Merge branch 'iv-org:master' into main

.github/workflows/ci.yml

@@ -38,11 +38,10 @@ jobs:
       matrix:
         stable: [true]
         crystal:
-          - 1.4.1
-          - 1.5.1
           - 1.6.2
           - 1.7.3
           - 1.8.2
+          - 1.9.2
         include:
           - crystal: nightly
             stable: false
@@ -53,7 +52,7 @@ jobs:
           submodules: true
 
       - name: Install Crystal
-        uses: crystal-lang/install-crystal@v1.7.0
+        uses: crystal-lang/install-crystal@v1.8.0
         with:
           crystal: ${{ matrix.crystal }}
 

.github/workflows/container-release.yml

@@ -25,9 +25,9 @@ jobs:
         uses: actions/checkout@v3
 
       - name: Install Crystal
-        uses: crystal-lang/install-crystal@v1.6.0
+        uses: crystal-lang/install-crystal@v1.8.0
         with:
-          crystal: 1.5.0
+          crystal: 1.9.2
 
       - name: Run lint
         run: |
@@ -77,4 +77,3 @@ jobs:
           tags: quay.io/invidious/invidious:${{ github.sha }}-arm64,quay.io/invidious/invidious:latest-arm64
           build-args: |
             "release=1"
-        

.github/workflows/stale.yml

@@ -14,7 +14,7 @@ jobs:
       with:
         repo-token: ${{ secrets.GITHUB_TOKEN }}
         days-before-stale: 365
-        days-before-pr-stale: 45 # PRs should be active. Anything that hasn't had activity in more than 45 days should be considered abandoned.
+        days-before-pr-stale: 90 
         days-before-close: 30
         exempt-pr-labels: blocked
         stale-issue-message: 'This issue has been automatically marked as stale and will be closed in 30 days because it has not had recent activity and is much likely outdated. If you think this issue is still relevant and applicable, you just have to post a comment and it will be unmarked.'

config/config.example.yml

@@ -161,6 +161,19 @@ https_only: false
 #force_resolve:
 
 
+##
+## Use Innertube's transcripts API instead of timedtext for closed captions
+##
+## Useful for larger instances as InnerTube is **not ratelimited**. See https://github.com/iv-org/invidious/issues/2567
+##
+## Subtitle experience may differ slightly on Invidious.
+##
+## Accepted values: true, false
+## Default: false
+##
+# use_innertube_for_captions: false
+
+
 # -----------------------------
 #  Logging
 # -----------------------------

src/invidious/config.cr

@@ -127,6 +127,9 @@ class Config
   # Pool size for HTTP requests to youtube.com and ytimg.com (each domain has a separate pool of `pool_size`)
   property pool_size : Int32 = 100
 
+  # Use Innertube's transcripts API instead of timedtext for closed captions
+  property use_innertube_for_captions : Bool = false
+
   # Saved cookies in "name1=value1; name2=value2..." format
   @[YAML::Field(converter: Preferences::StringToCookies)]
   property cookies : HTTP::Cookies = HTTP::Cookies.new

src/invidious/frontend/watch_page.cr

@@ -7,7 +7,7 @@ module Invidious::Frontend::WatchPage
     getter full_videos : Array(Hash(String, JSON::Any))
     getter video_streams : Array(Hash(String, JSON::Any))
     getter audio_streams : Array(Hash(String, JSON::Any))
-    getter captions : Array(Invidious::Videos::Caption)
+    getter captions : Array(Invidious::Videos::Captions::Metadata)
 
     def initialize(
       @full_videos,

src/invidious/playlists.cr

@@ -89,6 +89,7 @@ struct Playlist
   property views : Int64
   property updated : Time
   property thumbnail : String?
+  property subtitle : String?
 
   def to_json(offset, json : JSON::Builder, video_id : String? = nil)
     json.object do
@@ -100,6 +101,7 @@ struct Playlist
       json.field "author", self.author
       json.field "authorId", self.ucid
       json.field "authorUrl", "/channel/#{self.ucid}"
+      json.field "subtitle", self.subtitle
 
       json.field "authorThumbnails" do
         json.array do
@@ -356,6 +358,8 @@ def fetch_playlist(plid : String)
   updated = Time.utc
   video_count = 0
 
