Fix String#index and #rindex for Char::REPLACEMENT

spec/std/string_spec.cr

@@ -957,13 +957,16 @@ describe "String" do
       it { "日本語".index('本').should eq(1) }
       it { "bar".index('あ').should be_nil }
       it { "あいう_えお".index('_').should eq(3) }
+      it { "xyz\xFFxyz".index('\u{FFFD}').should eq(3) }
 
       describe "with offset" do
         it { "foobarbaz".index('a', 5).should eq(7) }
         it { "foobarbaz".index('a', -4).should eq(7) }
         it { "foo".index('g', 1).should be_nil }
         it { "foo".index('g', -20).should be_nil }
         it { "日本語日本語".index('本', 2).should eq(4) }
+        it { "xyz\xFFxyz".index('\u{FFFD}', 2).should eq(3) }
+        it { "xyz\xFFxyz".index('\u{FFFD}', 4).should be_nil }
 
         # Check offset type
         it { "foobarbaz".index('a', 5_i64).should eq(7) }
@@ -1106,6 +1109,7 @@ describe "String" do
       it { "foobar".rindex('g').should be_nil }
       it { "日本語日本語".rindex('本').should eq(4) }
       it { "あいう_えお".rindex('_').should eq(3) }
+      it { "xyz\xFFxyz".rindex('\u{FFFD}').should eq(3) }
 
       describe "with offset" do
         it { "bbbb".rindex('b', 2).should eq(2) }
@@ -1118,6 +1122,8 @@ describe "String" do
         it { "faobar".rindex('a', 3).should eq(1) }
         it { "faobarbaz".rindex('a', -3).should eq(4) }
         it { "日本語日本語".rindex('本', 3).should eq(1) }
+        it { "xyz\xFFxyz".rindex('\u{FFFD}', 4).should eq(3) }
+        it { "xyz\xFFxyz".rindex('\u{FFFD}', 2).should be_nil }
 
         # Check offset type
         it { "bbbb".rindex('b', 2_i64).should eq(2) }

src/string.cr

@@ -3349,11 +3349,21 @@ class String
   def index(search : Char, offset = 0) : Int32?
     # If it's ASCII we can delegate to slice
     if single_byte_optimizable?
-      # With `single_byte_optimizable?` there are only ASCII characters and invalid UTF-8 byte
-      # sequences and we can immediately reject any non-ASCII codepoint.
-      return unless search.ascii?
+      # With `single_byte_optimizable?` there are only ASCII characters and
+      # invalid UTF-8 byte sequences, and we can reject anything that is neither
+      # ASCII nor the replacement character.
+      case search
+      when .ascii?
+        return to_slice.fast_index(search.ord.to_u8!, offset)
+      when Char::REPLACEMENT
+        offset.upto(bytesize - 1) do |i|
+          if to_unsafe[i] >= 0x80
+            return i.to_i
+          end
+        end
+      end
 
-      return to_slice.fast_index(search.ord.to_u8, offset)
+      return nil
     end
 
     offset += size if offset < 0
@@ -3469,11 +3479,21 @@ class String
   def rindex(search : Char, offset = size - 1)
     # If it's ASCII we can delegate to slice
     if single_byte_optimizable?
-      # With `single_byte_optimizable?` there are only ASCII characters and invalid UTF-8 byte
-      # sequences and we can immediately reject any non-ASCII codepoint.
-      return unless search.ascii?
+      # With `single_byte_optimizable?` there are only ASCII characters and
+      # invalid UTF-8 byte sequences, and we can reject anything that is neither
+      # ASCII nor the replacement character.
+      case search
+      when .ascii?
+        return to_slice.rindex(search.ord.to_u8!, offset)
+      when Char::REPLACEMENT
+        offset.downto(0) do |i|
+          if to_unsafe[i] >= 0x80
+            return i.to_i
+          end
+        end
+      end
 
-      return to_slice.rindex(search.ord.to_u8, offset)
+      return nil
     end
 
     offset += size if offset < 0