fix(description): render custom emoji attachments in comments and descriptions

YouTube's new commentEntityPayload format carries emojis as attachmentRuns
(startIndex/length into the plain text plus an image reference), which
parse_description() ignored so far - making every custom channel emoji
disappear from comments and video descriptions.

Merge attachmentRuns with the existing commandRuns handling (they share the
same UTF-16 coordinate space):

- Images hosted on /ggpht-proxiable domains (yt3.ggpht.com custom channel
  emojis) are rendered as <img class="channel-emoji"> tags, reusing the
  proxy and CSS class from the legacy comment renderer.
- Attachments pointing at www.youtube.com (standard unicode emojis rendered
  as images by YouTube) are skipped, keeping their original characters -
  they render fine natively without contacting YouTube's servers.

Fixes #5888

spec/invidious/videos/description_attachment_runs_spec.cr

@@ -0,0 +1,50 @@
+require "../../parsers_helper.cr"
+
+Spectator.describe "parse_description" do
+  # Real InnerTube data captured from YouTube comments (2026-08).
+  # Standard emojis are attached as images hosted on www.youtube.com, while
+  # custom channel emojis are hosted on yt3.ggpht.com.
+
+  it "keeps standard emoji characters when their image is not proxyable" do
+    desc = JSON.parse(%q({"content":"Not going to lie. Your intro got me to sub!  Watching you, everyone deserves to grow❤️","attachmentRuns":[{"startIndex":84,"length":1,"element":{"type":{"imageType":{"image":{"sources":[{"url":"https://www.youtube.com/s/gaming/emoji/7ff574f2/emoji_u2764.png","width":16,"height":16}]},"playbackState":"IMAGE_PLAYBACK_STATE_STOPPED"}},"properties":{"layoutProperties":{"height":{"value":16,"unit":"DIMENSION_UNIT_POINT"},"width":{"value":16,"unit":"DIMENSION_UNIT_POINT"},"margin":{"left":{"value":2,"unit":"DIMENSION_UNIT_POINT"},"right":{"value":2,"unit":"DIMENSION_UNIT_POINT"}}},"accessibilityProperties":{"label":"❤"}}},"alignment":"ALIGNMENT_VERTICAL_CENTER"}]}))
+
+    html = parse_description(desc, "Gy3bmuzmWMM")
+
+    expect(html).to eq(%(Not going to lie. Your intro got me to sub!  Watching you, everyone deserves to grow\u{2764}\u{fe0f}))
+    expect(html.not_nil!.includes?("img")).to be_false
+  end
+
+  it "handles SMP emojis (2 UTF-16 units) right after multibyte-free text" do
+    desc = JSON.parse(%q({"content":"Feyrer is on 🔥! Thank you again, Master Jedi.","attachmentRuns":[{"startIndex":13,"length":2,"element":{"type":{"imageType":{"image":{"sources":[{"url":"https://www.youtube.com/s/gaming/emoji/7ff574f2/emoji_u1f525.png","width":16,"height":16}]},"playbackState":"IMAGE_PLAYBACK_STATE_STOPPED"}},"properties":{"layoutProperties":{"height":{"value":16,"unit":"DIMENSION_UNIT_POINT"},"width":{"value":16,"unit":"DIMENSION_UNIT_POINT"},"margin":{"left":{"value":2,"unit":"DIMENSION_UNIT_POINT"},"right":{"value":2,"unit":"DIMENSION_UNIT_POINT"}}},"accessibilityProperties":{"label":"🔥"}}},"alignment":"ALIGNMENT_VERTICAL_CENTER"}]}))
+
+    html = parse_description(desc, "Gy3bmuzmWMM")
+
+    expect(html).to eq("Feyrer is on \u{1F525}! Thank you again, Master Jedi.")
+  end
+
+  it "renders proxied channel emojis as images" do
+    # Fixture derived from a real capture: only the image URL was changed to
+    # a yt3.ggpht.com one (custom channel emoji) and its label set accordingly.
+    desc = JSON.parse(%q({"content":"Great video :_ToonThumbsUp: love it","attachmentRuns":[{"startIndex":12,"length":15,"element":{"type":{"imageType":{"image":{"sources":[{"url":"https://yt3.ggpht.com/UCxyz/aKgIYbrtFqmL8gT4576oCA=w24-h24-c-k-nd","width":24,"height":24}]}}},"properties":{"accessibilityProperties":{"label":"ToonThumbsUp"}}}}]}))
+
+    html = parse_description(desc, "Gy3bmuzmWMM")
+
+    expect(html).to eq(%(Great video <img alt="ToonThumbsUp" title="ToonThumbsUp" src="/ggpht/UCxyz/aKgIYbrtFqmL8gT4576oCA=w24-h24-c-k-nd" width="24" height="24" class="channel-emoji" /> love it))
+  end
+
+  it "renders links and channel emojis in order when both are present" do
+    # Minimal synthetic fixture: a command run without onTap stays as plain
+    # text, followed by a proxyable emoji attachment.
+    desc = JSON.parse(%q({"content":"see docs now :ToonFire:","attachmentRuns":[{"startIndex":13,"length":10,"element":{"type":{"imageType":{"image":{"sources":[{"url":"https://yt3.ggpht.com/x/fire.png"}]}}},"properties":{"accessibilityProperties":{"label":"ToonFire"}}}}],"commandRuns":[{"startIndex":4,"length":4}]}))
+
+    html = parse_description(desc, "Gy3bmuzmWMM")
+
+    expect(html).to eq(%(see docs now <img alt="ToonFire" title="ToonFire" src="/ggpht/x/fire.png" width="16" height="16" class="channel-emoji" />))
+  end
+
+  it "escapes HTML entities in the fast path" do
+    desc = JSON.parse(%q({"content":"a<b & c"}))
+
+    expect(parse_description(desc, "x")).to eq("a&lt;b &amp; c")
+  end
+end

