diff --git a/provision/hosts/torus/rss-bridge.nix b/provision/hosts/torus/rss-bridge.nix
index fdbc42d2..e0120a96 100644
--- a/provision/hosts/torus/rss-bridge.nix
+++ b/provision/hosts/torus/rss-bridge.nix
@@ -1,3 +1,6 @@
+# Mostly a copy of https://github.com/NixOS/nixpkgs/blob/nixos-unstable/nixos/modules/services/web-apps/rss-bridge.nix
+# Since I'm using a custom derivation of rss-bridge I must define my own services module.
+
{ config, lib, pkgs, ... }:
with lib;
let
diff --git a/provision/pkgs/rss-bridge/default.nix b/provision/pkgs/rss-bridge/default.nix
index b0d021d2..35d28979 100644
--- a/provision/pkgs/rss-bridge/default.nix
+++ b/provision/pkgs/rss-bridge/default.nix
@@ -2,17 +2,18 @@
stdenv.mkDerivation rec {
pname = "rss-bridge";
- version = "af9996ce3462b5b1ee0a8e6d95b5b010fc26b0a5";
+ version = "b037d1b4d1f0b0f422e21125ddef00a58e185ed1";
src = fetchFromGitHub {
- owner = "starr-dusT";
+ owner = "RSS-Bridge";
repo = "rss-bridge";
rev = version;
- sha256 = "sha256-e1VstKjKfnNmjYX0k2FswGF0Kj+AOOEcRqcGw6yyRB8=";
+ sha256 = "sha256-zyWnjSYE2NFK/OJLnsFsE5oEyf+yrJe8TT6MH4roBwU=";
};
patches = [
./paths.patch
+ ./youtube_bridge.patch
];
installPhase = ''
@@ -22,7 +23,7 @@ stdenv.mkDerivation rec {
meta = with lib; {
description = "The RSS feed for websites missing it";
- homepage = "https://github.com/starr-dusT/rss-bridge";
+ homepage = "https://github.com/RSS-Bridge/rss-bridge";
license = licenses.unlicense;
maintainers = with maintainers; [ starr-dusT ];
platforms = platforms.all;
diff --git a/provision/pkgs/rss-bridge/youtube_bridge.patch b/provision/pkgs/rss-bridge/youtube_bridge.patch
new file mode 100644
index 00000000..bc1a1e93
--- /dev/null
+++ b/provision/pkgs/rss-bridge/youtube_bridge.patch
@@ -0,0 +1,567 @@
+diff --git a/bridges/CustomYoutubeBridge.php b/bridges/CustomYoutubeBridge.php
+new file mode 100644
+index 00000000..d04c6ac0
+--- /dev/null
++++ b/bridges/CustomYoutubeBridge.php
+@@ -0,0 +1,561 @@
++ [
++ 'u' => [
++ 'name' => 'username',
++ 'exampleValue' => 'LinusTechTips',
++ 'required' => true
++ ]
++ ],
++ 'By channel id' => [
++ 'c' => [
++ 'name' => 'channel id',
++ 'exampleValue' => 'UCw38-8_Ibv_L6hlKChHO9dQ',
++ 'required' => true
++ ]
++ ],
++ 'By custom name' => [
++ 'custom' => [
++ 'name' => 'custom name',
++ 'exampleValue' => 'LinusTechTips',
++ 'required' => true
++ ]
++ ],
++ 'By playlist Id' => [
++ 'p' => [
++ 'name' => 'playlist id',
++ 'exampleValue' => 'PL8mG-RkN2uTzJc8N0EoyhdC54prvBBLpj',
++ 'required' => true
++ ]
++ ],
++ 'Search result' => [
++ 's' => [
++ 'name' => 'search keyword',
++ 'exampleValue' => 'LinusTechTips',
++ 'required' => true
++ ],
++ 'pa' => [
++ 'name' => 'page',
++ 'type' => 'number',
++ 'title' => 'This option is not work anymore, as YouTube will always return the same page',
++ 'exampleValue' => 1
++ ]
++ ],
++ 'global' => [
++ 'duration_min' => [
++ 'name' => 'min. duration (minutes)',
++ 'type' => 'number',
++ 'title' => 'Minimum duration for the video in minutes',
++ 'exampleValue' => 5
++ ],
++ 'duration_max' => [
++ 'name' => 'max. duration (minutes)',
++ 'type' => 'number',
++ 'title' => 'Maximum duration for the video in minutes',
++ 'exampleValue' => 10
++ ]
++ ]
++ ];
++
++ private $feedName = '';
++ private $feeduri = '';
++ private $feedIconUrl = '';
++ // This took from repo BetterVideoRss of VerifiedJoseph.