+  subtitle = extract_text(initial_data.dig?("header", "playlistHeaderRenderer", "subtitle"))
+
   playlist_info["stats"]?.try &.as_a.each do |stat|
     text = stat["runs"]?.try &.as_a.map(&.["text"].as_s).join("") || stat["simpleText"]?.try &.as_s
     next if !text
@@ -397,6 +401,7 @@ def fetch_playlist(plid : String)
     views:            views,
     updated:          updated,
     thumbnail:        thumbnail,
+    subtitle:         subtitle,
   })
 end
 

src/invidious/routes/api/v1/videos.cr

@@ -87,70 +87,78 @@ module Invidious::Routes::API::V1::Videos
       caption = caption[0]
     end
 
-    url = URI.parse("#{caption.base_url}&tlang=#{tlang}").request_target
+    if CONFIG.use_innertube_for_captions
+      params = Invidious::Videos::Transcript.generate_param(id, caption.language_code, caption.auto_generated)
+      initial_data = YoutubeAPI.get_transcript(params)
 
-    # Auto-generated captions often have cues that aren't aligned properly with the video,
+      webvtt = Invidious::Videos::Transcript.convert_transcripts_to_vtt(initial_data, caption.language_code)
-    # as well as some other markup that makes it cumbersome, so we try to fix that here
+    else
-    if caption.name.includes? "auto-generated"
+      # Timedtext API handling
-      caption_xml = YT_POOL.client &.get(url).body
+      url = URI.parse("#{caption.base_url}&tlang=#{tlang}").request_target
 
-      if caption_xml.starts_with?("<?xml")
+      # Auto-generated captions often have cues that aren't aligned properly with the video,
-        webvtt = caption.timedtext_to_vtt(caption_xml, tlang)
+      # as well as some other markup that makes it cumbersome, so we try to fix that here
-      else
+      if caption.name.includes? "auto-generated"
-        caption_xml = XML.parse(caption_xml)
+        caption_xml = YT_POOL.client &.get(url).body
 
-        webvtt = String.build do |str|
+        if caption_xml.starts_with?("<?xml")
-          str << <<-END_VTT
+          webvtt = caption.timedtext_to_vtt(caption_xml, tlang)
-          WEBVTT
+        else
-          Kind: captions
+          caption_xml = XML.parse(caption_xml)
-          Language: #{tlang || caption.language_code}
+
+          webvtt = String.build do |str|
+            str << <<-END_VTT
+            WEBVTT
+            Kind: captions
+            Language: #{tlang || caption.language_code}
 
 
-          END_VTT
+            END_VTT
 
-          caption_nodes = caption_xml.xpath_nodes("//transcript/text")
+            caption_nodes = caption_xml.xpath_nodes("//transcript/text")
-          caption_nodes.each_with_index do |node, i|
+            caption_nodes.each_with_index do |node, i|
-            start_time = node["start"].to_f.seconds
+              start_time = node["start"].to_f.seconds
-            duration = node["dur"]?.try &.to_f.seconds
+              duration = node["dur"]?.try &.to_f.seconds
-            duration ||= start_time
+              duration ||= start_time
 
