this problem, but it&#x2019;s still possible
    to bypass these features, wrecking your BLOBs. These examples
    demonstrate correct techniques.</para>


    <sect3 id="blob-save">
      <title>Loading a binary file into a BLOB column</title>

      <para>Above, I opined that it&#x2019;s usually incorrect to
      store image data in a database, particularly with web apps,
      of which CGI is a primitive form. Still, it makes a nice,
      simple example.</para>

      <para>Instead of a single example program, we have here a
      matched pair. The first example takes the name of a JPEG
      file on the command line along with all the other <link
      linkend="examples">common example program parameters</link>,
      loads that file into memory, and stores it in a BLOB column in
      the database.</para>

      <para>This example also demonstrates how to retrieve the
      value assigned to an auto-increment column in the previous
      insertion. This example uses that feature in the typical way,
      to create unique IDs for rows as they&#x2019;re inserted.</para>

      <para>Here is <filename>examples/load_jpeg.cpp</filename>:</para>

      <programlisting><xi:include href="load_jpeg.txt" parse="text"
      xmlns:xi="http://www.w3.org/2001/XInclude"/></programlisting>

      <para>Notice that we used the <type>escape</type> manipulator
      when building the INSERT query above. This is because
      <type>mysqlpp::sql_blob</type> is just an alias for one of the
      special MySQL++ <link linkend="string-types">string types</link>,
      which don&#x2019;t do automatic <link linkend="qescape">quoting
      and escaping</link>. They can&#x2019;t, because MySQL++ also
      uses these data types to hold raw SQL query strings, which
      would break due to doubled quoting and/or escaping if it were
      automatic.</para>
    </sect3>


    <sect3 id="blob-retreive">
      <title>Serving images from BLOB column via CGI</title>

      <para>The other example in this pair is rather short,
      considering how much it does. It parses a CGI query string
      giving the image ID, uses that to retreive data loaded into
      the database by <filename>load_jpeg</filename>, and writes
      it out in the form a web server wants when processing a CGI
      call, all with adequate real-world error handling. This is
      <filename>examples/cgi_jpeg.cpp</filename>:</para>

      <programlisting><xi:include href="cgi_jpeg.txt" parse="text"
      xmlns:xi="http://www.w3.org/2001/XInclude"/></programlisting>

      <para>While you can run it by hand, it&#x2019;s
      best to install this in a web server&#x2019;s
      CGI program directory, then call it with a URL like
      <filename>http://my.server.com/cgi-bin/cgi_jpeg?id=1</filename>.
      That retrieves the JPEG with ID 1 from the database and
      returns it to the web server, which will send it on to the
      browser.</para>

      <para>We&#x2019;ve included an image with MySQL++
      that you can use with this example pair,
      <filename>examples/logo.jpg</filename>.</para>
    </sect3>
  </sect2>


  <sect2 id="Transaction">
    <title>Using Transactions</title>

    <para>The <ulink type="classref" url="Transaction"/> class makes it
    easier to use SQL transactions in an exception-safe manner. Normally
    you create the <classname>Transaction</classname> object on the
    stack before you issue the queries in your transaction set. Then,
    when all the queries in the transaction set have been issued, you
    call <function>Transaction::commit()</function>, which commits the
    transaction set. If the <classname>Transaction</classname> object
    goes out of scope before you call <function>commit()</function>, the
    transaction set is rolled back. This ensures that if some code
    throws an exception after the transaction is started but before it
    is committed, the transaction isn&#x2019;t left unresolved.</para>

    <para><filename>examples/transaction.cpp</filename> illustrates
    this:</para>

    <programlisting><xi:include href="transaction.txt" parse="text"
    xmlns:xi="http://www.w3.org/2001/XInclude"/></programlisting>

    <para>One of the downsides of transactions is that the locking it
    requires in the database server is prone to deadlocks. The classic
    case where this happens is when two programs both want access to the
    same two rows within a single transaction each, but they modify them
    in opposite orders. If the timing is such that the programs
    interleave their lock acquisitions, the two come to an impasse:
    neither can get access to the other row they want to modify until
    the other program commits its transaction and thus release the row
    locks, but neither can finish the transaction because they&#x2019;re
    waiting on row locks the database server is holding on behalf of the
    other program.</para>