++ const URI_REGEX = '/(https?:\/\/(?:www\.)?(?:[a-zA-Z0-9-.]{2,256}\.[a-z]{2,20})(\:[0-9]{2 ,4})?(?:\/[a-zA-Z0-9@:%_\+.,~#"\'!?&\/\/=\-*]+|\/)?)/ims'; //phpcs:ignore
++
++ public function collectData()
++ {
++ $cacheKey = 'youtube_rate_limit';
++ if ($this->cache->get($cacheKey)) {
++ throw new HttpException('429 Too Many Requests', 429);
++ }
++ try {
++ $this->collectDataInternal();
++ } catch (HttpException $e) {
++ if ($e->getCode() === 429) {
++ $this->cache->set($cacheKey, true, 60 * 16);
++ }
++ throw $e;
++ }
++ }
++
++ private function collectDataInternal()
++ {
++ $html = '';
++ $url_feed = '';
++ $url_listing = '';
++
++ $username = $this->getInput('u');
++ $channel = $this->getInput('c');
++ $custom = $this->getInput('custom');
++ $playlist = $this->getInput('p');
++ $search = $this->getInput('s');
++
++ $durationMin = $this->getInput('duration_min');
++ $durationMax = $this->getInput('duration_max');
++
++ // Whether to discriminate videos by duration
++ $filterByDuration = $durationMin || $durationMax;
++
++ if ($username) {
++ // user and channel
++ $url_feed = self::URI . '/feeds/videos.xml?user=' . urlencode($username);
++ $url_listing = self::URI . '/user/' . urlencode($username) . '/videos';
++ } elseif ($channel) {
++ $url_feed = self::URI . '/feeds/videos.xml?channel_id=' . urlencode($channel);
++ $url_listing = self::URI . '/channel/' . urlencode($channel) . '/videos';
++ } elseif ($custom) {
++ $url_listing = self::URI . '/' . urlencode($custom) . '/videos';
++ }
++
++ if ($url_feed || $url_listing) {
++ // user, channel or custom
++ $this->feeduri = $url_listing;
++ if ($custom) {
++ // Extract the feed url for the custom name
++ $html = $this->fetch($url_listing);
++ $jsonData = $this->extractJsonFromHtml($html);
++ // Pluck out the rss feed url
++ $url_feed = $jsonData->metadata->channelMetadataRenderer->rssUrl;
++ $this->feedIconUrl = $jsonData->metadata->channelMetadataRenderer->avatar->thumbnails[0]->url;
++ }
++ if (!$custom) {
++ // Fetch the html page
++ $html = $this->fetch($url_listing);
++ $jsonData = $this->extractJsonFromHtml($html);
++ }
++ $channel_id = '';
++ if (isset($jsonData->contents)) {
++ $channel_id = $jsonData->metadata->channelMetadataRenderer->externalId;
++ $jsonData = $jsonData->contents->twoColumnBrowseResultsRenderer->tabs[1];
++ $jsonData = $jsonData->tabRenderer->content->richGridRenderer->contents;
++ // $jsonData = $jsonData->itemSectionRenderer->contents[0]->gridRenderer->items;
++ $this->fetchItemsFromFromJsonData($jsonData);
++ } else {
++ returnServerError('Unable to get data from YouTube');
++ }
++ $this->feedName = str_replace(' - YouTube', '', $html->find('title', 0)->plaintext);
++ } elseif ($playlist) {
++ // playlist
++ $url_feed = self::URI . '/feeds/videos.xml?playlist_id=' . urlencode($playlist);
++ $url_listing = self::URI . '/playlist?list=' . urlencode($playlist);
++ $html = $this->fetch($url_listing);
++ $jsonData = $this->extractJsonFromHtml($html);
++ // TODO: this method returns only first 100 video items
++ // if it has more videos, playlistVideoListRenderer will have continuationItemRenderer as last element
++ $jsonData = $jsonData->contents->twoColumnBrowseResultsRenderer->tabs[0];
++ $jsonData = $jsonData->tabRenderer->content->sectionListRenderer->contents[0]->itemSectionRenderer;
