tidy: prepare to remove in-context recovery hack for libxml < 2.13

This hack is not necessary with libxml 2.13 which improves fragment
recovery behavior.

Closes #2092

CHANGELOG.md

@@ -27,6 +27,7 @@ Nokogiri follows [Semantic Versioning](https://semver.org/), please see the [REA
 * [CRuby] When compiling packaged libraries from source, allow users' `AR` and `LD` environment variables to set the archiver and linker commands, respectively. This augments the existing `CC` environment variable to set the compiler command. [#3165] @ziggythehamster
 * [CRuby] The HTML5 parse methods accept a `:parse_noscript_content_as_text` keyword argument which will emulate the parsing behavior of a browser which has scripting enabled. [#3178, #3231] @stevecheckoway
 * [CRuby] `HTML5::DocumentFragment.parse` and `.new` accept a `:context` keyword argument that is the parse context node or element name. Previously this could only be passed in as a positional argument to `.new` and not at all to `.parse`. @flavorjones
+* [CRuby] The update to libxml v2.13 improves "in context" fragment parsing recovery. We removed our hacky workaround for recovery that led to silently-degraded functionality when parsing fragments with parse errors. Specifically, malformed XML fragments that used implicit namespace prefixes will now "link up" to the namespaces in the parent document or node, where previously they did not. [#2092] @flavorjones
 
 
 ### Fixed

lib/nokogiri/xml/node.rb

@@ -1086,9 +1086,11 @@ def parse(string_or_io, options = nil)
 
         error_count = document.errors.length
         node_set = in_context(contents, options.to_i)
+
         if document.errors.length > error_count
           raise document.errors[error_count] unless options.recover?
 
+          # TODO: remove this block when libxml2 < 2.13 is no longer supported
           if node_set.empty?
             # libxml2 < 2.13 does not obey the +recover+ option after encountering errors during
             # +in_context+ parsing, and so this horrible hack is here to try to emulate recovery
@@ -1115,6 +1117,7 @@ def parse(string_or_io, options = nil)
             node_set = fragment.children
           end
         end
+
         node_set
       end
 

test/xml/test_document_fragment.rb

@@ -309,6 +309,20 @@ def test_dup_creates_mutable_tree
           refute_nil(duplicate.at_css("b"))
         end
 
+        def test_in_context_fragment_parsing_recovery
+          skip("This tests behavior in libxml 2.13") unless Nokogiri.uses_libxml?(">= 2.13.0")
+
+          # https://github.com/sparklemotion/nokogiri/issues/2092
+          context_xml = "<root xmlns:n='https://example.com/foo'></root>"
+          context_doc = Nokogiri::XML::Document.parse(context_xml)
+          invalid_xml_fragment = "<n:a><b></n:a>" # note missing closing tag for `b`
+          fragment = context_doc.root.parse(invalid_xml_fragment)
+
+          assert_equal("a", fragment.first.name)
+          assert_equal("n", fragment.first.namespace.prefix)
+          assert_equal("https://example.com/foo", fragment.first.namespace.href)
+        end
+
         def test_for_libxml_in_context_fragment_parsing_bug_workaround
           skip_unless_libxml2("valgrind tests should only run with libxml2")