Fix encoding switches when RUBYOPT is empty or only spaces

* Follow-up of dbbc3583ba which broke this.

ruby.c

@@ -895,10 +895,11 @@ moreswitches(const char *s, ruby_cmdline_options_t *opt, int envopt)
     ruby_features_t warn = opt->warn;
     int backtrace_length_limit = opt->backtrace_length_limit;
 
-    opt->src.enc.name = opt->ext.enc.name = opt->intern.enc.name = 0;
-
     while (ISSPACE(*s)) s++;
     if (!*s) return;
+
+    opt->src.enc.name = opt->ext.enc.name = opt->intern.enc.name = 0;
+
     argstr = rb_str_tmp_new((len = strlen(s)) + (envopt!=0));
     argary = rb_str_tmp_new(0);
 

spec/ruby/command_line/dash_upper_u_spec.rb

@@ -6,6 +6,13 @@ describe "ruby -U" do
               options: '-U').should == 'UTF-8'
   end
 
+  it "sets Encoding.default_internal to UTF-8 when RUBYOPT is empty or only spaces" do
+    ruby_exe('p Encoding.default_internal',
+             options: '-U', env: { 'RUBYOPT' => '' }).should == "#<Encoding:UTF-8>\n"
+    ruby_exe('p Encoding.default_internal',
+             options: '-U', env: { 'RUBYOPT' => '  ' }).should == "#<Encoding:UTF-8>\n"
+  end
+
   it "does nothing different if specified multiple times" do
      ruby_exe('print Encoding.default_internal.name',
               options: '-U -U').should == 'UTF-8'