fix substring matching Unicode haystacks (#34)

CHANGELOG.md

@@ -8,6 +8,7 @@
 
 ## Bugfixes
 
+* avoid incorrect matches when searching for ASCII needles in a Unicode haystack
 * correctly handle Unicode normalization when there are normalizable characters in the pattern, for example characters with umlauts
 * when the needle is composed of a single char, return the score and index
   of the best position instead of always returning the first matched character

matcher/src/exact.rs

@@ -258,6 +258,9 @@ impl Matcher {
                 }
             }
         }
+        if max_score == 0 {
+            return None;
+        }
 
         let score = self.calculate_score::<INDICES, _, _>(
             haystack,

matcher/src/lib.rs

@@ -496,15 +496,8 @@ impl Matcher {
                         .substring_match_1_non_ascii::<INDICES>(haystack, needle, start, indices);
                     return Some(res);
                 }
-                let (start, end) = self.prefilter_non_ascii(haystack, needle_, false)?;
+                let (start, _) = self.prefilter_non_ascii(haystack, needle_, false)?;
-                self.fuzzy_match_optimal::<INDICES, char, char>(
+                self.substring_match_non_ascii::<INDICES, _>(haystack, needle, start, indices)
-                    haystack,
-                    needle,
-                    start,
-                    start + 1,
-                    end,
-                    indices,
-                )
             }
         }
     }

matcher/src/tests.rs

@@ -490,7 +490,13 @@ fn test_unicode() {
                 BONUS_BOUNDARY_WHITE * BONUS_FIRST_CHAR_MULTIPLIER - PENALTY_GAP_START,
             ),
         ],
-    )
+    );
+    assert_not_matches(
+        false,
+        false,
+        false,
+        &[("Flibbertigibbet / イタズラっ子たち", "lying")],
+    );
 }
 
 #[test]