    <para>The MySQL server is smart enough to detect this condition, but
    the best it can do is abort the second transaction. This breaks the
    impasse, allowing the first program to complete its
    transaction.</para>

    <para>The second program now has to deal with the fact that its
    transaction just got aborted. There&#x2019;s a subtlety in detecting
    this situation when using MySQL++. By default, MySQL++ signals
    errors like these with exceptions. In the exception handler, you
    might expect to get <constant>ER_LOCK_DEADLOCK</constant> from
    <methodname>Query::errnum()</methodname> (or
    <methodname>Connection::errnum()</methodname>, same thing), but what
    you&#x2019;ll almost certainly get instead is 0, meaning &#x201C;no
    error.&#x201D; Why? It&#x2019;s because you&#x2019;re probably using a
    <classname>Transaction</classname> object to get automatic
    roll-backs in the face of exceptions. In this case, the roll-back
    happens before your exception handler is called by issuing a
    <command>ROLLBACK</command> query to the database server. Thus,
    <methodname>Query::errnum()</methodname> returns the error code
    associated with this roll-back query, not the deadlocked transaction
    that caused the exception.</para>

    <para>To avoid this problem, a few of the exception objects as of
    MySQL++ v3.0 include this last error number in the exception object
    itself. It&#x2019;s populated at the point of the exception, so it
    can differ from the value you would get from
    <methodname>Query::errnum()</methodname> later on when the exception
    handler runs.</para>

    <para>The example <filename>examples/deadlock.cpp</filename>
    demonstrates the problem:</para>

    <programlisting><xi:include href="deadlock.txt" parse="text"
    xmlns:xi="http://www.w3.org/2001/XInclude"/></programlisting>

    <para>This example works a little differently than the others.  You
    run one copy of the example, then when it pauses waiting for you to
    press <keycap>Enter</keycap>, you run another copy.  Then, depending
    on which one you press <keycap>Enter</keycap> in, one of the two
    will abort with the deadlock exception. You can see from the error
    message you get that it matters which method you call to get the
    error number. What you do about it is up to you as it depends on
    your program&#x2019;s design and system architecture.</para>
  </sect2>


  <sect2 id="querytypes">
    <title>Which Query Type to Use?</title>

    <para>There are three major ways to execute a query in MySQL++:
    <methodname>Query::execute()</methodname>,
    <methodname>Query::store()</methodname>, and
    <methodname>Query::use()</methodname>. Which should you use, and
    why?</para>

    <para><methodname>execute()</methodname> is for queries that do not
    return data <emphasis>per se</emphasis>. For instance,
    <command>CREATE INDEX</command>. You do get back some information
    from the MySQL server, which <methodname>execute()</methodname>
    returns to its caller in a <ulink type="classref"
    url="SimpleResult"/> object. In addition to the obvious &mdash; a
    flag stating whether the query succeeded or not &mdash; this object
    also contains things like the number of rows that the query
    affected. If you only need the success status, it&#x2019;s a little
    more efficient to call <methodname>Query::exec()</methodname>
    instead, as it simply returns <type>bool</type>.</para>

    <para>If your query does pull data from the database, the simplest
    option is <methodname>store()</methodname>. (All of the examples up
    to this point have used this method.)  This returns a <ulink
    type="classref" url="StoreQueryResult"/> object, which contains the
    entire result set. It&#x2019;s especially convenient because
    <classname>StoreQueryResult</classname> derives from
    <classname>std::vector&lt;mysqlpp::Row&gt;</classname>, so it opens
    the whole panoply of STL operations for accessing the rows in the
    result set. Access rows randomly with subscript notation, iterate
    forwards and backwards over the result set, run STL algorithms on
    the set...it all works naturally.</para>

    <para>If you like the idea of storing your results in an STL
    container but don&#x2019;t want to use
    <classname>std::vector</classname>, you can call
    <methodname>Query::storein()</methodname> instead. It lets you store
    the results in any standard STL container (yes, both sequential and
    set-associative types) instead of using
    <classname>StoreQueryResult</classname>. You do miss out on some of
    the additional database information held by
    <classname>StoreQueryResult</classname>&#x2019;s other base class,
    <ulink type="classref" url="ResultBase"/>, however.</para>

