repo_zf1/documentation/manual/en/ref/performance-classloading.xml

<?xml version="1.0" encoding="UTF-8"?>
<!-- Reviewed: no -->
<sect1 id="performance.classloading">
    <title>Class Loading</title>

    <para>
        Anyone who ever performs profiling of a Zend Framework application will
        immediately recognize that class loading is relatively expensive in Zend
        Framework. Between the sheer number of class files that need to be
        loaded for many components, to the use of plugins that do not have a 1:1
        relationship between their class name and the file system, the various
        calls to <methodname>include_once()</methodname> and
        <methodname>require_once()</methodname> can be problematic. This chapter intends to provide
        some concrete solutions to these issues.
    </para>

    <sect2 id="performance.classloading.includepath">
        <title>How can I optimize my include_path?</title>

        <para>
            One trivial optimization you can do to increase the speed of class
            loading is to pay careful attention to your include_path. In
            particular, you should do four things: use absolute paths (or paths
            relative to absolute paths), reduce the number of include paths you
            define, have your Zend Framework include_path as early as possible,
            and only include the current directory path at the end of your
            include_path.
        </para>

        <sect3 id="performance.classloading.includepath.abspath">
            <title>Use absolute paths</title>

            <para>
                While this may seem a micro-optimization, the fact is that if
                you don't, you'll get very little benefit from <acronym>PHP</acronym>'s realpath
                cache, and as a result, opcode caching will not perform nearly
                as you may expect.
            </para>

            <para>
                There are two easy ways to ensure this. First, you can hardcode
                the paths in your <filename>php.ini</filename>, <filename>httpd.conf</filename>,
                or <filename>.htaccess</filename>. Second, you
                can use <acronym>PHP</acronym>'s <methodname>realpath()</methodname> function when
                setting your include_path:
            </para>

            <programlisting language="php"><![CDATA[
$paths = array(
    realpath(dirname(__FILE__) . '/../library'),
    '.',
);
set_include_path(implode(PATH_SEPARATOR, $paths);
]]></programlisting>

            <para>
                You <emphasis>can</emphasis> use relative paths -- so long as
                they are relative to an absolute path:
            </para>

            <programlisting language="php"><![CDATA[
define('APPLICATION_PATH', realpath(dirname(__FILE__)));
$paths = array(
    APPLICATION_PATH . '/../library'),
    '.',
);
set_include_path(implode(PATH_SEPARATOR, $paths);
]]></programlisting>

            <para>
                However, even so, it's typically a trivial task to simply pass
                the path to <methodname>realpath()</methodname>.
            </para>
        </sect3>

        <sect3 id="performance.classloading.includepath.reduce">
            <title>Reduce the number of include paths you define</title>

            <para>
                Include paths are scanned in the order in which they appear in
                the include_path. Obviously, this means that you'll get a result
                faster if the file is found on the first scan rather than the
                last. Thus, a rather obvious enhancement is to simply reduce the
                number of paths in your include_path to only what you need. Look
                through each include_path you've defined, and determine if you
                actually have any functionality in that path that is used in
                your application; if not, remove it.
            </para>

            <para>
                Another optimization is to combine paths. For instance, Zend
                Framework follows <acronym>PEAR</acronym> naming conventions; thus, if you are
                using <acronym>PEAR</acronym> libraries (or libraries from another framework or
                component library that follows <acronym>PEAR</acronym> CS), try to put all of these
                libraries on the same include_path. This can often be achieved
                by something as simple as symlinking one or more libraries into
                a common directory.
            </para>
        </sect3>

        <sect3 id="performance.classloading.includepath.early">
            <title>Define your Zend Framework include_path as early as possible</title>

            <para>
                Continuing from the previous suggestion, another obvious
                optimization is to define your Zend Framework include_path as
                early as possible in your include_path. In most cases, it should
                be the first path in the list. This ensures that files included
                from Zend Framework are found on the first scan.
            </para>
        </sect3>

