Fix CodeRabbit review findings

- Set description_node to nil for absent interactiveTabbedHeaderRenderer
  so description_node.as_s does not TypeCastError (PR #5984 finding)
- Validate byline browseId as non-empty string before selecting channel run
- Token-aware view-count/relative-date metadata matching in lockupViewModel
  parser prevents author rows like 'Chicago' from being misread
- linkify_hashtags accepts raw title, HTML-escapes internally, preventing
  false /hashtag/39 links from escaped apostrophes; template passes raw title
- Add regression tests for hashtag apostrophe handling

spec/invidious/utils_spec.cr

@@ -43,4 +43,17 @@ Spectator.describe "Utils" do
       expect(decode_date("8 years ago")).to be_close(Time.utc - 8.years, 500.milliseconds)
     end
   end
+
+  describe "linkify_hashtags" do
+    it "escapes apostrophes without creating false hashtag links" do
+      result = linkify_hashtags("Rock 'n' Roll `#music`")
+      expect(result).to_not contain("/hashtag/39")
+      expect(result).to contain("'")
+      expect(result).to contain("<a href=\"/hashtag/music\">#music</a>")
+    end
+
+    it "links a plain hashtag" do
+      expect(linkify_hashtags("hello #music")).to eq("hello <a href=\"/hashtag/music\">#music</a>")
+    end
+  end
 end

src/invidious/channels/about.cr

@@ -75,7 +75,7 @@ def get_about_info(ucid) : AboutChannel
         author_url = "https://www.youtube.com/channel/#{ucid}"
         author_thumbnail = phr_vm.try &.dig?("animatedImage", "contentPreviewImageViewModel", "image", "sources", 0, "url").try &.as_s || ""
         banner = nil
-        description_node = JSON.parse(%({"simpleText": ""}))
+        description_node = nil
         tags = [] of String
       else
         author = ithr["title"]["simpleText"].as_s

src/invidious/helpers/utils.cr

@@ -386,15 +386,15 @@ def parse_link_endpoint(endpoint : JSON::Any, text : String, video_id : String)
   return text
 end
 
-def linkify_hashtags(title : String) : String
+def linkify_hashtags(raw_title : String) : String
   String.build do |str|
     pos = 0
-    title.scan(/#([a-zA-Z0-9_]+)/) do |match|
-      str << title[pos...match.begin(0)]
-      str << %(<a href="/hashtag/#{URI.encode_path(match[1])}">#{match[0]}</a>)
+    raw_title.scan(/#([a-zA-Z0-9_]+)/) do |match|
+      str << HTML.escape(raw_title[pos...match.begin(0)])
+      str << %(<a href="/hashtag/#{URI.encode_path(match[1])}">#{HTML.escape(match[0])}</a>)
       pos = match.end(0)
     end
-    str << title[pos..]
+    str << HTML.escape(raw_title[pos..])
   end
 end
 

src/invidious/videos/parser.cr

@@ -20,14 +20,16 @@ module Invidious::Videos::Parser
     end
 
     # Both have "short", so the "long" option shouldn't be required
-    # Use the first run that has a valid browse endpoint.
+    # Use the first run that has a non-empty string browseId.
     # For collaboration videos, the first run is a plain text summary
     # without a link, while subsequent runs contain the channel links.
     byline_runs = (related["shortBylineText"]? || related["longBylineText"]?)
       .try &.dig?("runs").try &.as_a
 
     channel_info = byline_runs.try &.find do |run|
-      run.dig?("navigationEndpoint", "browseEndpoint", "browseId")
+      browse_id = run.dig?("navigationEndpoint", "browseEndpoint", "browseId")
+      text_id = browse_id.try &.as_s?
+      !text_id.nil? && !text_id.empty?
     end || byline_runs.try &.[0]?
 
     author = channel_info.try &.dig?("text")

src/invidious/views/watch.ecr

@@ -94,7 +94,7 @@ we're going to need to do it here in order to allow for translations.
 <div class="h-box">
     <h1>
         <% if video.title.matches?(/#\w/) %>
-            <%= linkify_hashtags(title) %>
+            <%= linkify_hashtags(video.title) %>
         <% else %>
             <%= title %>
         <% end %>

src/invidious/yt_backend/extractors.cr

@@ -642,6 +642,13 @@ private module Parsers
     extend self
     include BaseParser
 
+    private VIEW_COUNT_TOKEN = /\d+\s+views/
+    private RELATIVE_DATE_TOKEN = /\d+\s+(second|minute|hour|day|week|month|year)s?\s+ago/
+
+    private def metadata_part_text(item : JSON::Any) : String?
+      item.dig?("text", "content").try &.as_s
+    end
+
     def process(item : JSON::Any, author_fallback : AuthorFallback)
       if item_contents = item["lockupViewModel"]?
         return self.parse(item_contents, author_fallback)
@@ -664,14 +671,16 @@ private module Parsers
         # Contains the views of the video and the published time of the video.
         # For collaboration videos, the first row contains the author names
         # instead, so we scan all rows for the one with view/publish info.
+        # Token-aware matching prevents author rows (e.g. "Chicago") from
+        # being mistaken for view-count or relative-date metadata.
         metadata_parts = metadata.dig?("metadata", "contentMetadataViewModel", "metadataRows")
           .try &.as_a
             .compact_map { |row| row.dig?("metadataParts").try &.as_a }
-            .find { |parts| parts.any? { |item| item.dig?("text", "content").try &.as_s.includes?("views") || item.dig?("text", "content").try &.as_s.includes?("ago") } }
+            .find { |parts| parts.any? { |item| metadata_part_text(item).try { |t| t.matches?(VIEW_COUNT_TOKEN) || t.matches?(RELATIVE_DATE_TOKEN) } } }
 
-        view_count_text = metadata_parts.try &.find { |item| item["icon"]?.nil? && item.dig?("text", "content").try &.as_s.includes?("views") }
+        view_count_text = metadata_parts.try &.find { |item| item["icon"]?.nil? && metadata_part_text(item).try(&.matches?(VIEW_COUNT_TOKEN)) }
           .try &.dig("text", "content").as_s
-        published = metadata_parts.try &.find { |item| item["icon"]?.nil? && item.dig?("text", "content").try &.as_s.includes?("ago") }
+        published = metadata_parts.try &.find { |item| item["icon"]?.nil? && metadata_part_text(item).try(&.matches?(RELATIVE_DATE_TOKEN)) }
           .try { |item| decode_date(item.dig("text", "content").as_s) } || Time.local
 
         view_count = short_text_to_number(view_count_text || "0")