++ $jsonData = $jsonData->contents[0]->playlistVideoListRenderer->contents;
++ $item_count = count($jsonData);
++
++ if ($item_count > 15 || $filterByDuration) {
++ $this->fetchItemsFromFromJsonData($jsonData);
++ } else {
++ $xml = $this->fetch($url_feed);
++ $this->extractItemsFromXmlFeed($xml);
++ }
++ $this->feedName = 'Playlist: ' . str_replace(' - YouTube', '', $html->find('title', 0)->plaintext);
++ usort($this->items, function ($item1, $item2) {
++ if (!is_int($item1['timestamp']) && !is_int($item2['timestamp'])) {
++ $item1['timestamp'] = strtotime($item1['timestamp']);
++ $item2['timestamp'] = strtotime($item2['timestamp']);
++ }
++ return $item2['timestamp'] - $item1['timestamp'];
++ });
++ } elseif ($search) {
++ // search
++ $url_listing = self::URI . '/results?search_query=' . urlencode($search) . '&sp=CAI%253D';
++ $html = $this->fetch($url_listing);
++ $jsonData = $this->extractJsonFromHtml($html);
++ $jsonData = $jsonData->contents->twoColumnSearchResultsRenderer->primaryContents;
++ $jsonData = $jsonData->sectionListRenderer->contents;
++ foreach ($jsonData as $data) {
++ // Search result includes some ads, have to filter them
++ if (isset($data->itemSectionRenderer->contents[0]->videoRenderer)) {
++ $jsonData = $data->itemSectionRenderer->contents;
++ break;
++ }
++ }
++ $this->fetchItemsFromFromJsonData($jsonData);
++ $this->feeduri = $url_listing;
++ $this->feedName = 'Search: ' . $search;
++ } else {
++ returnClientError("You must either specify either:\n - YouTube username (?u=...)\n - Channel id (?c=...)\n - Playlist id (?p=...)\n - Search (?s=...)");
++ }
++ }
++
++ private function fetchVideoDetails($videoId, &$author, &$description, &$timestamp)
++ {
++ $url = self::URI . "/watch?v=$videoId";
++ $html = $this->fetch($url, true);
++
++ // Skip unavailable videos
++ if (strpos($html->innertext, 'IS_UNAVAILABLE_PAGE') !== false) {
++ return;
++ }
++
++ $elAuthor = $html->find('span[itemprop=author] > link[itemprop=name]', 0);
++ if (!is_null($elAuthor)) {
++ $author = $elAuthor->getAttribute('content');
++ }
++
++ $elDatePublished = $html->find('meta[itemprop=datePublished]', 0);
++ if (!is_null($elDatePublished)) {
++ $timestamp = strtotime($elDatePublished->getAttribute('content'));
++ }
++
++ $jsonData = $this->extractJsonFromHtml($html);
++ if (!isset($jsonData->contents)) {
++ return;
++ }
++
++ $jsonData = $jsonData->contents->twoColumnWatchNextResults->results->results->contents ?? null;
++ if (!$jsonData) {
++ throw new \Exception('Unable to find json data');
++ }
++ $videoSecondaryInfo = null;
++ foreach ($jsonData as $item) {
++ if (isset($item->videoSecondaryInfoRenderer)) {
++ $videoSecondaryInfo = $item->videoSecondaryInfoRenderer;
++ break;
++ }
++ }
++ if (!$videoSecondaryInfo) {
++ returnServerError('Could not find videoSecondaryInfoRenderer. Error at: ' . $videoId);
++ }
++
++ $description = $videoSecondaryInfo->attributedDescription->content ?? '';
++
++ // Default whitespace chars used by trim + non-breaking spaces (https://en.wikipedia.org/wiki/Non-breaking_space)
++ $whitespaceChars = " \t\n\r\0\x0B\u{A0}\u{2060}\u{202F}\u{2007}";
++ $descEnhancements = $this->ytBridgeGetVideoDescriptionEnhancements($videoSecondaryInfo, $description, self::URI, $whitespaceChars);
