test: add test coverage describing how pushparser handles empty docs (#…

…3269)

**What problem is this PR intended to solve?**

The behavior is different between Java and C impls in a way that I don't
care enough to fix. Let's document the difference and move on with our
lives.

Closes #1758

**Have you included adequate test coverage?**

Yes.


**Does this change affect the behavior of either the C or the Java
implementations?**

N/A

lib/nokogiri/xml/sax/push_parser.rb

@@ -52,6 +52,9 @@ def write(chunk, last_chunk = false)
         ###
         # Finish the parsing.  This method is only necessary for
         # Nokogiri::XML::SAX::Document#end_document to be called.
+        #
+        # ⚠ Note that empty documents are treated as an error when using the libxml2-based
+        # implementation (CRuby), but are fine when using the Xerces-based implementation (JRuby).
         def finish
           write("", true)
         end

test/xml/sax/test_push_parser.rb

@@ -32,9 +32,36 @@
 
   it :test_empty_doc do
     parser.options |= Nokogiri::XML::ParseOptions::RECOVER
-    parser.write("", true)
-    assert_nil parser.document.start_elements
-    assert_nil parser.document.end_elements
+    parser.finish
+
+    assert_nil(parser.document.start_elements)
+    assert_nil(parser.document.end_elements)
+    if Nokogiri.jruby?
+      assert_empty(parser.document.errors)
+    elsif Nokogiri.uses_libxml?(">= 2.12.0") # gnome/libxml2@53050b1d
+      assert_match(/Document is empty/, parser.document.errors.first)
+    end
+    assert(parser.document.end_document_called)
+  end
+
+  it :test_empty_doc_without_recovery do
+    # behavior is different between implementations
+    # https://github.com/sparklemotion/nokogiri/issues/1758
+    if Nokogiri.jruby?
+      parser.finish
+
+      assert_nil(parser.document.start_elements)
+      assert_nil(parser.document.end_elements)
+      assert_empty(parser.document.errors)
+      assert(parser.document.end_document_called)
+    else
+      e = assert_raises(Nokogiri::XML::SyntaxError) do
+        parser.finish
+      end
+      if Nokogiri.uses_libxml?(">= 2.12.0") # gnome/libxml2@53050b1d
+        assert_match(/Document is empty/, e.message)
+      end
+    end
   end
 
   it :test_finish_should_rethrow_last_error do