fix/search: measure regexp literals in runes (#1149)

The trigram optimizer operates on rune trigrams, but its eligibility check measured UTF-8 bytes. This sent short non-ASCII literals through an inconsistent planning path and could produce surprising composite trees.

Use the parsed rune count so the optimizer and substring planner apply the same threshold.

Author: keegancsmith

index/eval.go

@@ -625,7 +625,7 @@ func (d *indexData) regexpToMatchTreeRecursive(r *syntax.Regexp, minTextSize int
 	switch r.Op {
 	case syntax.OpLiteral:
 		s := string(r.Rune)
-		if len(s) >= minTextSize {
+		if len(r.Rune) >= minTextSize {
 			ignoreCase := syntax.FoldCase == (r.Flags & syntax.FoldCase)
 			mt, err := d.newSubstringMatchTree(&query.Substring{Pattern: s, FileName: fileName, CaseSensitive: !ignoreCase && caseSensitive})
 			return mt, true, !strings.Contains(s, "\n"), err

index/eval_test.go

@@ -90,6 +90,9 @@ func TestRegexpParse(t *testing.T) {
 	cases := []testcase{
 		{"(foo|)bar", substrMT("bar"), false, false},
 		{"(foo|)", &bruteForceMatchTree{}, false, false},
+		{"éa", &bruteForceMatchTree{}, false, false},
+		{"éab", substrMT("éab"), true, false},
+		{"(éa|êb)", &bruteForceMatchTree{}, false, false},
 		{"(foo|bar)baz.*bla", &andMatchTree{[]matchTree{
 			&orMatchTree{[]matchTree{
 				substrMT("foo"),