Merge 08d5420ad58dda71d004405f86a564f298f0644a into 3b21f78e99c1416f65dd3e64f1aa43b2f4dcad7a

spec/invidious/videos/description_spec.cr

@@ -0,0 +1,130 @@
+require "../../parsers_helper.cr"
+
+# Fixtures below are taken verbatim (structurally) from the InnerTube payloads
+# that Invidious receives for `commentEntityPayload.properties.content` and for
+# video `attributedDescription`s. Custom channel emoji are shipped as an image
+# in `attachmentRuns`, with only a `:_EmojiName:` placeholder (followed by
+# U+25A1 padding) left behind in `content`. Plain unicode emoji are the opposite:
+# they are already in `content` and their attachment run points at YouTube's own
+# emoji sprite on `www.youtube.com`, which cannot be served through /ggpht.
+private def attachment_run(label : String, start_index : Int32, length : Int32, url : String) : String
+  <<-JSON
+    {
+      "startIndex": #{start_index},
+      "length": #{length},
+      "element": {
+        "type": {
+          "imageType": {
+            "image": {
+              "sources": [{"url": "#{url}", "width": 16, "height": 16}]
+            }
+          }
+        },
+        "properties": {
+          "accessibilityProperties": {"label": "#{label}"}
+        }
+      },
+      "alignment": "ALIGNMENT_VERTICAL_CENTER"
+    }
+    JSON
+end
+
+private def command_run(start_index : Int32, length : Int32, url : String) : String
+  <<-JSON
+    {
+      "startIndex": #{start_index},
+      "length": #{length},
+      "onTap": {"innertubeCommand": {"urlEndpoint": {"url": "#{url}"}}}
+    }
+    JSON
+end
+
+CUSTOM_EMOJI_URL = "https://yt3.googleusercontent.com/6-Hzg3v1TFZI9DifzoYrJZYYTsu4quApCRT461xkMjDhaR4xUP8FbsDZe4xIdcTQi_oGPOg4=s16-w24-h24-c-k-nd"
+# YouTube's own emoji sprite, served from www.youtube.com (not /ggpht-able).
+UNICODE_EMOJI_URL = "https://www.youtube.com/s/gaming/emoji/7ff574f2/emoji_u1f600.png"
+
+Spectator.describe "parse_description" do
+  it "returns an empty string for nil content" do
+    expect(parse_description(nil, "vid")).to eq("")
+  end
+
+  it "returns an empty string for empty content" do
+    expect(parse_description(JSON.parse(%({"content": ""})), "vid")).to eq("")
+  end
+
+  it "escapes plain content when there are no runs" do
+    desc = JSON.parse(%({"content": "a & b <c>"}))
+    expect(parse_description(desc, "vid")).to eq("a &amp; b &lt;c&gt;")
+  end
+
+  it "renders custom channel emoji as proxied images" do
+    content = "hi :_EmojiOne:\u25A1\u25A1 and :_EmojiTwo: end"
+    desc = JSON.parse(
+      %({"content": "#{content}", "attachmentRuns": [) +
+      attachment_run("EmojiOne", 3, 11, CUSTOM_EMOJI_URL) + "," +
+      attachment_run("EmojiTwo", 21, 11, CUSTOM_EMOJI_URL) + "]}"
+    )
+
+    html = parse_description(desc, "vid").to_s
+
+    expect(html).to eq(
+      %(hi ) +
+      %(<img alt="EmojiOne" src="/ggpht/6-Hzg3v1TFZI9DifzoYrJZYYTsu4quApCRT461xkMjDhaR4xUP8FbsDZe4xIdcTQi_oGPOg4=s16-w24-h24-c-k-nd" title="EmojiOne" class="channel-emoji" />) +
+      %( and ) +
+      %(<img alt="EmojiTwo" src="/ggpht/6-Hzg3v1TFZI9DifzoYrJZYYTsu4quApCRT461xkMjDhaR4xUP8FbsDZe4xIdcTQi_oGPOg4=s16-w24-h24-c-k-nd" title="EmojiTwo" class="channel-emoji" />) +
+      %( end)
+    )
+  end
+
+  it "does not leave placeholders or padding behind for custom emoji" do
+    content = "hi :_EmojiOne:\u25A1\u25A1"
+    desc = JSON.parse(
+      %({"content": "#{content}", "attachmentRuns": [) +
+      attachment_run("EmojiOne", 3, 11, CUSTOM_EMOJI_URL) + "]}"
+    )
+
+    html = parse_description(desc, "vid").to_s
+
+    expect(html).to_not contain(":_EmojiOne:")
+    expect(html).to_not contain("\u25A1")
+  end
+
+  it "leaves plain unicode emoji alone instead of substituting an image" do
+    content = "Smiley: \u{1F600} ok"
+    desc = JSON.parse(
+      %({"content": "#{content}", "attachmentRuns": [) +
+      attachment_run("\u{1F600}", 8, 2, UNICODE_EMOJI_URL) + "]}"
+    )
+
+    html = parse_description(desc, "vid").to_s
+
+    expect(html).to eq("Smiley: \u{1F600} ok")
+    expect(html).to_not contain("<img")
+  end
+
+  it "still renders commandRuns (links and timestamps)" do
+    desc = JSON.parse(
+      %({"content": "see here", "commandRuns": [) +
+      command_run(4, 4, "https://example.com/x") + "]}"
+    )
+
+    html = parse_description(desc, "vid").to_s
+
+    expect(html).to contain(%(<a href="https://example.com/x">here</a>))
+  end
+
+  it "renders commandRuns and attachmentRuns interleaved in index order" do
+    desc = JSON.parse(
+      %({"content": "0:48 :_E1: tail", "commandRuns": [) +
+      command_run(0, 4, "https://example.com/t") +
+      %(], "attachmentRuns": [) +
+      attachment_run("E1", 5, 5, CUSTOM_EMOJI_URL) + "]}"
+    )
+
+    html = parse_description(desc, "vid").to_s
+
+    expect(html).to contain(%(<a href="https://example.com/t">0:48</a>))
+    expect(html).to contain(%(class="channel-emoji" />))
+    expect(html).to contain(" tail")
+  end
+end

