FEATURE: Optional strip_trailing_invisibles config flag

lib/markbridge.rb

@@ -156,7 +156,19 @@ def build_renderer(tag_library:)
       Renderers::Discourse::Renderer.new(tag_library:, escaper:)
     end
 
+    # Trailing-invisibles set, applied only when opted into via
+    # `Markbridge.configuration.strip_trailing_invisibles = true`.
+    # NBSP (U+00A0) plus the zero-width format chars that render as
+    # nothing — ZWSP U+200B, ZWNJ U+200C, ZWJ U+200D, WJ U+2060,
+    # ZWNBSP/BOM U+FEFF. Deliberately excludes ASCII space and tab so
+    # Markdown's "two trailing spaces = hard line break" rule still
+    # works. The `$` anchors to end-of-line (default Ruby regex mode),
+    # so this strips per line without consuming the line break.
+    TRAILING_INVISIBLE_RE = /[\u00A0\u200B\u200C\u200D\u2060\uFEFF]+$/
+    private_constant :TRAILING_INVISIBLE_RE
+
     def cleanup_markdown(text)
+      text = text.gsub(TRAILING_INVISIBLE_RE, "") if configuration.strip_trailing_invisibles
       text
         .gsub(/\n{3,}/, "\n\n") # Max 2 consecutive newlines
         .gsub(/^[ \t]+$/, "") # Remove whitespace-only lines

lib/markbridge/configuration.rb

@@ -2,10 +2,18 @@
 
 module Markbridge
   class Configuration
-    attr_accessor :escape_hard_line_breaks
+    attr_accessor :escape_hard_line_breaks, :strip_trailing_invisibles
 
     def initialize
       @escape_hard_line_breaks = false
+      # When true, `cleanup_markdown` rstrips a small set of invisible
+      # characters (NBSP, ZWSP, ZWNJ, ZWJ, WJ, ZWNBSP/BOM) at each line
+      # end. Useful for cleaning Outlook/Word HTML exports where these
+      # show up as soft-break hints and spacer-paragraph fillers. Adds
+      # one regex pass over the rendered output (~4-5% slowdown on a
+      # mixed-content benchmark), so default off; opt in if the polish
+      # matters more than throughput.
+      @strip_trailing_invisibles = false
     end
   end
 end

spec/markbridge_spec.rb

@@ -231,6 +231,44 @@ def render(_element, _interface, **_kwargs)
     it "strips leading and trailing whitespace from the final output" do
       expect(described_class.bbcode_to_markdown("   hi   ")).to eq("hi")
     end
+
+    it "preserves trailing invisible characters by default (config flag off)" do
+      # Default config keeps the output as-is — no extra regex pass on
+      # cleanup. The U+200B byte sequence (e2 80 8b in UTF-8) survives
+      # at the end of the rendered output.
+      result = described_class.html_to_markdown("<p>Hello&#8203;</p>")
+
+      expect(result.bytes.last(4)).to eq([0x6f, 0xe2, 0x80, 0x8b]) # "o" + ZWSP
+    end
+
+    context "with strip_trailing_invisibles enabled" do
+      before { described_class.configure { |c| c.strip_trailing_invisibles = true } }
+
+      it "strips a trailing zero-width space at the end of a paragraph" do
+        # Outlook-style soft-break ZWSP after content.
+        expect(described_class.html_to_markdown("<p>Hello&#8203;</p>")).to eq("Hello")
+      end
+
+      it "drops nbsp-only spacer paragraphs by stripping their trailing nbsp" do
+        # `<p>&nbsp;</p>` collapses to "<nbsp>\n\n"; rstripping the nbsp
+        # leaves an empty line which the existing \n{3,} collapse drops.
+        expect(described_class.html_to_markdown("<p>a</p><p>&nbsp;</p><p>b</p>")).to eq("a\n\nb")
+      end
+
+      it "preserves trailing ASCII spaces — they are the Markdown hard-break syntax" do
+        # `hello  \nworld` (two trailing spaces) is the hard-line-break form;
+        # the trailing-invisibles strip must not touch ASCII spaces.
+        expect(described_class.bbcode_to_markdown("hello  \nworld")).to eq("hello  \nworld")
+      end
+
+      it "strips trailing invisibles on every affected line, not just the first" do
+        # gsub vs sub: with sub, only the first paragraph would get cleaned
+        # and the second's ZWSP would leak through.
+        result = described_class.html_to_markdown("<p>first&#8203;</p><p>second&#8203;</p>")
+
+        expect(result).to eq("first\n\nsecond")
+      end
+    end
   end
 
   describe ".parse_html" do

spec/system/html_to_markdown_spec.rb

@@ -286,6 +286,45 @@
     end
   end
 
+  describe "trailing invisible characters with strip_trailing_invisibles" do
+    around do |example|
+      Markbridge.configure { |c| c.strip_trailing_invisibles = true }
+      example.run
+    ensure
+      Markbridge.reset_defaults!
+    end
+
+    it "strips trailing zero-width space at the end of a paragraph" do
+      html = "<p>Hello Specialist&#8203;</p><p>Our customer are unhappy</p>"
+
+      result = Markbridge.html_to_markdown(html)
+      expect(result).to eq("Hello Specialist\n\nOur customer are unhappy")
+    end
+
+    it "drops an Outlook-style nbsp-only spacer paragraph between content" do
+      html = '<p>before</p><p class="MsoNormal">&nbsp;</p><p>after</p>'
+
+      result = Markbridge.html_to_markdown(html)
+      expect(result).to eq("before\n\nafter")
+    end
+
+    it "preserves leading nbsp (author intent — used as indentation)" do
+      html = "<p>&nbsp;Hello</p>"
+
+      result = Markbridge.html_to_markdown(html)
+      expect(result).to eq(" Hello")
+    end
+
+    it "preserves invisibles in the middle of content" do
+      # Mid-content ZWSP is a meaningful soft-break hint (long URLs, CJK),
+      # only line-end invisibles get stripped.
+      html = "<p>before​inline​text</p>"
+
+      result = Markbridge.html_to_markdown(html)
+      expect(result).to eq("before​inline​text")
+    end
+  end
+
   describe "complex combinations" do
     it "converts mixed content" do
       html = <<~HTML

spec/unit/markbridge/configuration_spec.rb

@@ -13,4 +13,15 @@
       expect(configuration.escape_hard_line_breaks).to be true
     end
   end
+
+  describe "#strip_trailing_invisibles" do
+    it "defaults to false" do
+      expect(configuration.strip_trailing_invisibles).to be false
+    end
+
+    it "can be set to true" do
+      configuration.strip_trailing_invisibles = true
+      expect(configuration.strip_trailing_invisibles).to be true
+    end
+  end
 end