    <para><methodname>store*()</methodname> queries are convenient, but
    the cost of keeping the entire result set in main memory can
    sometimes be too high. It can be surprisingly costly, in fact. A
    MySQL database server stores data compactly on disk, but it returns
    query data to the client in a textual form. This results in a kind
    of data bloat that affects numeric and BLOB types the most. MySQL++
    and the underlying C API library also have their own memory
    overheads in addition to this. So, if you happen to know that the
    database server stores every record of a particular table in 1 KB,
    pulling a million records from that table could easily take several
    GB of memory with a <methodname>store()</methodname> query,
    depending on what&#x2019;s actually stored in that table.</para>

    <para>For these large result sets, the superior option is a
    <methodname>use()</methodname> query. This returns a <ulink
    type="classref" url="UseQueryResult"/> object, which is similar to
    <classname>StoreQueryResult</classname>, but without all of the
    random-access features. This is because a &#x201C;use&#x201D; query
    tells the database server to send the results back one row at a
    time, to be processed linearly. It&#x2019;s analogous to a C++
    stream&#x2019;s input iterator, as opposed to a random-access
    iterator that a container like vector offers. By accepting this
    limitation, you can process arbitrarily large result sets. This
    technique is demonstrated in
    <filename>examples/simple3.cpp</filename>:</para>

    <programlisting><xi:include href="simple3.txt" parse="text"
    xmlns:xi="http://www.w3.org/2001/XInclude"/></programlisting>

    <para>This example does the same thing as
    <filename>simple2</filename>, only with a &#x201C;use&#x201D; query
    instead of a &#x201C;store&#x201D; query.</para>

    <para>Valuable as <methodname>use()</methodname> queries are, they
    should not be the first resort in solving problems of excessive
    memory use. It&#x2019;s better if you can find a way to simply not
    pull as much data from the database in the first place. Maybe
    you&#x2019;re saying <command>SELECT *</command> even though you
    don&#x2019;t immedidately need all the columns from the table. Or,
    maybe you&#x2019;re filtering the result set with C++ code after you
    get it from the database server. If you can do that filtering with a
    more restrictive <command>WHERE</command> clause on the
    <command>SELECT</command>, it&#x2019;ll not only save memory,
    it&#x2019;ll save bandwidth between the database server and client,
    and can even save CPU time. If the filtering criteria can&#x2019;t be
    expressed in a <command>WHERE</command> clause, however, read on to
    the next section.</para>
  </sect2>


  <sect2 id="store_if">
    <title>Conditional Result Row Handling</title>

    <para>Sometimes you must pull more data from the database server
    than you actually need and filter it in memory. SQL&#x2019;s
    <command>WHERE</command> clause is powerful, but not as powerful as
    C++. Instead of storing the full result set and then picking over it
    to find the rows you want to keep, use
    <methodname>Query::store_if()</methodname>. This is
    <filename>examples/store_if.cpp</filename>:</para>

    <programlisting><xi:include href="store_if.txt" parse="text"
    xmlns:xi="http://www.w3.org/2001/XInclude"/></programlisting>

    <para>I doubt anyone really needs to select rows from a table that
    have a prime number in a given field. This example is meant to be
    just barely more complex than SQL can manage, to avoid obscuring the
    point. That point being, the
    <methodname>Query::store_if()</methodname> call here gives you a
    container full of results meeting a criterion that you probably
    can&#x2019;t express in SQL. You will no doubt have much more useful
    criteria in your own programs.</para>

    <para>If you need a more complex query than the one
    <methodname>store_if()</methodname> knows how to build when given an
    SSQLS examplar, there are two overloads that let you use your own
    query string. One overload takes the query string directly, and the
    other uses the query string built with
    <classname>Query</classname>&#x2019;s stream interface.</para>
  </sect2>


  <sect2 id="for_each">
    <title>Executing Code for Each Row In a Result Set</title>