src/invidious/comments/youtube.cr

@@ -150,7 +150,7 @@ module Invidious::Comments
                   toolbar_mutation = mutations.find { |i| i.dig?("entityKey") == toolbar_key }
 
                   if !comment_mutation.nil? && !toolbar_mutation.nil?
-                    # todo parse styleRuns, commandRuns and attachmentRuns for comments
+                    # todo parse styleRuns for comments
                     html_content = parse_description(comment_mutation.dig("payload", "commentEntityPayload", "properties", "content"), id)
                     comment_author = comment_mutation.dig("payload", "commentEntityPayload", "author")
                     json.field "authorId", comment_author["channelId"].as_s

src/invidious/videos/description.cr

@@ -1,3 +1,4 @@
+require "html"
 require "json"
 require "uri"
 
@@ -12,6 +13,20 @@ private def utf16_length(content : String) : Int32
   u16_size
 end
 
+private def skip_string(iter : Iterator, count : Int) : Int
+  skipped = 0
+  while skipped < count
+    cp = iter.next
+    break if cp.is_a?(Iterator::Stop)
+
+    # A codepoint from the SMP counts twice
+    skipped += 1 if cp > 0xFFFF
+    skipped += 1
+  end
+
+  return skipped
+end
+
 private def copy_string(str : String::Builder, iter : Iterator, count : Int) : Int
   copied = 0
   while copied < count
@@ -40,19 +55,60 @@ private def copy_string(str : String::Builder, iter : Iterator, count : Int) : I
   return copied
 end
 
+# YouTube ships custom channel emoji as an image inside an `attachmentRuns`
+# entry, while `content` only holds a `:_EmojiName:` placeholder followed by
+# U+25A1 padding characters. Plain unicode emoji are the opposite: they are
+# already present verbatim in `content`, and their attachment run points at
+# YouTube's own emoji sprite (`www.youtube.com/s/gaming/emoji/...`), which the
+# `/ggpht` proxy cannot serve. Therefore only the former are substituted with
+# an image tag; the latter are left as the text characters they already are.
+private def attachment_run_to_html(run : JSON::Any) : String?
+  image = run.dig?("element", "type", "imageType", "image")
+  return if image.nil?
+
+  sources = image["sources"].as_a?
+  return if sources.nil? || sources.empty?
+
+  image_url = sources[0]["url"].as_s?
+  return if image_url.nil?
+
+  uri = URI.parse(image_url)
+  return unless uri.host.try &.ends_with?("googleusercontent.com")
+
+  alt = run.dig?("element", "properties", "accessibilityProperties", "label")
+    .try &.as_s || ""
+  alt = HTML.escape(alt)
+
+  return String.build do |str|
+    str << %(<img alt=") << alt << "\" "
+    str << %(src="/ggpht) << uri.request_target << "\" "
+    str << %(title=") << alt << "\" "
+    str << %(class="channel-emoji" />)
+  end
+end
+
 def parse_description(desc, video_id : String) : String?
   return "" if desc.nil?
 
   content = desc["content"].as_s
   return "" if content.empty?
 
