Extending DQL in Doctrine 2: Custom AST Walkers

Section author: Benjamin Eberlei <kontakt@beberlei.de>

The Doctrine Query Language (DQL) is a proprietary sql-dialect that substitutes tables and columns for Entity names and their fields. Using DQL you write a query against the database using your entities. With the help of the metadata you can write very concise, compact and powerful queries that are then translated into SQL by the Doctrine ORM.

In Doctrine 1 the DQL language was not implemented using a real parser. This made modifications of the DQL by the user impossible. Doctrine 2 in contrast has a real parser for the DQL language, which transforms the DQL statement into an Abstract Syntax Tree and generates the appropriate SQL statement for it. Since this process is deterministic Doctrine heavily caches the SQL that is generated from any given DQL query, which reduces the performance overhead of the parsing process to zero.

You can modify the Abstract syntax tree by hooking into DQL parsing process by adding a Custom Tree Walker. A walker is an interface that walks each node of the Abstract syntax tree, thereby generating the SQL statement.

There are two types of custom tree walkers that you can hook into the DQL parser:

  • An output walker. This one actually generates the SQL, and there is only ever one of them. We implemented the default SqlWalker implementation for it.
  • A tree walker. There can be many tree walkers, they cannot generate the sql, however they can modify the AST before its rendered to sql.

Now this is all awfully technical, so let me come to some use-cases fast to keep you motivated. Using walker implementation you can for example:

  • Modify the AST to generate a Count Query to be used with a paginator for any given DQL query.
  • Modify the Output Walker to generate vendor-specific SQL (instead of ANSI).
  • Modify the AST to add additional where clauses for specific entities (example ACL, country-specific content...)
  • Modify the Output walker to pretty print the SQL for debugging purposes.

In this cookbook-entry I will show examples on the first two points. There are probably much more use-cases.

Generic count query for pagination

Say you have a blog and posts all with one category and one author. A query for the front-page or any archive page might look something like:

SELECT p, c, a FROM BlogPost p JOIN p.category c JOIN p.author a WHERE ...

Now in this query the blog post is the root entity, meaning its the one that is hydrated directly from the query and returned as an array of blog posts. In contrast the comment and author are loaded for deeper use in the object tree.

A pagination for this query would want to approximate the number of posts that match the WHERE clause of this query to be able to predict the number of pages to show to the user. A draft of the DQL query for pagination would look like:

SELECT count(DISTINCT p.id) FROM BlogPost p JOIN p.category c JOIN p.author a WHERE ...

Now you could go and write each of these queries by hand, or you can use a tree walker to modify the AST for you. Lets see how the API would look for this use-case:

<?php
$pageNum = 1;
$query = $em->createQuery($dql);
$query->setFirstResult( ($pageNum-1) * 20)->setMaxResults(20);

$totalResults = Paginate::count($query);
$results = $query->getResult();

The Paginate::count(Query $query) looks like:

<?php
class Paginate
{
    static public function count(Query $query)
    {
        /* @var $countQuery Query */
        $countQuery = clone $query;

        $countQuery->setHint(Query::HINT_CUSTOM_TREE_WALKERS, array('DoctrineExtensions\Paginate\CountSqlWalker'));
        $countQuery->setFirstResult(null)->setMaxResults(null);

        return $countQuery->getSingleScalarResult();
    }
}

It clones the query, resets the limit clause first and max results and registers the CountSqlWalker customer tree walker which will modify the AST to execute a count query. The walkers implementation is:

<?php
class CountSqlWalker extends TreeWalkerAdapter
{
    /**
     * Walks down a SelectStatement AST node, thereby generating the appropriate SQL.
     *
     * @return string The SQL.
     */
    public function walkSelectStatement(SelectStatement $AST)
    {
        $parent = null;
        $parentName = null;
        foreach ($this->_getQueryComponents() AS $dqlAlias => $qComp) {
            if ($qComp['parent'] === null && $qComp['nestingLevel'] == 0) {
                $parent = $qComp;
                $parentName = $dqlAlias;
                break;
            }
        }

        $pathExpression = new PathExpression(
            PathExpression::TYPE_STATE_FIELD | PathExpression::TYPE_SINGLE_VALUED_ASSOCIATION, $parentName,
            $parent['metadata']->getSingleIdentifierFieldName()
        );
        $pathExpression->type = PathExpression::TYPE_STATE_FIELD;

        $AST->selectClause->selectExpressions = array(
            new SelectExpression(
                new AggregateExpression('count', $pathExpression, true), null
            )
        );
    }
}

This will delete any given select expressions and replace them with a distinct count query for the root entities primary key. This will only work if your entity has only one identifier field (composite keys won’t work).

Modify the Output Walker to generate Vendor specific SQL

Most RMDBS have vendor-specific features for optimizing select query execution plans. You can write your own output walker to introduce certain keywords using the Query Hint API. A query hint can be set via Query::setHint($name, $value) as shown in the previous example with the HINT_CUSTOM_TREE_WALKERS query hint.

We will implement a custom Output Walker that allows to specify the SQL_NO_CACHE query hint.

<?php
$dql = "SELECT p, c, a FROM BlogPost p JOIN p.category c JOIN p.author a WHERE ...";
$query = $m->createQuery($dql);
$query->setHint(Query::HINT_CUSTOM_OUTPUT_WALKER, 'DoctrineExtensions\Query\MysqlWalker');
$query->setHint("mysqlWalker.sqlNoCache", true);
$results = $query->getResult();

Our MysqlWalker will extend the default SqlWalker. We will modify the generation of the SELECT clause, adding the SQL_NO_CACHE on those queries that need it:

<?php
class MysqlWalker extends SqlWalker
{
     /**
     * Walks down a SelectClause AST node, thereby generating the appropriate SQL.
     *
     * @param $selectClause
     * @return string The SQL.
     */
    public function walkSelectClause($selectClause)
    {
        $sql = parent::walkSelectClause($selectClause);

        if ($this->getQuery()->getHint('mysqlWalker.sqlNoCache') === true) {
            if ($selectClause->isDistinct) {
                $sql = str_replace('SELECT DISTINCT', 'SELECT DISTINCT SQL_NO_CACHE', $sql);
            } else {
                $sql = str_replace('SELECT', 'SELECT SQL_NO_CACHE', $sql);
            }
        }

        return $sql;
    }
}

Writing extensions to the Output Walker requires a very deep understanding of the DQL Parser and Walkers, but may offer your huge benefits with using vendor specific features. This would still allow you write DQL queries instead of NativeQueries to make use of vendor specific features.




Fork me on GitHub