gistfile1.diff

diff --git a/re.c b/re.c
index 523ba17..0df6349 100644
--- a/re.c
+++ b/re.c
@@ -696,34 +696,56 @@ rb_reg_matcher_new(VALUE re, VALUE str)
                u_errorName(status));
     }
 
-    long chars_len = 0;
-    UChar *chars = rb_str_xcopy_uchars(str, &chars_len);
-
-    if (chars_len == 0) {
-       // uregex_setText() will complain if we pass a NULL pattern or a
-       // pattern length of 0, so we do pass an empty pattern with a length
-       // of -1 which indicates it's terminated by \0.
-       chars = (UChar *)xmalloc(sizeof(UChar));
-       *chars = '\0';
-       chars_len = -1;
+    // Fast path when applying a regexp on an UTF-8 encoded text string.
+    // (Only if ICU is 4.6 or higher).
+    bool need_uchars = true;
+#if U_ICU_VERSION_MAJOR_NUM >= 4 && U_ICU_VERSION_MINOR_NUM >= 6
+    if (IS_RSTR(str)
+           && (IS_UTF8_ENC(RSTR(str)->encoding)
+               || IS_ASCII_ENC(RSTR(str)->encoding))) {
+       UText *text = utext_openUTF8(NULL, RSTR(str)->bytes,
+               RSTR(str)->length_in_bytes, &status);
+       if (status == U_ZERO_ERROR) {
+           uregex_setUText(match_pattern, text, &status);
+           utext_close(text);
+           if (status == U_ZERO_ERROR) {
+               need_uchars = false;
+           }
+       }
+       status = U_ZERO_ERROR;
     }
+#endif
+
+    // Slow path, converting the text string into a buffer of uchars.
+    if (need_uchars) {
+       long chars_len = 0;
+       UChar *chars = rb_str_xcopy_uchars(str, &chars_len);
+
+       if (chars_len == 0) {
+           // uregex_setText() will complain if we pass a NULL pattern or a
+           // pattern length of 0, so we do pass an empty pattern with a
+           // length of -1 which indicates it's terminated by \0.
+           chars = (UChar *)xmalloc(sizeof(UChar));
+           *chars = '\0';
+           chars_len = -1;
+       }
 
-    uregex_setText(match_pattern, chars, chars_len, &status);
+       uregex_setText(match_pattern, chars, chars_len, &status);
 
-    if (status != U_ZERO_ERROR) {
-       uregex_close(match_pattern);
-       rb_raise(rb_eRegexpError, "can't set pattern text: %s",
-               u_errorName(status));   
+       if (status != U_ZERO_ERROR) {
+           uregex_close(match_pattern);
+           rb_raise(rb_eRegexpError, "can't set pattern text: %s",
+                   u_errorName(status));       
+       }
+       // Apparently uregex_setText doesn't copy the given string, so we need
+       // to keep it around until we finally destroy the matcher object.
+       GC_WB(&matcher->text_chars, chars);
     }
 
     matcher->pattern = match_pattern;
     matcher->frozen_str = 0; // set lazily
     GC_WB(&matcher->orig_str, str);
 
-    // Apparently uregex_setText doesn't copy the given string, so we need
-    // to keep it around until we finally destroy the matcher object.
-    GC_WB(&matcher->text_chars, chars);
-
     return (VALUE)matcher;
 }
 

gistfile2.txt

$ time ./miniruby -e "s=File.read('parse.c'); 1000.times { s.match(/./) }"

real	0m3.868s
user	0m3.863s
sys	0m0.197s
$ time macruby -e "s=File.read('parse.c'); 1000.times { s.match(/./) }"

real	0m3.269s
user	0m2.907s
sys	0m0.550s