diff --git a/file/error.go b/file/error.go
index 8ff85dfa5..04253b6f6 100644
--- a/file/error.go
+++ b/file/error.go
@@ -3,7 +3,6 @@ package file
 import (
 	"fmt"
 	"strings"
-	"unicode/utf8"
 )
 
 type Error struct {
@@ -19,43 +18,47 @@ func (e *Error) Error() string {
 	return e.format()
 }
 
+var tabReplacer = strings.NewReplacer("\t", " ")
+
 func (e *Error) Bind(source Source) *Error {
+	src := source.String()
+
+	var runeCount, lineStart int
 	e.Line = 1
-	for i, r := range source {
-		if i == e.From {
+	e.Column = 0
+	for i, r := range src {
+		if runeCount == e.From {
 			break
 		}
 		if r == '\n' {
+			lineStart = i
 			e.Line++
 			e.Column = 0
-		} else {
-			e.Column++
 		}
+		runeCount++
+		e.Column++
+	}
+
+	lineEnd := lineStart + strings.IndexByte(src[lineStart:], '\n')
+	if lineEnd < lineStart {
+		lineEnd = len(src)
+	}
+	if lineStart == lineEnd {
+		return e
 	}
-	if snippet, found := source.Snippet(e.Line); found {
-		snippet := strings.Replace(snippet, "\t", " ", -1)
-		srcLine := "\n | " + snippet
-		var bytes = []byte(snippet)
-		var indLine = "\n | "
-		for i := 0; i < e.Column && len(bytes) > 0; i++ {
-			_, sz := utf8.DecodeRune(bytes)
-			bytes = bytes[sz:]
-			if sz > 1 {
-				goto noind
-			} else {
-				indLine += "."
-			}
-		}
-		if _, sz := utf8.DecodeRune(bytes); sz > 1 {
-			goto noind
-		} else {
-			indLine += "^"
-		}
-		srcLine += indLine
 
-	noind:
-		e.Snippet = srcLine
+	const prefix = "\n | "
+	line := src[lineStart:lineEnd]
+	snippet := new(strings.Builder)
+	snippet.Grow(2*len(prefix) + len(line) + e.Column + 1)
+	snippet.WriteString(prefix)
+	tabReplacer.WriteString(snippet, line)
+	snippet.WriteString(prefix)
+	for i := 0; i < e.Column; i++ {
+		snippet.WriteByte('.')
 	}
+	snippet.WriteByte('^')
+	e.Snippet = snippet.String()
 	return e
 }
 
diff --git a/file/source.go b/file/source.go
index 8e2b2d154..b11bb5f9d 100644
--- a/file/source.go
+++ b/file/source.go
@@ -1,48 +1,36 @@
 package file
 
-import (
-	"strings"
-	"unicode/utf8"
-)
+import "strings"
 
-type Source []rune
+type Source struct {
+	raw string
+}
 
 func NewSource(contents string) Source {
-	return []rune(contents)
+	return Source{
+		raw: contents,
+	}
 }
 
