Search

    The unified search combines a variable number of search providers into a unified search result for the user. To improve the user experience with search, the search results should be displayed quickly. Therefore parallelism is used to split the process into several requests that can be processed concurrently, to give the client (e.g. JavaScript in the browser) the ability to display partial search results as they come on.

    Hence the search process consists of two steps.

    These two steps have to be run consecutively, but the individual requests in the second step can be dispatched and processed concurrently.

    This will return a structure like

    GET https://cloud.domain/ocs/v2.php/search/providers/files/search?term=cat

    1. {
    2.     "ocs": {
    3.         "meta": {
    4.             …
    5.         },
    6.         "data": {
    7.             "name": "Files",
    8.             "isPaginated": false,
    9.             "entries": [
    10.                 {
    11.                     "thumbnailUrl": "/core/preview?x=32&y=32&fileId=9261",
    12.                     "title": "my cute cats.jpg",
    13.                     "subline": "/my cute cats.jpg",
    14.                     "resourceUrl": "/apps/files/?dir=/&scrollto=my%20cute%20cats.jpg"
    15.                 },
    16.                 {
    17.                     "thumbnailUrl": "/core/preview?x=32&y=32&fileId=1553",
    18.                     "title": "cat (2).png",
    19.                     "subline": "/cat (2).png",
    20.                     "resourceUrl": "/apps/files/?dir=/&scrollto=cat%20%282%29.png"
    21.                 }
    22.             ],
    23.             "cursor": null
    24.         }
    25.     }
    26. }

    A search provider is a class the implements the interface \OCP\Search\IProvider.

    The method getId returns a string identifier of the registered provider. It has to be globally unique, hence must not conflict with any other apps. Therefore it’s advised to use just the app ID (e.g. mail) as ID or an ID that is prefixed with the app id, like mail_recipients. getName is a translated name for your search results.

    The method search transforms a search request into a search result.

    The class would typically be saved into a file in lib/Search of your app but you are free to put it elsewhere as long as it’s loadable by Nextcloud’s dependency injection container.

    The provider class is registered via the of the Application class.

    1. <?php
    3. declare(strict_types=1);
    5. namespace OCA\MyApp\AppInfo;
    6. use OCA\MyApp\Search\Provider;
    7. use OCP\AppFramework\App;
    9. use OCP\AppFramework\Bootstrap\IBootstrap;
    10. use OCP\AppFramework\Bootstrap\IRegistrationContext;
    12. class Application extends App implements IBootstrap {
    14.     public function register(IRegistrationContext $context): void {
    15.         $context->registerSearchProvider(Provider::class);
    16.     }
    18.     public function boot(IBootContext $context): void {}
    20. }

    Search requests are processed in the search method. The $user object is the user who the result shall be generated for. $query gives context information like the search term, the sort order, the route information, the size limit of a request and the cursor for follow-up request of paginated results.

    The result is encapsulated in the SearchResult class that offers two static factory methods complete and paginated. Both of these methods take an array of SearchResultEntry objects.

    Next, you’ll see a dummy provider that returns a static set of results.

    Each of the result result entries has

    • A thumbnail or icon that is a (relative) URL
    • A title, e.g. the name of a file
    • A subline, e.g. the path to a file
    • A resource URL that makes it possible to navigate to the details of this result
    • Optional icon CSS class that is applied then the thumbnail URL was not set
    • A boolean rounded, whether the thumbnail should be rounded, e.g. when it’s an avatar

    Apps may return the full result in search, but in most cases the size of the result set can become too big to fit into one HTTP request and is complicated to display to the user, hence the set should be split into chunks – it should be paginated.

    There are two ways to use the cursor: offset-based pagination and cursor-based pagination.

    For offset-based pagination you return $query->getLimit() results and specify this number as cursor. Any subsequent call where $query->getCursor() does not return null you take the value as offset for the next page. The following example shall demonstrate this use case.

    1. <?php
    3. declare(strict_types=1);
    5. namespace OCA\MyApp\Search;
    7. use OCA\MyApp\AppInfo\Application;
    8. use OCP\IL10N;
    9. use OCP\IURLGenerator;
    10. use OCP\IUser;
    11. use OCP\Search\IProvider;
    12. use OCP\Search\SearchResult;
    14. class Provider implements IProvider {
    16.     /** @var IL10N */
    17.     private $l10n;
    19.     /** @var IURLGenerator */
    20.     private $urlGenerator;
    21.     public function __construct(IL10N $l10n,
    22.                                 IURLGenerator $urlGenerator) {
    23.         $this->l10n = $l10n;
    24.         $this->urlGenerator = $urlGenerator;
    25.     }
    27.     public function getId(): string {
    28.         return 'mysearchprovider';
    29.     }
    31.     public function getName(): string {
    32.         return $this->l->t('My app');
    33.     }
    35.     public function getOrder(string $route, array $routeParameters): int {
    36.         if (strpos($route, Application::APP_ID . '.') === 0) {
    37.             // Active app, prefer my results
    38.             return -1;
    39.         }
    41.         return 25;
    42.     }
    44.     public function search(IUser $user, ISearchQuery $query): SearchResult {
    45.         $offset = ($query->getCursor() ?? 0);
    46.         $limit = $query->getLimit();
    48.         $data = []; // Fill this with $limit entries, where the first entry is row $offset
    50.         return SearchResult::paginated(
    51.             $this->l10n->t('My app'),
    52.             $data,
    53.             $offset + $limit
    54.         );
    55.     }
    56. }

    So the first call will get a cursor of null and a limit of, say, 20. So the first 20 rows are fetched. The next call will have a cursor of 20, so the 20st to 39th rows are fetched.

    The downside of a offset-based pagination is that when the underlying data changes (new entries are inserted into or deleted from the database, files change), the offset might be out of sync from on request to its successor. Therefor, if possible, a true cursor-based pagination is preferable.

    For a cursor-based pagination a app-specific property is used to know a reference to the last element of the previous search request. The presumption of this algorithm is that the result set is sorted by an attribute and this attribute is an int or string. The attribute value of the last element in the result page determines the cursor for the next search request. Again, a small example shall demonstrate how this works.

    The unified search is available via OCS, which means client application like the mobile apps can use it to get access to the server search mechanism. The default properties of a search result entry might be difficult to parse and interpret in those clients, hence it’s possible to add optional string attributes to each entry.

    1. <?php
    3. $entry = new SearchResultEntry(/* same arguments as above */);
    4. $entry->addAttribute("type", "deckCard");
    5. $entry->addAttribute("cardId", "1234");

    Note

    This method was added in Nextcloud 21. If your app also targets Nextcloud 20 you should either not use it or add a version check to invoke the method only conditionally.