        <sect3 id="performance.classloading.includepath.currentdir">
            <title>Define the current directory last, or not at all</title>

            <para>
                Most include_path examples show using the current directory, or
                '.'. This is convenient for ensuring that scripts in the same
                directory as the file requiring them can be loaded. However,
                these same examples typically show this path item as the first
                item in the include_path -- which means that the current
                directory tree is always scanned first. In most cases, with Zend
                Framework applications, this is not desired, and the path may be
                safely pushed to the last item in the list.
            </para>

            <example id="performance.classloading.includepath.example">
                <title>Example: Optimized include_path</title>

                <para>
                    Let's put all of these suggestions together. Our assumption will be that you
                    are using one or more <acronym>PEAR</acronym> libraries in conjunction with
                    Zend Framework -- perhaps the PHPUnit and <classname>Archive_Tar</classname>
                    libraries -- and that you occasionally need to include
                    files relative to the current file.
                </para>

                <para>
                    First, we'll create a library directory in our project. Inside that
                    directory, we'll symlink our Zend Framework's <filename>library/Zend</filename>
                    directory, as well as the necessary directories from our <acronym>PEAR</acronym>
                    installation:
                </para>

                <programlisting language="php"><![CDATA[
library
    Archive/
    PEAR/
    PHPUnit/
    Zend/
]]></programlisting>

                <para>
                    This allows us to add our own library code if necessary, while
                    keeping shared libraries intact.
                </para>

                <para>
                    Next, we'll opt to create our include_path programmatically
                    within our <filename>public/index.php</filename> file. This allows us to move
                    our code around on the file system, without needing to edit the
                    include_path every time.
                </para>

                <para>
                    We'll borrow ideas from each of the suggestions above: we'll use
                    absolute paths, as determined using <methodname>realpath()</methodname>;
                    we'll include Zend Framework's include path early; we've
                    already consolidated include_paths; and we'll put the current
                    directory as the last path. In fact, we're doing really well
                    here -- we're going to end up with only two paths.
                </para>

                <programlisting language="php"><![CDATA[
$paths = array(
    realpath(dirname(__FILE__) . '/../library'),
    '.'
);
set_include_path(implode(PATH_SEPARATOR, $paths));
]]></programlisting>
            </example>
        </sect3>
    </sect2>

    <sect2 id="performance.classloading.striprequires">
        <title>How can I eliminate unnecessary require_once statements?</title>

        <para>
            Lazy loading is an optimization technique designed to push the
            expensive operation of loading a class file until the last possible
            moment -- i.e., when instantiating an object of that class, calling
            a static class method, or referencing a class constant or static
            property. <acronym>PHP</acronym> supports this via autoloading, which allows you to
            define one or more callbacks to execute in order to map a class name
            to a file.
        </para>

        <para>
            However, most benefits you may reap from autoloading are negated if
            your library code is still performing <methodname>require_once()</methodname> calls --
            which is precisely the case with Zend Framework. So, the question is: how can
            you eliminate those <methodname>require_once()</methodname> calls in order to maximize
            autoloader performance?
        </para>

        <sect3 id="performance.classloading.striprequires.sed">
            <title>Strip require_once calls with find and sed</title>

            <para>
                An easy way to strip <methodname>require_once()</methodname> calls is to use the
                <acronym>UNIX</acronym> utilities 'find' and 'sed' in conjunction to comment out
                each call. Try executing the following statements (where '%'
                indicates the shell prompt):
            </para>

            <programlisting language="shell"><![CDATA[
% cd path/to/ZendFramework/library
% find . -name '*.php' -not -wholename '*/Loader/Autoloader.php' \
  -not -wholename '*/Application.php' -print0 | \
  xargs -0 sed --regexp-extended --in-place 's/(require_once)/\/\/ \1/g'
]]></programlisting>

