add ASCII-only option, to mimic default RE2 behaviour

This is a workaround, motivated by the difference in handling non-valid UTF8
bytes that Oniriguma has, compared to Go's default RE2.

See src-d/enry#225 (comment)

Summary of changes:
 - c: prevent `NewOnigRegex()` from hard-coding UTF8
 - c: `NewOnigRegex()` now propely calls to `onig_initialize()` [1]
 - go: expose new `MustCompileASCII()` \w default charecter class matching only ASCII
 - go: `MustCompile()` refactored, `initRegexp()` extracted for common UTF8/ASCII logic

Encoding was not exposed on Go API level intentionaly for simplisity,
in order to avoid introducing complex struct type [2] to API surface.

 1. https://github.com/kkos/oniguruma/blob/83572e983928243d741f61ac290fc057d69fefc3/doc/API#L6
 2. https://github.com/kkos/oniguruma/blob/83572e983928243d741f61ac290fc057d69fefc3/src/oniguruma.h#L121

Signed-off-by: Alexander Bezzubov <[email protected]>

chelper.c

@@ -17,7 +17,8 @@ int NewOnigRegex( char *pattern, int pattern_length, int option,
     *error_info = (OnigErrorInfo *) malloc(sizeof(OnigErrorInfo));
     memset(*error_info, 0, sizeof(OnigErrorInfo));
 
-    *encoding = (void*)ONIG_ENCODING_UTF8;
+    OnigEncoding use_encs[] = { *encoding };
+    onig_initialize(use_encs, sizeof(use_encs)/sizeof(use_encs[0]));
 
     *error_buffer = (char*) malloc(ONIG_MAX_ERROR_MESSAGE_LEN * sizeof(char));
 

regex.go

@@ -48,13 +48,22 @@ type Regexp struct {
 }
 
 func NewRegexp(pattern string, option int) (re *Regexp, err error) {
-	re = &Regexp{pattern: pattern}
-	patternCharPtr := C.CString(pattern)
-	defer C.free(unsafe.Pointer(patternCharPtr))
+	re = &Regexp{pattern: pattern, encoding: C.ONIG_ENCODING_UTF8}
+	return initRegexp(re, option)
+}
+
+func NewRegexpASCII(pattern string, option int) (re *Regexp, err error) {
+	re = &Regexp{pattern: pattern, encoding: C.ONIG_ENCODING_ASCII}
+	return initRegexp(re, option)
+}
 
+func initRegexp(re *Regexp, option int) (*Regexp, error) {
+	var err error
+	patternCharPtr := C.CString(re.pattern)
+	defer C.free(unsafe.Pointer(patternCharPtr))
 	mutex.Lock()
 	defer mutex.Unlock()
-	error_code := C.NewOnigRegex(patternCharPtr, C.int(len(pattern)), C.int(option), &re.regex, &re.region, &re.encoding, &re.errorInfo, &re.errorBuf)
+	error_code := C.NewOnigRegex(patternCharPtr, C.int(len(re.pattern)), C.int(option), &re.regex, &re.region, &re.encoding, &re.errorInfo, &re.errorBuf)
 	if error_code != C.ONIG_NORMAL {
 		err = errors.New(C.GoString(re.errorBuf))
 	} else {
@@ -95,6 +104,15 @@ func MustCompileWithOption(str string, option int) *Regexp {
 	return regexp
 }
 
+// MustCompileASCII equivalent of MustCompile but with char matching only ASCII.
+func MustCompileASCII(str string) *Regexp {
+	regexp, error := NewRegexpASCII(str, ONIG_OPTION_DEFAULT)
+	if error != nil {
+		panic("regexp: compiling " + str + ": " + error.Error())
+	}
+	return regexp
+}
+
 func (re *Regexp) Free() {
 	mutex.Lock()
 	if re.regex != nil {