    <para>SQL is more than just a database query language. Modern
    database engines can actually do some calculations on the data on
    the server side. But, this isn&#x2019;t always the best way to get
    something done. When you need to mix code and a query,
    MySQL++&#x2019;s <methodname>Query::for_each()</methodname> facility
    might be just what you need. This is
    <filename>examples/for_each.cpp</filename>:</para>

    <programlisting><xi:include href="for_each.txt" parse="text"
    xmlns:xi="http://www.w3.org/2001/XInclude"/></programlisting>

    <para>You only need to read the <function>main()</function> function
    to get a good idea of what the program does. The key line of code
    passes an SSQLS examplar and a functor to
    <methodname>Query::for_each()</methodname>.
    <methodname>for_each()</methodname> uses the SSQLS instance to build
    a <computeroutput>select * from TABLE</computeroutput> query,
    <computeroutput>stock</computeroutput> in this case. It runs that
    query internally, calling <classname>gather_stock_stats</classname>
    on each row. This is a pretty contrived example; you could actually
    do this in SQL, but we&#x2019;re trying to prevent the complexity of
    the code from getting in the way of the demonstration here.</para>

    <para>Just as with <methodname>store_if()</methodname>, described
    above, there are two other overloads for
    <methodname>for_each()</methodname> that let you use your own query
    string.</para>
  </sect2>


  <sect2 id="connopts" xreflabel="connection options">
    <title>Connection Options</title>

    <para>MySQL has a large number of options that control how it makes
    the connection to the database server, and how that connection
    behaves. The defaults are sufficient for most programs, so only one
    of the MySQL++ example programs make any connection option changes.
    Here is <filename>examples/multiquery.cpp</filename>:</para>

    <programlisting><xi:include href="multiquery.txt" parse="text"
    xmlns:xi="http://www.w3.org/2001/XInclude"/></programlisting>

    <para>This is a fairly complex example demonstrating the multi-query
    and stored procedure features in newer versions of MySQL. Because
    these are new features, and they change the communication between
    the client and server, you have to enable these features in a
    connection option. The key line is right up at the top of
    <function>main()</function>, where it creates a <ulink
    type="classref" url="MultiStatementsOption"/> object and passes it
    to <methodname>Connection::set_option()</methodname>. That method
    will take a pointer to any derivative of <ulink type="classref"
    url="Option"/>: you just create such an object on the heap and pass
    it in, which gives <classname>Connection</classname> the data values
    it needs to set the option. You don&#x2019;t need to worry about
    releasing the memory used by the <classname>Option</classname>
    objects; it&#x2019;s done automatically.</para>

    <para>The only tricky thing about setting options is that only a few
    of them can be set after the connection is up. Most need to be set
    just as shown in the example above: create an unconnected
    <classname>Connection</classname> object, set your connection
    options, and only then establish the connection. The option setting
    mechanism takes care of applying the options at the correct time in
    the connection establishment sequence.</para>

    <para>If you&#x2019;re familiar with setting connection options in
    the MySQL C API, you&#x2019;ll have to get your head around the fact
    that MySQL++&#x2019;s connection option mechanism is a much simpler,
    higher-level design that doesn&#x2019;t resemble the C API in any
    way. The C API has something like half a dozen different mechanisms
    for setting options that control the connection. The flexibility of
    the C++ type system allows us to wrap all of these up into a single
    high-level mechanism while actually getting greater type safety than
    the C API allows.</para>
  </sect2>


  <sect2 id="conn-timeout">
    <title>Dealing with Connection Timeouts</title>

    <para>By default, current MySQL servers have an 8 hour idle
    timeout on connections. This is not a problem if your program
    never has to run for more than 8 hours or reliably queries the
    database more often than that. And, it&#x2019;s a good thing for
    the database server, because even an idle connection takes up
    server resources.</para>

    <para>Many programs must run continually, however, and may
    experience long idle periods, such as nights and weekends
    when no one is around to make the program issue database
    queries. It&#x2019;s therefore common for people writing such
    programs to get a bug report from the field complaining that the
    program died overnight or over a long weekend, usually with some
    error message about the database server going away. They then check
    the DB server, find that it&#x2019;s still running and never did
    restart and scratch their heads wondering what happened. What
    happened is that the server&#x2019;s connection idle timeout
    expired, so it closed the connection to the client.</para>