-  commands = desc["commandRuns"]?.try &.as_a
-  if commands.nil?
-    # Slightly faster than HTML.escape, as we're only doing one pass on
-    # the string instead of five for the standard library
-    return String.build do |str|
-      content_size = content.ascii_only? ? content.size : utf16_length(content)
-      copy_string(str, content.each_codepoint, content_size)
+  # Both `commandRuns` (links, timestamps, hashtags) and `attachmentRuns`
+  # (emoji) are indexed in UTF-16 code units over `content`, and the two sets
+  # can be interleaved, so they have to be rendered in a single ordered pass.
+  runs = [] of NamedTuple(start: Int32, length: Int32, node: JSON::Any, kind: Symbol)
+
+  desc["commandRuns"]?.try &.as_a.try do |command_runs|
+    command_runs.each do |run|
+      runs << {start: run["startIndex"].as_i, length: run["length"].as_i,
+               node: run, kind: :command}
+    end
+  end
+
+  desc["attachmentRuns"]?.try &.as_a.try do |attachment_runs|
+    attachment_runs.each do |run|
+      runs << {start: run["startIndex"].as_i, length: run["length"].as_i,
+               node: run, kind: :attachment}
     end
   end
 
@@ -60,36 +116,79 @@ def parse_description(desc, video_id : String) : String?
   # (0x10000 and above) are encoded as UTF-16 surrogate pairs, which are
   # automatically decoded by the JSON parser. It means that we need to count
   # copied byte in a special manner, preventing the use of regular string copy.
+  content_size = content.ascii_only? ? content.size : utf16_length(content)
+
+  if runs.empty?
+    # Slightly faster than HTML.escape, as we're only doing one pass on
+    # the string instead of five for the standard library
+    return String.build do |str|
+      copy_string(str, content.each_codepoint, content_size)
+    end
+  end
+
+  runs.sort_by! { |run| run[:start] }
+
   iter = content.each_codepoint
 
   index = 0
+  strip_padding = false
 
   return String.build do |str|
-    commands.each do |command|
-      cmd_start = command["startIndex"].as_i
-      cmd_length = command["length"].as_i
+    runs.each do |run|
+      # Skip runs that overlap an already rendered one: consuming their span
+      # again would advance the iterator past the text they cover.
+      next if run[:start] < index
 
-      # Copy the text chunk between this command and the previous if needed.
-      length = cmd_start - index
-      index += copy_string(str, iter, length)
+      # Copy the text chunk between this run and the previous one.
+      chunk = String.build do |str2|
+        copy_string(str2, iter, run[:start] - index)
+      end
 
+      # Drop the placeholder padding that follows a substituted channel emoji.
+      if strip_padding
+        chunk = chunk.sub(/\A\x{25A1}+/, "")
+        strip_padding = false
+      end
+
+      str << chunk
+      index = run[:start]
+
+      case run[:kind]
+      when :command
         # We need to copy the command's text using the iterator
         # and the special function defined above.
-      cmd_content = String.build(cmd_length) do |str2|
-        copy_string(str2, iter, cmd_length)
+        cmd_content = String.build(run[:length]) do |str2|
+          copy_string(str2, iter, run[:length])
         end
 
         link = cmd_content
-      if on_tap = command.dig?("onTap", "innertubeCommand")
+        if on_tap = run[:node].dig?("onTap", "innertubeCommand")
           link = parse_link_endpoint(on_tap, cmd_content, video_id)
         end
         str << link
-      index += cmd_length
+      when :attachment
+        if emoji = attachment_run_to_html(run[:node])
+          skip_string(iter, run[:length])
+          str << emoji
+          strip_padding = true
+        else
+          # Not a channel emoji: the span is already the emoji itself.
+          copy_string(str, iter, run[:length])
+        end
       end
 
-    # Copy the end of the string (past the last command).
-    content_size = content.ascii_only? ? content.size : utf16_length(content)
-    remaining_length = content_size - index
-    copy_string(str, iter, remaining_length) if remaining_length > 0
+      index += run[:length]
+    end
+
+    # Copy the end of the string (past the last run).
+    chunk = String.build do |str2|
+      copy_string(str2, iter, content_size - index)
+    end
+
+    if strip_padding
+      chunk = chunk.sub(/\A\x{25A1}+/, "")
+    end
+
+    str << chunk
   end
 end