Stop diff output from corrupting the document on structural tag changes

sdiff aligned an inserted paragraph break correctly, but the renderer
wrapped the raw </p>/<p> tokens in <ins> regardless -- invalid nesting
that browsers silently repair by dropping the unmatched closing tag,
leaving everything downstream rendered inside an <ins> with nothing
left to close it. Same failure mode as the retired cacycle_diff.js,
different cause.

Tag tokens now render as a plain-text pilcrow with a tooltip naming
the actual tag, never as literal markup inside <ins>/<del>. Applies to
both inline and side-by-side. Side-by-side's new panel flattening a
real paragraph break into a marker, rather than showing its own true
structure, is a known, accepted trade-off for now -- not the same
problem, and not fixed here.

2 files changed, 47 insertions, 7 deletions

diff --git a/lib/html_word_diff.rb b/lib/html_word_diff.rb
index 1f28cf5..aa0cb86 100644
--- a/lib/html_word_diff.rb
+++ b/lib/html_word_diff.rb
@@ -1,4 +1,5 @@
 require 'diff/lcs'
+require 'cgi'
 
 module HtmlWordDiff
   TOKEN_REGEXP = /<[^>]*>|[[:space:]]+|[[:alnum:]]+|[^[:space:][:alnum:]<>]/
@@ -10,11 +11,12 @@ module HtmlWordDiff
       when '='
         html << new_token
       when '-'
-        html << "<del>#{old_token}</del>"
+        html << wrap_removed(old_token)
       when '+'
-        html << "<ins>#{new_token}</ins>"
+        html << wrap_inserted(new_token)
       when '!'
-        html << "<del>#{old_token}</del><ins>#{new_token}</ins>"
+        html << wrap_removed(old_token)
+        html << wrap_inserted(new_token)
       end
     end
     html
@@ -29,12 +31,12 @@ module HtmlWordDiff
         old_out << old_token
         new_out << new_token
       when '-'
-        old_out << "<del>#{old_token}</del>"
+        old_out << wrap_removed(old_token)
       when '+'
-        new_out << "<ins>#{new_token}</ins>"
+        new_out << wrap_inserted(new_token)
       when '!'
-        old_out << "<del>#{old_token}</del>"
-        new_out << "<ins>#{new_token}</ins>"
+        old_out << wrap_removed(old_token)
+        new_out << wrap_inserted(new_token)
       end
     end
     [old_out, new_out]
@@ -49,4 +51,24 @@ module HtmlWordDiff
   def self.tokenize(html)
     html.to_s.scan(TOKEN_REGEXP)
   end
+
+  def self.tag_token?(token)
+    token.to_s.match?(/\A<[^>]*>\z/)
+  end
+
+  def self.wrap_removed(token)
+    if tag_token?(token)
+      %(<del class="diff_structural" title="removed: #{CGI.escapeHTML(token)}">\u00b6</del>)
+    else
+      "<del>#{token}</del>"
+    end
+  end
+
+  def self.wrap_inserted(token)
+    if tag_token?(token)
+      %(<ins class="diff_structural" title="added: #{CGI.escapeHTML(token)}">\u00b6</ins>)
+    else
+      "<ins>#{token}</ins>"
+    end
+  end
 end
diff --git a/test/models/page_test.rb b/test/models/page_test.rb
index ac5691a..3868a53 100644
--- a/test/models/page_test.rb
+++ b/test/models/page_test.rb
@@ -210,4 +210,22 @@ class PageTest < ActiveSupport::TestCase
     assert_match "<del>Old</del>", old_html
     assert_match "<ins>New</ins>", new_html
   end
+
+  test "diff_against handles an inserted paragraph split without corrupting the document" do
+    n = Node.root.children.create! :slug => "paragraph_split_test"
+    d = n.find_or_create_draft @user1
+    d.body = "<p>Der Vortragsraum ist ab 19 Uhr geöffnet, der Zugang erfolgt über den Hinterhof.</p>"
+    d.save!
+    n.publish_draft!
+
+    d2 = n.find_or_create_draft @user1
+    d2.body = "<p>Der Vortragsraum ist ab 19 Uhr geöffnet,</p>\n<p>der Zugang erfolgt über den Hinterhof.</p>"
+    d2.save!
+
+    diff = d2.diff_against(n.head)
+    fragment = Nokogiri::HTML::DocumentFragment.parse(diff[:body])
+
+    assert_equal 2, fragment.css('ins.diff_structural').length
+    assert_match "der Zugang erfolgt über den Hinterhof.", fragment.text
+  end
 end