-            if caption_nodes.size > i + 1
+              if caption_nodes.size > i + 1
-              end_time = caption_nodes[i + 1]["start"].to_f.seconds
+                end_time = caption_nodes[i + 1]["start"].to_f.seconds
-            else
+              else
-              end_time = start_time + duration
+                end_time = start_time + duration
+              end
+
+              start_time = "#{start_time.hours.to_s.rjust(2, '0')}:#{start_time.minutes.to_s.rjust(2, '0')}:#{start_time.seconds.to_s.rjust(2, '0')}.#{start_time.milliseconds.to_s.rjust(3, '0')}"
+              end_time = "#{end_time.hours.to_s.rjust(2, '0')}:#{end_time.minutes.to_s.rjust(2, '0')}:#{end_time.seconds.to_s.rjust(2, '0')}.#{end_time.milliseconds.to_s.rjust(3, '0')}"
+
+              text = HTML.unescape(node.content)
+              text = text.gsub(/<font color="#[a-fA-F0-9]{6}">/, "")
+              text = text.gsub(/<\/font>/, "")
+              if md = text.match(/(?<name>.*) : (?<text>.*)/)
+                text = "<v #{md["name"]}>#{md["text"]}</v>"
+              end
+
+              str << <<-END_CUE
+              #{start_time} --> #{end_time}
+              #{text}
+
+
+              END_CUE
             end
-
-            start_time = "#{start_time.hours.to_s.rjust(2, '0')}:#{start_time.minutes.to_s.rjust(2, '0')}:#{start_time.seconds.to_s.rjust(2, '0')}.#{start_time.milliseconds.to_s.rjust(3, '0')}"
-            end_time = "#{end_time.hours.to_s.rjust(2, '0')}:#{end_time.minutes.to_s.rjust(2, '0')}:#{end_time.seconds.to_s.rjust(2, '0')}.#{end_time.milliseconds.to_s.rjust(3, '0')}"
-
-            text = HTML.unescape(node.content)
-            text = text.gsub(/<font color="#[a-fA-F0-9]{6}">/, "")
-            text = text.gsub(/<\/font>/, "")
-            if md = text.match(/(?<name>.*) : (?<text>.*)/)
-              text = "<v #{md["name"]}>#{md["text"]}</v>"
-            end
-
-            str << <<-END_CUE
-            #{start_time} --> #{end_time}
-            #{text}
-
-
-            END_CUE
           end
         end
-      end
-    else
-      # Some captions have "align:[start/end]" and "position:[num]%"
-      # attributes. Those are causing issues with VideoJS, which is unable
-      # to properly align the captions on the video, so we remove them.
-      #
-      # See: https://github.com/iv-org/invidious/issues/2391
-      webvtt = YT_POOL.client &.get("#{url}&format=vtt").body
-      if webvtt.starts_with?("<?xml")
-        webvtt = caption.timedtext_to_vtt(webvtt)
       else
+        # Some captions have "align:[start/end]" and "position:[num]%"
+        # attributes. Those are causing issues with VideoJS, which is unable
+        # to properly align the captions on the video, so we remove them.
+        #
+        # See: https://github.com/iv-org/invidious/issues/2391
         webvtt = YT_POOL.client &.get("#{url}&format=vtt").body
-          .gsub(/([0-9:.]{12} --> [0-9:.]{12}).+/, "\\1")
+        if webvtt.starts_with?("<?xml")
+          webvtt = caption.timedtext_to_vtt(webvtt)
+        else
+          webvtt = YT_POOL.client &.get("#{url}&format=vtt").body
+            .gsub(/([0-9:.]{12} --> [0-9:.]{12}).+/, "\\1")
+        end
       end
     end
 

src/invidious/videos.cr

@@ -24,7 +24,7 @@ struct Video
   property updated : Time
 
   @[DB::Field(ignore: true)]
-  @captions = [] of Invidious::Videos::Caption
+  @captions = [] of Invidious::Videos::Captions::Metadata
 
   @[DB::Field(ignore: true)]
   property adaptive_fmts : Array(Hash(String, JSON::Any))?
@@ -215,9 +215,9 @@ struct Video
     keywords.includes? "YouTube Red"
   end
 
-  def captions : Array(Invidious::Videos::Caption)
+  def captions : Array(Invidious::Videos::Captions::Metadata)
     if @captions.empty? && @info.has_key?("captions")
-      @captions = Invidious::Videos::Caption.from_yt_json(info["captions"])
+      @captions = Invidious::Videos::Captions::Metadata.from_yt_json(info["captions"])
     end
 
     return @captions

src/invidious/videos/caption.cr

@@ -1,100 +1,106 @@
 require "json"
 
 module Invidious::Videos
-  struct Caption
+  module Captions
-    property name : String
+    struct Metadata
-    property language_code : String
+      property name : String
-    property base_url : String
+      property language_code : String
+      property base_url : String
 
