Remove unnecessary work in negotiating the encoding to use in a Regexp match.

The old approach did not align with MRI and ended negotiating the encoding twice. The result of the first negotiation was used only to detect errors and then subsequently ignored. Moreover, that additional negotiation was semantically different than what MRI does. Part of this clean-up involved implementing MRI's `rb_reg_prepare_enc` and using that for all match operations.

As an additional benefit, by synchronizing with MRI's implementation, we now handle a warning case that we did not previously, allowing us to untag some MRI tests.

Author: nirvdrum

language/regexp/encoding_spec.rb

@@ -38,6 +38,10 @@
     /#{/./}/n.match("\303\251").to_a.should == ["\303"]
   end
 
+  it "warns when using /n with a match string with non-ASCII characters and an encoding other than ASCII-8BIT" do
+    -> { /./n.match("\303\251".force_encoding('utf-8')) }.should complain(%r{historical binary regexp match /.../n against UTF-8 string})
+  end
+
   it 'uses US-ASCII as /n encoding if all chars are 7-bit' do
     /./n.encoding.should == Encoding::US_ASCII
   end
@@ -117,6 +121,15 @@
     -> { /\A[[:space:]]*\z/ =~ " ".encode("UTF-16LE") }.should raise_error(Encoding::CompatibilityError)
   end
 
+  it "raises Encoding::CompatibilityError when the regexp has a fixed, non-ASCII-compatible encoding" do
+    -> { Regexp.new("".force_encoding("UTF-16LE"), Regexp::FIXEDENCODING) =~ " ".encode("UTF-8") }.should raise_error(Encoding::CompatibilityError)
+  end
+
+  it "raises Encoding::CompatibilityError when the regexp has a fixed encoding and the match string has non-ASCII characters" do
+    -> { Regexp.new("".force_encoding("US-ASCII"), Regexp::FIXEDENCODING) =~ "\303\251".force_encoding('UTF-8') }.should raise_error(Encoding::CompatibilityError)
+  end
+
+
   it "computes the Regexp Encoding for each interpolated Regexp instance" do
     make_regexp = -> str { /#{str}/ }