fix: don't crash on utf18 autocompletion

fixes #52

Author: alexanderadam

lib/irb/completion.rb

@@ -280,7 +280,7 @@ def retrieve_completion_data(input, bind:, doc_namespace:)
           rescue EncodingError
             # ignore
           end
-          candidates.grep(/^#{Regexp.quote(sym)}/)
+          safe_grep(candidates, /^#{Regexp.quote(sym)}/)
         end
       when /^::([A-Z][^:\.\(\)]*)$/
         # Absolute Constant or class methods
@@ -291,7 +291,7 @@ def retrieve_completion_data(input, bind:, doc_namespace:)
         if doc_namespace
           candidates.find { |i| i == receiver }
         else
-          candidates.grep(/^#{Regexp.quote(receiver)}/).collect{|e| "::" + e}
+          safe_grep(candidates, /^#{Regexp.quote(receiver)}/).collect{|e| "::" + e}
         end
 
       when /^([A-Z].*)::([^:.]*)$/
@@ -378,7 +378,7 @@ def retrieve_completion_data(input, bind:, doc_namespace:)
         if doc_namespace
           all_gvars.find{ |i| i == gvar }
         else
-          all_gvars.grep(Regexp.new(Regexp.quote(gvar)))
+          safe_grep(all_gvars, Regexp.new(Regexp.quote(gvar)))
         end
 
       when /^([^.:"].*)(\.|::)([^.]*)$/
@@ -453,16 +453,28 @@ def retrieve_completion_data(input, bind:, doc_namespace:)
         else
           candidates = (bind.eval_methods | bind.eval_private_methods | bind.local_variables | bind.eval_instance_variables | bind.eval_class_constants).collect{|m| m.to_s}
           candidates |= RubyLex::RESERVED_WORDS.map(&:to_s)
-          candidates.grep(/^#{Regexp.quote(input)}/).sort
+          safe_grep(candidates, /^#{Regexp.quote(input)}/).sort
         end
       end
     end
 
     # Set of available operators in Ruby
     Operators = %w[% & * ** + - / < << <= <=> == === =~ > >= >> [] []= ^ ! != !~]
 
+    def safe_grep(candidates, pattern)
+      target_encoding = Encoding.default_external
+      candidates.filter_map do |candidate|
+        converted = candidate.encoding == target_encoding ? candidate : candidate.encode(target_encoding)
+
+        converted if pattern.match?(converted)
+      rescue Encoding::UndefinedConversionError, Encoding::InvalidByteSequenceError, Encoding::CompatibilityError, EncodingError
+        # Skip candidates that cannot be converted to the target encoding
+        nil
+      end
+    end
+
     def select_message(receiver, message, candidates, sep = ".")
-      candidates.grep(/^#{Regexp.quote(message)}/).collect do |e|
+      safe_grep(candidates, /^#{Regexp.quote(message)}/).collect do |e|
         case e
         when /^[a-zA-Z_]/
           receiver + sep + e

test/irb/test_completion.rb

@@ -343,5 +343,18 @@ def test_regexp_completor_handles_encoding_errors_gracefully
         Encoding.default_external = original_encoding
       end
     end
+
+    def test_utf16_method_name_does_not_crash
+      # Reproduces issue #52: https://github.com/ruby/irb/issues/52
+      method_name = "a".encode(Encoding::UTF_16)
+      test_obj = Object.new
+      test_obj.define_singleton_method(method_name) {}
+      test_bind = test_obj.instance_eval { binding }
+
+      completor = IRB::RegexpCompletor.new
+      assert_nothing_raised do
+        completor.completion_candidates('', 'a', '', bind: test_bind)
+      end
+    end
   end
 end