-    def initialize(@name, @language_code, @base_url)
+      property auto_generated : Bool
-    end
 
-    # Parse the JSON structure from Youtube
+      def initialize(@name, @language_code, @base_url, @auto_generated)
-    def self.from_yt_json(container : JSON::Any) : Array(Caption)
-      caption_tracks = container
-        .dig?("playerCaptionsTracklistRenderer", "captionTracks")
-        .try &.as_a
-
-      captions_list = [] of Caption
-      return captions_list if caption_tracks.nil?
-
-      caption_tracks.each do |caption|
-        name = caption["name"]["simpleText"]? || caption["name"]["runs"][0]["text"]
-        name = name.to_s.split(" - ")[0]
-
-        language_code = caption["languageCode"].to_s
-        base_url = caption["baseUrl"].to_s
-
-        captions_list << Caption.new(name, language_code, base_url)
       end
 
-      return captions_list
+      # Parse the JSON structure from Youtube
-    end
+      def self.from_yt_json(container : JSON::Any) : Array(Captions::Metadata)
+        caption_tracks = container
+          .dig?("playerCaptionsTracklistRenderer", "captionTracks")
+          .try &.as_a
 
-    def timedtext_to_vtt(timedtext : String, tlang = nil) : String
+        captions_list = [] of Captions::Metadata
-      # In the future, we could just directly work with the url. This is more of a POC
+        return captions_list if caption_tracks.nil?
-      cues = [] of XML::Node
-      tree = XML.parse(timedtext)
-      tree = tree.children.first
 
-      tree.children.each do |item|
+        caption_tracks.each do |caption|
-        if item.name == "body"
+          name = caption["name"]["simpleText"]? || caption["name"]["runs"][0]["text"]
-          item.children.each do |cue|
+          name = name.to_s.split(" - ")[0]
-            if cue.name == "p" && !(cue.children.size == 1 && cue.children[0].content == "\n")
+
-              cues << cue
+          language_code = caption["languageCode"].to_s
+          base_url = caption["baseUrl"].to_s
+
+          auto_generated = (caption["kind"]? == "asr")
+
+          captions_list << Captions::Metadata.new(name, language_code, base_url, auto_generated)
+        end
+
+        return captions_list
+      end
+
+      def timedtext_to_vtt(timedtext : String, tlang = nil) : String
+        # In the future, we could just directly work with the url. This is more of a POC
+        cues = [] of XML::Node
+        tree = XML.parse(timedtext)
+        tree = tree.children.first
+
+        tree.children.each do |item|
+          if item.name == "body"
+            item.children.each do |cue|
+              if cue.name == "p" && !(cue.children.size == 1 && cue.children[0].content == "\n")
+                cues << cue
+              end
             end
+            break
           end
-          break
         end
-      end
+        result = String.build do |result|
-      result = String.build do |result|
+          result << <<-END_VTT
-        result << <<-END_VTT
+          WEBVTT
-        WEBVTT
+          Kind: captions
-        Kind: captions
+          Language: #{tlang || @language_code}
-        Language: #{tlang || @language_code}
 
 
-        END_VTT
+          END_VTT
 
-        result << "\n\n"
+          result << "\n\n"
 
-        cues.each_with_index do |node, i|
+          cues.each_with_index do |node, i|
-          start_time = node["t"].to_f.milliseconds
+            start_time = node["t"].to_f.milliseconds
 
-          duration = node["d"]?.try &.to_f.milliseconds
+            duration = node["d"]?.try &.to_f.milliseconds
 
