Recently, I've been working on the search interface for McGill University's course catalog. The University wants to allow students to browse courses at friendly URLs like:

  • arts/undergraduate/courses
  • science/undergraduate/courses
  • arts/graduate/courses

Instead of unreadable URLs like:

  • search/apachesolr_search/?filters=type%3Acatalog%20ss_faculty%3AAR%20sm_level%3AUndergraduate
  • search/apachesolr_search/?filters=type%3Acatalog%20ss_faculty%3ASC%20sm_level%3AUndergraduate
  • search/apachesolr_search/?filters=type%3Acatalog%20ss_faculty%3AAR%20sm_level%3AGraduate

First, let's use hook_menu to define a new search path, "arts/undergraduate/courses", with the page callback mcgill_courses_search:

function mcgill_menu() {
  $items['arts/undergraduate/courses'] = array(
    'page callback' => 'mcgill_courses_search',
    'access arguments' => array('search content'),
    'type' => MENU_CALLBACK,
  );
  return $items;
}

Now, let's define that page callback. mcgill_courses_search will be a copy of apachesolr_search_view from apachesolr_search.module, with two important modifications: it will replace search_get_keys with mcgill_get_keys, and it will replace search_data($keys, $type) with search_data($keys, 'mcgill'). A minor modification is setting $type to "apachesolr_search" in the function signature; this is just to get the page callback working with the fewest modifications. Once we define the page callback here, we'll have a closer look at search_get_keys and search_data.

// START MINOR MODIFICATION
function mcgill_courses_search($type = 'apachesolr_search') {
// END
  $results = '';
  if (!isset($_POST['form_id'])) {
    if (empty($type)) {
      drupal_goto('search/apachesolr_search');
    }
    // START MODIFICATION
    $keys = trim(mcgill_get_keys());
    // END
    $filters = '';
    if ($type == 'apachesolr_search' && isset($_GET['filters'])) {
      $filters = trim($_GET['filters']);
    }
    if ($keys || $filters) {
      // Log the search keys:
      $log = $keys;
      if ($filters) {
        $log .= 'filters='. $filters;
      }
      watchdog(
        'search',
        '%keys (@type).', array(
          '%keys' => $log,
          '@type' => t('Search')
        ),
        WATCHDOG_NOTICE,
        l(t('results'), 'search/'. $type .'/'. $keys));

      // START MODIFICATION
      $results = search_data($keys, 'mcgill');
      // END

      if ($results) {
        $results = theme('box', t('Search results'), $results);
      }
      else {
        $results = theme('box', t('Your search yielded no results'),
          variable_get('apachesolr_search_noresults',
            apachesolr_search_noresults()));
      }
    }
  }
  return drupal_get_form('search_form', NULL, $keys, $type) .
    $results;
}

The above page callback looks big and scary, but, remember, we're just changing two lines from apachesolr_search_view.


So, why don't we use Drupal's search_get_keys? The reason is, search_get_keys assumes your search path looks like "foo/bar/keywords", and that's not the case here. So, in mcgill_courses_search, we replaced search_get_keys with mcgill_get_keys. Let's implement mcgill_get_keys, and then explain the code:

function mcgill_get_keys() {
  static $return;
  if (!isset($return)) {
    $parts = explode('/', $_GET['q']);
    if (count($parts) == 4) {
      $return = array_pop($parts);
    }
    else {
      $return = empty($_REQUEST['keys']) ? '' : $_REQUEST['keys'];
    }
  }
  return $return;
}

mcgill_get_keys inspects the path. If it has four parts - for example, "arts/undergraduate/courses/english" - it will extract the last part as the keywords - in this case, "english". If it has fewer parts, it will behave like search_get_keys, in most cases returning the empty string. Keep in mind that, if we add more search paths, we may need to tweak mcgill_get_keys above.


In apachesolr_search_view, if $type is "apachesolr_search", search_data will invoke apachesolr_search_search. What's wrong with that? The problem is, apachesolr_search_search assumes your search path looks like "search/something/keywords", and that's not the case here. So, in mcgill_courses_search, we force search_data to invoke mcgill_search by passing it "mcgill" as an argument. Let's implement mcgill_search, which will be a copy of apachesolr_search_search from apachesolr_search.module, with one modification:

function mcgill_search($op = 'search', $keys = NULL) {
  switch ($op) {
    case 'name':
      return t('Search');

    case 'reset':
      apachesolr_clear_last_index('apachesolr_search');
      return;

    case 'status':
      return apachesolr_index_status('apachesolr_search');

    case 'search':
      $filters = isset($_GET['filters']) ? $_GET['filters'] : '';
      $solrsort = isset($_GET['solrsort']) ? $_GET['solrsort'] : '';
      $page = isset($_GET['page']) ? $_GET['page'] : 0;
      try {
        $results = apachesolr_search_execute(
          $keys,
          $filters,
          $solrsort,
          // START MODIFICATION
          'arts/undergraduate/courses', $page);
          // END
        return $results;
      }
      catch (Exception $e) {
        watchdog(
          'Apache Solr',
          nl2br(check_plain($e->getMessage())),
          NULL, WATCHDOG_ERROR);
        apachesolr_failure(t('Solr search'), $keys);
      }
      break;
  } // switch
}

The one modification is that mcgill_search replaces 'search/' . arg(1) with "arts/undergraduate/courses". Of course, if we add more search paths, we will have to modify this line. For now, let's keep things simple.

At this point, we have all we need for the search to run at the custom search path. We will of course want to add default filters based on the path. For example, if the path is "arts/undergraduate/courses", we want to filter the list of courses down to those within the Arts faculty at the undergraduate level. For that, come to my session at DrupalCon SF!