perf: fix IndexAt() to extract runes from segment directly

The previous implementation tracked byte lengths but compared against a
rune index, then called s.Runes()[idx] which reconverted the entire
string to runes - O(n) per index lookup.

Now we track cumulative rune lengths and extract the character directly
from the segment, making IndexAt() O(n) for the first call and avoiding
the redundant full-string conversion.

Author: amterp

core/testing/string_test.go

@@ -33,3 +33,16 @@ print(a[len(a) - 3])
 	assertOnlyOutput(t, stdOutBuffer, "i\n")
 	assertNoErrors(t)
 }
+
+func TestString_MultiByteIndexing(t *testing.T) {
+	script := `
+s = "a😀b"
+print(s[0])
+print(s[1])
+print(s[2])
+print(s[-1])
+`
+	setupAndRunCode(t, script, "--color=never")
+	assertOnlyOutput(t, stdOutBuffer, "a\n😀\nb\nb\n")
+	assertNoErrors(t)
+}

core/type_string.go

@@ -120,14 +120,16 @@ func (s *RadString) Index(i *Interpreter, idxNode *ts.Node) RadString {
 
 // assumes idx is valid for this string
 func (s *RadString) IndexAt(idx int64) RadString {
-	cumLen := 0
+	cumRuneLen := int64(0)
 	for _, segment := range s.Segments {
-		nextSegmentLen := len(segment.String)
-		if cumLen+nextSegmentLen > int(idx) {
-			char := s.Runes()[idx] // todo inefficient, should just look up in segment
+		segmentRunes := []rune(segment.String)
+		segmentRuneLen := int64(len(segmentRunes))
+		if cumRuneLen+segmentRuneLen > idx {
+			offsetInSegment := idx - cumRuneLen
+			char := segmentRunes[offsetInSegment]
 			return newRadStringWithAttr(string(char), segment)
 		}
-		cumLen += +nextSegmentLen
+		cumRuneLen += segmentRuneLen
 	}
 	RP.RadErrorExit("Bug! IndexAt called with invalid index")
 	panic(UNREACHABLE)