            <para>
                This one-liner (broken into two lines for readability) iterates
                through each <acronym>PHP</acronym> file and tells it to replace each instance of
                'require_once' with '// require_once', effectively commenting
                out each such statement. (It selectively keeps
                <methodname>require_once()</methodname> calls within
                <classname>Zend_Application</classname> and
                <classname>Zend_Loader_Autoloader</classname>, as these classes will fail without
                them.)
            </para>

            <para>
                This command could be added to an automated build or release
                process trivially, helping boost performance in your production
                application. It should be noted, however, that if you use this
                technique, you <emphasis>must</emphasis> utilize autoloading;
                you can do that from your "<filename>public/index.php</filename>" file with the
                following code:
            </para>

            <programlisting language="php"><![CDATA[
require_once 'Zend/Loader/Autoloader.php';
Zend_Loader_Autoloader::getInstance();
]]></programlisting>
        </sect3>
    </sect2>

    <sect2 id="performance.classloading.pluginloader">
        <title>How can I speed up plugin loading?</title>

        <para>
            Many components have plugins, which allow you to create your own
            classes to utilize with the component, as well as to override
            existing, standard plugins shipped with Zend Framework. This
            provides important flexibility to the framework, but at a price:
            plugin loading is a fairly expensive task.
        </para>

        <para>
            The plugin loader allows you to register class prefix / path pairs,
            allowing you to specify class files in non-standard paths. Each
            prefix can have multiple paths associated with it.
            Internally, the plugin loader loops through each prefix, and then
            through each path attached to it, testing to see if the file exists
            and is readable on that path. It then loads it, and tests to see
            that the class it is looking for is available. As you might imagine,
            this can lead to many stat calls on the file system.
        </para>

        <para>
            Multiply this by the number of components that use the PluginLoader,
            and you get an idea of the scope of this issue. At the time of this
            writing, the following components made use of the PluginLoader:
        </para>

        <itemizedlist>
            <listitem>
                <para>
                    <classname>Zend_Controller_Action_HelperBroker</classname>: helpers
                </para>
            </listitem>

            <listitem>
                <para>
                    <classname>Zend_Dojo</classname>: view helpers, form elements and decorators
                </para>
            </listitem>

            <listitem>
                <para>
                    <classname>Zend_File_Transfer</classname>: adapters
                </para>
            </listitem>

            <listitem>
                <para>
                    <classname>Zend_Filter_Inflector</classname>: filters (used by the
                    ViewRenderer action helper and <classname>Zend_Layout</classname>)
                </para>
            </listitem>

            <listitem>
                <para>
                    <classname>Zend_Filter_Input</classname>: filters and validators
                </para>
            </listitem>

            <listitem>
                <para>
                    <classname>Zend_Form</classname>: elements, validators, filters,
                    decorators, captcha and file transfer adapters
                </para>
            </listitem>

            <listitem>
                <para>
                    <classname>Zend_Paginator</classname>: adapters
                </para>
            </listitem>

            <listitem>
                <para>
                    <classname>Zend_View</classname>: helpers, filters
                </para>
            </listitem>
        </itemizedlist>

        <para>
            How can you reduce the number of such calls made?
        </para>

        <sect3
            id="performance.classloading.pluginloader.includefilecache">
            <title>Use the PluginLoader include file cache</title>

            <para>
                Zend Framework 1.7.0 adds an include file cache to the
                PluginLoader. This functionality writes "<methodname>include_once()</methodname>"
                calls to a file, which you can then include in your bootstrap. While this
                introduces extra <methodname>include_once()</methodname> calls to your code, it
                also ensures that the PluginLoader returns as early as possible.
            </para>

            <para>
                The PluginLoader documentation <link
                    linkend="zend.loader.pluginloader.performance.example">includes
                a complete example of its use</link>.
            </para>
        </sect3>
    </sect2>
</sect1>
<!--
vim:se ts=4 sw=4 et:
-->