src/invidious/videos/description.cr

@@ -40,14 +40,60 @@ private def copy_string(str : String::Builder, iter : Iterator, count : Int) : I
   return copied
 end
 
+private def skip_string(iter : Iterator, count : Int) : Nil
+  copied = 0
+  while copied < count
+    cp = iter.next
+    break if cp.is_a?(Iterator::Stop)
+
+    # A codepoint from the SMP counts twice
+    copied += 1 if cp > 0xFFFF
+    copied += 1
+  end
+end
+
+# Builds an <img> tag for a given attachment run. Only images hosted on
+# domains proxied through /ggpht are supported; attachments pointing at
+# www.youtube.com (standard unicode emojis rendered as images by YouTube)
+# return nil, as the underlying text characters already render fine.
+private def attachment_to_img(attachment : JSON::Any) : String?
+  source = attachment.dig?("element", "type", "imageType", "image", "sources", 0)
+  url = source.try &.dig?("url").try &.as_s
+  return unless url
+
+  uri = URI.parse(url)
+  host = uri.host.try &.downcase
+  return unless host.try(&.ends_with?("ggpht.com")) ||
+                host.try(&.ends_with?("googleusercontent.com"))
+
+  alt = attachment.dig?("element", "properties", "accessibilityProperties", "label").try &.as_s || ""
+  width = source.try &.dig?("width").try &.as_i || 16
+  height = source.try &.dig?("height").try &.as_i || 16
+
+  String.build do |str|
+    str << %(<img alt=") << HTML.escape(alt) << "\" "
+    str << %(title=") << HTML.escape(alt) << "\" "
+    str << %(src="/ggpht) << uri.request_target << "\" "
+    str << %(width=") << width << "\" "
+    str << %(height=") << height << "\" "
+    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
+  commands = (desc["commandRuns"]?.try &.as_a) || Array(JSON::Any).new
-  if commands.nil?
+  attachments = (desc["attachmentRuns"]?.try &.as_a) || Array(JSON::Any).new
+  # Only image attachments are supported for now
+  attachments = attachments.select do |attachment|
+    attachment.dig?("element", "type", "imageType", "image", "sources", 0, "url")
+  end
+
+  if commands.empty? && attachments.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|
@@ -60,34 +106,53 @@ 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.
+  #
+  # Links (commandRuns) and image attachments (attachmentRuns, used for custom
+  # channel emojis) share the same startIndex/length coordinate space and are
+  # merged into one ordered list of runs below.
+  runs = [] of {Int32, Int32, JSON::Any, Bool}
+  commands.each do |command|
+    runs << {command["startIndex"].as_i, command["length"].as_i, command, false}
+  end
+  attachments.each do |attachment|
+    runs << {attachment["startIndex"].as_i, attachment["length"].as_i, attachment, true}
+  end
+  runs.sort_by!(&.[0])
+
   iter = content.each_codepoint
 
   index = 0
 
   return String.build do |str|
-    commands.each do |command|
+    runs.each do |(start, length, node, is_attachment)|
-      cmd_start = command["startIndex"].as_i
+      # Defensive check against overlapping runs
-      cmd_length = command["length"].as_i
+      next if start < index
 
-      # Copy the text chunk between this command and the previous if needed.
+      # Copy the text chunk between this run and the previous one if needed.
-      length = cmd_start - index
+      gap = start - index
-      index += copy_string(str, iter, length)
+      index += copy_string(str, iter, gap)
 
-      # We need to copy the command's text using the iterator
+      if img = is_attachment ? attachment_to_img(node) : nil
-      # and the special function defined above.
+        # Skip the attached characters: the image replaces them entirely.
-      cmd_content = String.build(cmd_length) do |str2|
+        skip_string(iter, length)
-        copy_string(str2, iter, cmd_length)
+        str << img
+      else
+        # We need to copy the command's text using the iterator
+        # and the special function defined above.
+        cmd_content = String.build(length) do |str2|
+          copy_string(str2, iter, length)
+        end
+
+        link = cmd_content
+        if !is_attachment && (on_tap = node.dig?("onTap", "innertubeCommand"))
+          link = parse_link_endpoint(on_tap, cmd_content, video_id)
+        end
+        str << link
       end
-
+      index += length
-      link = cmd_content
-      if on_tap = command.dig?("onTap", "innertubeCommand")
-        link = parse_link_endpoint(on_tap, cmd_content, video_id)
-      end
-      str << link
-      index += cmd_length
     end
 
-    # Copy the end of the string (past the last command).
+    # Copy the end of the string (past the last run).
     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