++ foreach ($descEnhancements as $descEnhancement) {
++ if (isset($descEnhancement['url'])) {
++ $descBefore = mb_substr($description, 0, $descEnhancement['pos']);
++ $descValue = mb_substr($description, $descEnhancement['pos'], $descEnhancement['len']);
++ $descAfter = mb_substr($description, $descEnhancement['pos'] + $descEnhancement['len'], null);
++
++ // Extended trim for the display value of internal links, e.g.:
++ // FAVICON • Video Name
++ // FAVICON / @ChannelName
++ $descValue = trim($descValue, $whitespaceChars . '•/');
++
++ $description = sprintf('%s%s%s', $descBefore, $descEnhancement['url'], $descValue, $descAfter);
++ }
++ }
++ }
++
++ private function ytBridgeGetVideoDescriptionEnhancements(
++ object $videoSecondaryInfo,
++ string $descriptionContent,
++ string $baseUrl,
++ string $whitespaceChars
++ ): array {
++ $commandRuns = $videoSecondaryInfo->attributedDescription->commandRuns ?? [];
++ if (count($commandRuns) <= 0) {
++ return [];
++ }
++
++ $enhancements = [];
++
++ $boundaryWhitespaceChars = mb_str_split($whitespaceChars);
++ $boundaryStartChars = array_merge($boundaryWhitespaceChars, [':', '-', '(']);
++ $boundaryEndChars = array_merge($boundaryWhitespaceChars, [',', '.', "'", ')']);
++ $hashtagBoundaryEndChars = array_merge($boundaryEndChars, ['#', '-']);
++
++ $descriptionContentLength = mb_strlen($descriptionContent);
++
++ $minPositionOffset = 0;
++
++ $prevStartPosition = 0;
++ $totalLength = 0;
++ $maxPositionByStartIndex = [];
++ foreach (array_reverse($commandRuns) as $commandRun) {
++ $endPosition = $commandRun->startIndex + $commandRun->length;
++ if ($endPosition < $prevStartPosition) {
++ $totalLength += 1;
++ }
++ $totalLength += $commandRun->length;
++ $maxPositionByStartIndex[$commandRun->startIndex] = $totalLength;
++ $prevStartPosition = $commandRun->startIndex;
++ }
++
++ foreach ($commandRuns as $commandRun) {
++ $commandMetadata = $commandRun->onTap->innertubeCommand->commandMetadata->webCommandMetadata ?? null;
++ if (!isset($commandMetadata)) {
++ continue;
++ }
++
++ $enhancement = null;
++
++ /*
++ $commandRun->startIndex can be offset by few positions in the positive direction
++ when some multibyte characters (e.g. emojis, but maybe also others) are used in the plain text video description.
++ (probably some difference between php and javascript in handling multibyte characters)
++ This loop should correct the position in most cases. It searches for the next word (determined by a set of boundary chars) with the expected length.
++ Several safeguards ensure that the correct word is chosen. When a link can not be matched,
++ everything will be discarded to prevent corrupting the description.
++ Hashtags require a different set of boundary chars.
++ */
++ $isHashtag = $commandMetadata->webPageType === 'WEB_PAGE_TYPE_BROWSE';
++ $prevEnhancement = end($enhancements);
++ $minPosition = $prevEnhancement === false ? 0 : $prevEnhancement['pos'] + $prevEnhancement['len'];
++ $maxPosition = $descriptionContentLength - $maxPositionByStartIndex[$commandRun->startIndex];
++ $position = min($commandRun->startIndex - $minPositionOffset, $maxPosition);
++ while ($position >= $minPosition) {
++ // The link display value can only ever include a new line at the end (which will be removed further below), never in between.