-          duration ||= start_time
+            duration ||= start_time
 
-          if cues.size > i + 1
+            if cues.size > i + 1
-            end_time = cues[i + 1]["t"].to_f.milliseconds
+              end_time = cues[i + 1]["t"].to_f.milliseconds
-          else
+            else
-            end_time = start_time + duration
+              end_time = start_time + duration
+            end
+
+            # start_time
+            result << start_time.hours.to_s.rjust(2, '0')
+            result << ':' << start_time.minutes.to_s.rjust(2, '0')
+            result << ':' << start_time.seconds.to_s.rjust(2, '0')
+            result << '.' << start_time.milliseconds.to_s.rjust(3, '0')
+
+            result << " --> "
+
+            # end_time
+            result << end_time.hours.to_s.rjust(2, '0')
+            result << ':' << end_time.minutes.to_s.rjust(2, '0')
+            result << ':' << end_time.seconds.to_s.rjust(2, '0')
+            result << '.' << end_time.milliseconds.to_s.rjust(3, '0')
+
+            result << "\n"
+
+            node.children.each do |s|
+              result << s.content
+            end
+            result << "\n"
+            result << "\n"
           end
-
-          # start_time
-          result << start_time.hours.to_s.rjust(2, '0')
-          result << ':' << start_time.minutes.to_s.rjust(2, '0')
-          result << ':' << start_time.seconds.to_s.rjust(2, '0')
-          result << '.' << start_time.milliseconds.to_s.rjust(3, '0')
-
-          result << " --> "
-
-          # end_time
-          result << end_time.hours.to_s.rjust(2, '0')
-          result << ':' << end_time.minutes.to_s.rjust(2, '0')
-          result << ':' << end_time.seconds.to_s.rjust(2, '0')
-          result << '.' << end_time.milliseconds.to_s.rjust(3, '0')
-
-          result << "\n"
-
-          node.children.each do |s|
-            result << s.content
-          end
-          result << "\n"
-          result << "\n"
         end
+        return result
       end
-      return result
     end
 
     # List of all caption languages available on Youtube.

src/invidious/videos/transcript.cr

@@ -0,0 +1,103 @@
+module Invidious::Videos
+  # Namespace for methods primarily relating to Transcripts
+  module Transcript
+    record TranscriptLine, start_ms : Time::Span, end_ms : Time::Span, line : String
+
+    def self.generate_param(video_id : String, language_code : String, auto_generated : Bool) : String
+      kind = auto_generated ? "asr" : ""
+
+      object = {
+        "1:0:string" => video_id,
+
+        "2:base64" => {
+          "1:string" => kind,
+          "2:string" => language_code,
+          "3:string" => "",
+        },
+
+        "3:varint" => 1_i64,
+        "5:string" => "engagement-panel-searchable-transcript-search-panel",
+        "6:varint" => 1_i64,
+        "7:varint" => 1_i64,
+        "8:varint" => 1_i64,
+      }
+
+      params = object.try { |i| Protodec::Any.cast_json(i) }
+        .try { |i| Protodec::Any.from_json(i) }
+        .try { |i| Base64.urlsafe_encode(i) }
+        .try { |i| URI.encode_www_form(i) }
+
+      return params
+    end
+
+    def self.convert_transcripts_to_vtt(initial_data : Hash(String, JSON::Any), target_language : String) : String
+      # Convert into array of TranscriptLine
+      lines = self.parse(initial_data)
+
+      # Taken from Invidious::Videos::Captions::Metadata.timedtext_to_vtt()
+      vtt = String.build do |vtt|
+        vtt << <<-END_VTT
+        WEBVTT
+        Kind: captions
+        Language: #{target_language}
+
+
+        END_VTT
+
+        vtt << "\n\n"
+
+        lines.each do |line|
+          start_time = line.start_ms
+          end_time = line.end_ms
+
+          # start_time
+          vtt << start_time.hours.to_s.rjust(2, '0')
+          vtt << ':' << start_time.minutes.to_s.rjust(2, '0')
+          vtt << ':' << start_time.seconds.to_s.rjust(2, '0')
+          vtt << '.' << start_time.milliseconds.to_s.rjust(3, '0')
+
+          vtt << " --> "
+
+          # end_time
+          vtt << end_time.hours.to_s.rjust(2, '0')
+          vtt << ':' << end_time.minutes.to_s.rjust(2, '0')
+          vtt << ':' << end_time.seconds.to_s.rjust(2, '0')
+          vtt << '.' << end_time.milliseconds.to_s.rjust(3, '0')
+
+          vtt << "\n"
+          vtt << line.line
+
+          vtt << "\n"
+          vtt << "\n"
+        end
+      end
+
+      return vtt
+    end
+
+    private def self.parse(initial_data : Hash(String, JSON::Any))
+      body = initial_data.dig("actions", 0, "updateEngagementPanelAction", "content", "transcriptRenderer",
+        "content", "transcriptSearchPanelRenderer", "body", "transcriptSegmentListRenderer",
+        "initialSegments").as_a
+
+      lines = [] of TranscriptLine
+      body.each do |line|
+        # Transcript section headers. They are not apart of the captions and as such we can safely skip them.
+        if line.as_h.has_key?("transcriptSectionHeaderRenderer")
+          next
+        end
+
+        line = line["transcriptSegmentRenderer"]
+
+        start_ms = line["startMs"].as_s.to_i.millisecond
+        end_ms = line["endMs"].as_s.to_i.millisecond
+
+        text = extract_text(line["snippet"]) || ""
+
+        lines << TranscriptLine.new(start_ms, end_ms, text)
+      end
+
+      return lines
+    end
+  end
+end