    <para>You cannot detect this condition by calling
    <methodname>Connection::connected()</methodname>. When
    that returns <symbol>true</symbol>, it just means
    that either the connect-on-create constructor or the
    <methodname>connect()</methodname> call succeeded and that we
    haven&#x2019;t observed the connection to be down since then.
    When the database server closes an idle connection, you won&#x2019;t
    know it until after you try to issue a query. This is simply due
    to the nature of network programming.</para>

    <para>One way around this problem is to <ulink
    url="http://dev.mysql.com/doc/refman/5.0/en/gone-away.html">configure
    MySQL</ulink> to have a longer idle timeout. This timeout is
    in seconds, so the default of 8 hours is 28,800 seconds. You
    would want to figure out the longest possible time that your
    program could be left idle, then pick a value somewhat longer
    than that. For instance, you might decide that the longest
    reasonable idle time is a long 4-day weekend &mdash; 345,600
    seconds &mdash; which you could round up to 350,000 or 400,000
    to allow for a little bit of additional idle time on either end
    of that period.</para>

    <para>Another way around this, on a per-connection basis from
    the client side, would be to set the <ulink type="classref"
    url="ReconnectOption"/> <link linkend="connopts">connection
    option</link>. This will cause MySQL++ to reconnect to the server
    automatically if it drops the connection. Beware that unless
    you&#x2019;re using MySQL 5.1.6 or higher, you have to set this
    only after the connection is established, or it won&#x2019;t take
    effect. This means there&#x2019;s a potential race condition:
    it&#x2019;s possible the connection could drop shortly enough
    after being established that you don&#x2019;t have time to apply
    the option, so it won&#x2019;t come back up automatically. MySQL
    5.1.6+ fixes this by allowing this option to be set before the
    connection is established.</para>

    <para>A completely different way to tackle this, if your program
    doesn&#x2019;t block forever waiting on I/O while idle, is to
    periodically call <methodname>Connection::ping()</methodname>.

    <footnote>
      <para>Don&#x2019;t ping the server too often! It takes a tiny
      amount of processing capability to handle a ping, which can add
      up to a significant amount if done often enough by a client, or
      even just rarely by enough clients. Also, a lower ping frequency
      can let your program ride through some types of network faults
      &mdash; a switch reboot, for instance &mdash; without needing
      a reconnect. I like to ping the DB server no more often than
      half the connection timeout. With the default of 8 hours, then,
      I&#x2019;d ping between every 4 and 7 hours.</para>
    </footnote>

    This sends the smallest possible amount of data to the
    database server, which will reset its idle timer and cause
    it to respond, so <methodname>ping()</methodname> returns
    <symbol>true</symbol>. If it returns <symbol>false</symbol>
    instead, you know you need to reconnect to the server. Periodic
    pinging is easiest to do if your program uses asynchronous I/O,
    <link linkend="threads">threads</link>, or some kind of event
    loop to ensure that you can call something periodically even
    while the rest of the program has nothing to do.</para>

    <para>An interesting variant on this strategy is to ping the server
    before each query, or, better, before each group of queries within
    a larger operation. It has an advantage over pinging during idle
    time in that the client is about to use far more server resources
    to handle the query than it will take to handle the ping, so the
    ping time gets lost in the overhead. On the other hand, if the
    client issues queries frequently when not idle, it can result
    in a lot more pings than would happen if you just pinged every
    N hours while idle.</para>

    <para>Finally, some programmers prefer to wrap the querying
    mechanism in an error handler that catches the &#x201C;server has
    gone away&#x201D; error and tries to reestablish the connection and
    reissue the query. This adds some complexity, but it makes your
    program more robust without taking up unnecessary resources. If you
    did this, you could even change the server to drop idle connections
    more often, thus tying up fewer TCP/IP stack resources.</para>
  </sect2>


  <sect2 id="concurrentqueries">
    <title>Concurrent Queries on a Connection</title>