++ $newLinePosition = mb_strpos($descriptionContent, "\n", $position);
++ if ($newLinePosition !== false && $newLinePosition < $position + ($commandRun->length - 1)) {
++ $position = $newLinePosition - ($commandRun->length - 1);
++ continue;
++ }
++
++ $firstChar = mb_substr($descriptionContent, $position, 1);
++ $boundaryStart = mb_substr($descriptionContent, $position - 1, 1);
++ $boundaryEndIndex = $position + $commandRun->length;
++ $boundaryEnd = mb_substr($descriptionContent, $boundaryEndIndex, 1);
++
++ $boundaryStartIsValid = $position === 0 ||
++ in_array($boundaryStart, $boundaryStartChars) ||
++ ($isHashtag && $firstChar === '#');
++ $boundaryEndIsValid = $boundaryEndIndex === $descriptionContentLength ||
++ in_array($boundaryEnd, $isHashtag ? $hashtagBoundaryEndChars : $boundaryEndChars);
++
++ if ($boundaryStartIsValid && $boundaryEndIsValid) {
++ $minPositionOffset = $commandRun->startIndex - $position;
++ $enhancement = [
++ 'pos' => $position,
++ 'len' => $commandRun->length,
++ ];
++ break;
++ }
++
++ $position--;
++ }
++
++ if (!isset($enhancement)) {
++ $this->logger->debug(sprintf('Position %d cannot be corrected in "%s"', $commandRun->startIndex, substr($descriptionContent, 0, 50) . '...'));
++ // Skip to prevent the description from becoming corrupted
++ continue;
++ }
++
++ // $commandRun->length sometimes incorrectly includes the newline as last char
++ $lastChar = mb_substr($descriptionContent, $enhancement['pos'] + $enhancement['len'] - 1, 1);
++ if ($lastChar === "\n") {
++ $enhancement['len'] -= 1;
++ }
++
++ $commandUrl = parse_url($commandMetadata->url);
++ if ($commandUrl['path'] === '/redirect') {
++ parse_str($commandUrl['query'], $commandUrlQuery);
++ $enhancement['url'] = urldecode($commandUrlQuery['q']);
++ } elseif (isset($commandUrl['host'])) {
++ $enhancement['url'] = $commandMetadata->url;
++ } else {
++ $enhancement['url'] = $baseUrl . $commandMetadata->url;
++ }
++
++ $enhancements[] = $enhancement;
++ }
++
++ if (count($enhancements) !== count($commandRuns)) {
++ // At least one link can not be matched. Discard everything to prevent corrupting the description.
++ return [];
++ }
++
++ // Sort by position in descending order to be able to safely replace values
++ return array_reverse($enhancements);
++ }
++
++ private function extractItemsFromXmlFeed($xml)
++ {
++ $this->feedName = $this->decodeTitle($xml->find('feed > title', 0)->plaintext);
++
++ foreach ($xml->find('entry') as $element) {
++ $videoId = str_replace('yt:video:', '', $element->find('id', 0)->plaintext);
++ if (strpos($videoId, 'googleads') !== false) {
++ continue;
++ }
++ $title = $this->decodeTitle($element->find('title', 0)->plaintext);
++ $author = $element->find('name', 0)->plaintext;
++ $desc = $element->find('media:description', 0)->innertext;
++ $desc = htmlspecialchars($desc);
++ $desc = nl2br($desc);
++ $desc = preg_replace(self::URI_REGEX, '$1 ', $desc);
++ $time = strtotime($element->find('published', 0)->plaintext);
++ $this->addItem($videoId, $title, $author, $desc, $time);
++ }
++ }
++
++ private function fetch($url, bool $cache = false)
++ {
++ $header = ['Accept-Language: en-US'];
++ $ttl = 86400 * 3; // 3d
++ $stripNewlines = false;
++ if ($cache) {
++ return getSimpleHTMLDOMCached($url, $ttl, $header, [], true, true, DEFAULT_TARGET_CHARSET, $stripNewlines);
++ }
++ return getSimpleHTMLDOM($url, $header, [], true, true, DEFAULT_TARGET_CHARSET, $stripNewlines);
++ }
++
++ private function extractJsonFromHtml($html)
++ {
++ $scriptRegex = '/var ytInitialData = (.*?);<\/script>/';
++ $result = preg_match($scriptRegex, $html, $matches);
++ if (! $result) {
++ $this->logger->debug('Could not find ytInitialData');
++ return null;
++ }
++ $data = json_decode($matches[1]);
++ return $data;
++ }
++
++ private function fetchItemsFromFromJsonData($jsonData)
++ {
++ $minimumDurationSeconds = ($this->getInput('duration_min') ?: -1) * 60;
++ $maximumDurationSeconds = ($this->getInput('duration_max') ?: INF) * 60;
++
++ foreach ($jsonData as $item) {
++ $wrapper = null;
++ if (isset($item->gridVideoRenderer)) {
++ $wrapper = $item->gridVideoRenderer;
++ } elseif (isset($item->videoRenderer)) {
++ $wrapper = $item->videoRenderer;
++ } elseif (isset($item->playlistVideoRenderer)) {
++ $wrapper = $item->playlistVideoRenderer;
++ } elseif (isset($item->richItemRenderer)) {
++ $wrapper = $item->richItemRenderer->content->videoRenderer;
++ } else {
++ continue;
++ }
++
++ // 01:03:30 | 15:06 | 1:24
++ $lengthText = $wrapper->lengthText->simpleText ?? null;
++ // 6,875 views
++ $viewCount = $wrapper->viewCountText->simpleText ?? null;
++ // Dc645M8Het8
++ $videoId = $wrapper->videoId;
++ // Jumbo frames - transfer more data faster!