src/invidious/views/playlist.ecr

@@ -70,7 +70,12 @@
             </b>
         <% else %>
             <b>
-                <a href="/channel/<%= playlist.ucid %>"><%= author %></a> |
+                <% if !author.empty? %>
+                    <a href="/channel/<%= playlist.ucid %>"><%= author %></a> |
+                <% elsif !playlist.subtitle.nil? %>
+                    <% subtitle = playlist.subtitle || "" %>
+                    <span><%= HTML.escape(subtitle[0..subtitle.rindex(" • ") || subtitle.size]) %></span> |
+                <% end %>
                 <%= translate_count(locale, "generic_videos_count", playlist.video_count) %> |
                 <%= translate(locale, "Updated `x` ago", recode_date(playlist.updated, locale)) %>
             </b>

src/invidious/views/user/preferences.ecr

@@ -89,7 +89,7 @@
                 <label for="captions[0]"><%= translate(locale, "preferences_captions_label") %></label>
                 <% preferences.captions.each_with_index do |caption, index| %>
                     <select class="pure-u-1-6" name="captions[<%= index %>]" id="captions[<%= index %>]">
-                        <% Invidious::Videos::Caption::LANGUAGES.each do |option| %>
+                        <% Invidious::Videos::Captions::LANGUAGES.each do |option| %>
                             <option value="<%= option %>" <% if preferences.captions[index] == option %> selected <% end %>><%= translate(locale, option.blank? ? "none" : option) %></option>
                         <% end %>
                     </select>

src/invidious/yt_backend/youtube_api.cr

@@ -557,6 +557,30 @@ module YoutubeAPI
     return self._post_json("/youtubei/v1/search", data, client_config)
   end
 
+  ####################################################################
+  # get_transcript(params, client_config?)
+  #
+  # Requests the youtubei/v1/get_transcript endpoint with the required headers
+  # and POST data in order to get a JSON reply.
+  #
+  # The requested data is a specially encoded protobuf string that denotes the specific language requested.
+  #
+  # An optional ClientConfig parameter can be passed, too (see
+  # `struct ClientConfig` above for more details).
+  #
+
+  def get_transcript(
+    params : String,
+    client_config : ClientConfig | Nil = nil
+  ) : Hash(String, JSON::Any)
+    data = {
+      "context" => self.make_context(client_config),
+      "params"  => params,
+    }
+
+    return self._post_json("/youtubei/v1/get_transcript", data, client_config)
+  end
+
   ####################################################################
   # _post_json(endpoint, data, client_config?)
   #