    <para>An important limitation of the MySQL C API library &mdash;
    which MySQL++ is built atop, so it shares this limitation &mdash;
    is that you can only have one query in progress on each connection
    to the database server. If you try to issue a second query while
    one is still in progress, you get an obscure error message about
    &#x201C;Commands out of sync&#x201D; from the underlying C API
    library. (You normally get this message in a MySQL++ exception
    unless you have exceptions disabled, in which case you get a
    failure code and <methodname>Connection::error()</methodname>
    returns this message.)</para>

    <para>There are lots of ways to run into this limitation:</para>

    <itemizedlist>
      <listitem>
        <para>The easiest way is to try to use a single <ulink
        type="classref" url="Connection"/> object in a multithreaded
        program, with more than one thread attempting to use it to
        issue queries.  Unless you put in a lot of work to synchronize
        access, this is almost guaranteed to fail at some point, giving
        the dread &#x201C;Commands out of sync&#x201D; error.</para>
      </listitem>

      <listitem>
        <para>You might then think to give each thread that issues
        queries its own <classname>Connection</classname> object.
        You can still run into trouble if you pass the data you get
        from queries around to other threads. What can happen is
        that one of these child objects indirectly calls back to the
        <classname>Connection</classname> at a time where it&#x2019;s
        involved with another query. This is properly covered
        elsewhere, in <xref linkend="thread-data-sharing"/>.)</para>
      </listitem>

      <listitem>
        <para>One way to run into this problem without using
        threads is with &#x201C;use&#x201D; queries, discussed <link
        linkend="querytypes">above</link>. If you don&#x2019;t
        consume all rows from a query before you issue another on
        that connection, you are effectively trying to have multiple
        concurrent queries on a single connection. Here&#x2019;s a
        recipie for this particular disaster:</para>

        <programlisting>
UseQueryResult r1 = query.use("select garbage from plink where foobie='tamagotchi'");
UseQueryResult r2 = query.use("select blah from bonk where bletch='smurf'");</programlisting>

        <para>The second <methodname>use()</methodname> call fails because
        the first result set hasn&#x2019;t been consumed yet.</para>
      </listitem>

      <listitem>
        <para>Still another way to run into this limitation
        is if you use MySQL&#x2019;s multi-query feature. This
        lets you give multiple queries in a single call,
        separated by semicolons, and get back the results for
        each query separately. If you issue three queries using
        <methodname>Query::store()</methodname>, you only get
        back the first query&#x2019;s results with that call, and
        then have to call <methodname>store_next()</methodname>
        to get the subsequent query results.  MySQL++ provides
        <methodname>Query::more_results()</methodname> so
        you know whether you&#x2019;re done, or need to call
        <methodname>store_next()</methodname> again.  Until you reach
        the last result set, you can&#x2019;t issue another query on
        that connection.</para>
      </listitem>

      <listitem>
        <para>Finally, there&#x2019;s a way to run into this
        that surprises almost everyone sooner or later: stored
        procedures. MySQL normally returns <emphasis>at least
        two</emphasis> result sets for a stored procedure call. The
        simple case is that the stored procedure contains a single
        SQL query, and it succeeds: you get two results, first the
        results of the embedded SQL query, and then the result
        of the call itself. If there are multiple SQL queries
        within the stored procedure, you get more than two result
        sets. Until you consume them all, you can&#x2019;t start a
        new query on the connection. As above, you want to have
        a loop calling <methodname>more_results()</methodname>
        and <methodname>store_next()</methodname> to work your
        way through all of the result sets produced by the stored
        procedure call.</para>
      </listitem>
    </itemizedlist>
  </sect2>


  <sect2 id="fieldinf">
    <title>Getting Field Meta-Information</title>

    <para>The following example demonstrates how to get information
    about the fields in a result set, such as the name of the field and
    the SQL type. This is
    <filename>examples/fieldinf.cpp</filename>:</para>

    <programlisting><xi:include href="fieldinf.txt" parse="text"
    xmlns:xi="http://www.w3.org/2001/XInclude"/></programlisting>
  </sect2>
</sect1>