++ $title = $wrapper->title->runs[0]->text ?? $wrapper->title->accessibility->accessibilityData->label ?? null;
++ $author = null;
++ $description = $wrapper->descriptionSnippet->runs[0]->text ?? null;
++ // 5 days ago | 1 month ago
++ $publishedTimeText = $wrapper->publishedTimeText->simpleText ?? $wrapper->videoInfo->runs[2]->text ?? null;
++ $timestamp = null;
++ if ($publishedTimeText) {
++ try {
++ $publicationDate = new \DateTimeImmutable($publishedTimeText);
++ // Hard-code hour, minute and second
++ $publicationDate = $publicationDate->setTime(0, 0, 0);
++ $timestamp = $publicationDate->getTimestamp();
++ } catch (\Exception $e) {
++ }
++ }
++
++ $durationText = 0;
++ if ($lengthText) {
++ $durationText = $lengthText;
++ } else {
++ foreach ($wrapper->thumbnailOverlays as $overlay) {
++ if (isset($overlay->thumbnailOverlayTimeStatusRenderer)) {
++ $durationText = $overlay->thumbnailOverlayTimeStatusRenderer->text;
++ break;
++ }
++ }
++ }
++ if (is_string($durationText)) {
++ if (preg_match('/([\d]{1,2})\:([\d]{1,2})\:([\d]{2})/', $durationText)) {
++ $durationText = preg_replace('/([\d]{1,2})\:([\d]{1,2})\:([\d]{2})/', '$1:$2:$3', $durationText);
++ } else {
++ $durationText = preg_replace('/([\d]{1,2})\:([\d]{2})/', '00:$1:$2', $durationText);
++ }
++ sscanf($durationText, '%d:%d:%d', $hours, $minutes, $seconds);
++ $duration = $hours * 3600 + $minutes * 60 + $seconds;
++ if ($duration < $minimumDurationSeconds || $duration > $maximumDurationSeconds) {
++ continue;
++ }
++ }
++ # Re-fetch better details from xml
++ $this->fetchVideoDetails($videoId, $author, $description, $timestamp);
++ $this->addItem($videoId, $title, $author, $description, $timestamp, $durationText);
++ if (count($this->items) >= 99) {
++ break;
++ }
++ }
++ }
++
++ private function addItem($videoId, $title, $author, $description, $timestamp, $durationText, $thumbnail = '')
++ {
++ $description = nl2br($description);
++
++ $item = [];
++ // This should probably be uid?
++ $item['id'] = $videoId;
++ $item['title'] = $title . " [" . $durationText . "]";
++ $item['author'] = $author ?? '';
++ $item['timestamp'] = $timestamp;
++ $item['uri'] = self::URI . '/watch?v=' . $videoId;
++ if (!$thumbnail) {
++ // Fallback to default thumbnail if there aren't any provided.
++ $thumbnail = '0';
++ }
++ $thumbnailUri = str_replace('/www.', '/img.', self::URI) . '/vi/' . $videoId . '/' . $thumbnail . '.jpg';
++ $item['content'] = sprintf('
%s', $item['uri'], $thumbnailUri, $description);
++ $this->items[] = $item;
++ }
++
++ private function decodeTitle($title)
++ {
++ // convert both Ӓ and " to UTF-8
++ return html_entity_decode($title, ENT_QUOTES, 'UTF-8');
++ }
++
++ public function getURI()
++ {
++ if (!is_null($this->getInput('p'))) {
++ return static::URI . '/playlist?list=' . $this->getInput('p');
++ } elseif ($this->feeduri) {
++ return $this->feeduri;
++ }
++
++ return parent::getURI();
++ }
++
++ public function getName()
++ {
++ switch ($this->queriedContext) {
++ case 'By username':
++ case 'By channel id':
++ case 'By custom name':
++ case 'By playlist Id':
++ case 'Search result':
++ return htmlspecialchars_decode($this->feedName) . ' - YouTube';
++ default:
++ return parent::getName();
++ }
++ }
++
++ public function getIcon()
++ {
++ if (empty($this->feedIconUrl)) {
++ return parent::getIcon();
++ } else {
++ return $this->feedIconUrl;
++ }
++ }
++}