 func (s Source) String() string {
-	return string(s)
+	return s.raw
 }
 
 func (s Source) Snippet(line int) (string, bool) {
-	if s == nil {
+	if s.raw == "" {
 		return "", false
 	}
-	lines := strings.Split(string(s), "\n")
-	lineOffsets := make([]int, len(lines))
-	var offset int
-	for i, line := range lines {
-		offset = offset + utf8.RuneCountInString(line) + 1
-		lineOffsets[i] = offset
-	}
-	charStart, found := getLineOffset(lineOffsets, line)
-	if !found || len(s) == 0 {
-		return "", false
+	var start int
+	for i := 1; i < line; i++ {
+		pos := strings.IndexByte(s.raw[start:], '\n')
+		if pos < 0 {
+			return "", false
+		}
+		start += pos + 1
 	}
-	charEnd, found := getLineOffset(lineOffsets, line+1)
-	if found {
-		return string(s[charStart : charEnd-1]), true
-	}
-	return string(s[charStart:]), true
-}
-
-func getLineOffset(lineOffsets []int, line int) (int, bool) {
-	if line == 1 {
-		return 0, true
-	} else if line > 1 && line <= len(lineOffsets) {
-		offset := lineOffsets[line-2]
-		return offset, true
+	end := start + strings.IndexByte(s.raw[start:], '\n')
+	if end < start {
+		end = len(s.raw)
 	}
-	return -1, false
+	return s.raw[start:end], true
 }
diff --git a/internal/ring/ring.go b/internal/ring/ring.go
new file mode 100644
index 000000000..cc9e727b0
--- /dev/null
+++ b/internal/ring/ring.go
@@ -0,0 +1,85 @@
+package ring
+
+// Ring is a very simple ring buffer implementation that uses a slice. The
+// internal slice will only grow, never shrink. When it grows, it grows in
+// chunks of "chunkSize" (given as argument in the [New] function). Pointer and
+// reference types can be safely used because memory is cleared.
+type Ring[T any] struct {
+	data                 []T
+	back, len, chunkSize int
+}
+
+func New[T any](chunkSize int) *Ring[T] {
+	if chunkSize < 1 {
+		panic("chunkSize must be greater than zero")
+	}
+	return &Ring[T]{
+		chunkSize: chunkSize,
+	}
+}
+
+func (r *Ring[T]) Len() int {
+	return r.len
+}
+
+func (r *Ring[T]) Cap() int {
+	return len(r.data)
+}
+
+func (r *Ring[T]) Reset() {
+	var zero T
+	for i := range r.data {
+		r.data[i] = zero // clear mem, optimized by the compiler, in Go 1.21 the "clear" builtin can be used
+	}
+	r.back = 0
+	r.len = 0
+}
+
+// Nth returns the n-th oldest value (zero-based) in the ring without making
+// any change.
+func (r *Ring[T]) Nth(n int) (v T, ok bool) {
+	if n < 0 || n >= r.len || len(r.data) == 0 {
+		return v, false
+	}
+	n = (n + r.back) % len(r.data)
+	return r.data[n], true
+}
+
+// Dequeue returns the oldest value.
+func (r *Ring[T]) Dequeue() (v T, ok bool) {
+	if r.len == 0 {
+		return v, false
+	}
+	v, r.data[r.back] = r.data[r.back], v // retrieve and clear mem
+	r.len--
+	r.back = (r.back + 1) % len(r.data)
+	return v, true
+}
+
+// Enqueue adds an item to the ring.
+func (r *Ring[T]) Enqueue(v T) {
+	if r.len == len(r.data) {
+		r.grow()
+	}
+	writePos := (r.back + r.len) % len(r.data)
+	r.data[writePos] = v
+	r.len++
+}
+
+func (r *Ring[T]) grow() {
+	s := make([]T, len(r.data)+r.chunkSize)
+	if r.len > 0 {
+		chunk1 := r.back + r.len
+		if chunk1 > len(r.data) {
+			chunk1 = len(r.data)
+		}
+		copied := copy(s, r.data[r.back:chunk1])
+
+		if copied < r.len { // wrapped slice
+			chunk2 := r.len - copied
+			copy(s[copied:], r.data[:chunk2])
+		}
+	}
+	r.back = 0
+	r.data = s
+}
diff --git a/internal/ring/ring_test.go b/internal/ring/ring_test.go
new file mode 100644
index 000000000..b7457cd70
--- /dev/null
+++ b/internal/ring/ring_test.go
@@ -0,0 +1,140 @@
+package ring
+
+import (
+	"fmt"
+	"testing"
+)
+
+func TestRing(t *testing.T) {
+	type op = ringOp[int]
+	testRing(t, New[int](3),
+		// noops on empty ring
+		op{cap: 0, opType: opRst, value: 0, items: []int{}},
+		op{cap: 0, opType: opDeq, value: 0, items: []int{}},
+
+		// basic
+		op{cap: 3, opType: opEnq, value: 1, items: []int{1}},
+		op{cap: 3, opType: opDeq, value: 1, items: []int{}},
+
+		// wrapping
+		op{cap: 3, opType: opEnq, value: 2, items: []int{2}},
+		op{cap: 3, opType: opEnq, value: 3, items: []int{2, 3}},
+		op{cap: 3, opType: opEnq, value: 4, items: []int{2, 3, 4}},
+		op{cap: 3, opType: opDeq, value: 2, items: []int{3, 4}},
+		op{cap: 3, opType: opDeq, value: 3, items: []int{4}},
+		op{cap: 3, opType: opDeq, value: 4, items: []int{}},
+
+		// resetting
+		op{cap: 3, opType: opEnq, value: 2, items: []int{2}},
+		op{cap: 3, opType: opRst, value: 0, items: []int{}},
+		op{cap: 3, opType: opDeq, value: 0, items: []int{}},
+
+		// growing without wrapping
+		op{cap: 3, opType: opEnq, value: 5, items: []int{5}},
+		op{cap: 3, opType: opEnq, value: 6, items: []int{5, 6}},
+		op{cap: 3, opType: opEnq, value: 7, items: []int{5, 6, 7}},
+		op{cap: 6, opType: opEnq, value: 8, items: []int{5, 6, 7, 8}},
+		op{cap: 6, opType: opRst, value: 0, items: []int{}},
+		op{cap: 6, opType: opDeq, value: 0, items: []int{}},
+
+		// growing and wrapping
+		op{cap: 6, opType: opEnq, value: 9, items: []int{9}},
+		op{cap: 6, opType: opEnq, value: 10, items: []int{9, 10}},
+		op{cap: 6, opType: opEnq, value: 11, items: []int{9, 10, 11}},
+		op{cap: 6, opType: opEnq, value: 12, items: []int{9, 10, 11, 12}},
+		op{cap: 6, opType: opEnq, value: 13, items: []int{9, 10, 11, 12, 13}},
+		op{cap: 6, opType: opEnq, value: 14, items: []int{9, 10, 11, 12, 13, 14}},
+		op{cap: 6, opType: opDeq, value: 9, items: []int{10, 11, 12, 13, 14}},
+		op{cap: 6, opType: opDeq, value: 10, items: []int{11, 12, 13, 14}},
+		op{cap: 6, opType: opEnq, value: 15, items: []int{11, 12, 13, 14, 15}},
+		op{cap: 6, opType: opEnq, value: 16, items: []int{11, 12, 13, 14, 15, 16}},
+		op{cap: 9, opType: opEnq, value: 17, items: []int{11, 12, 13, 14, 15, 16, 17}}, // grows wrapped
+		op{cap: 9, opType: opDeq, value: 11, items: []int{12, 13, 14, 15, 16, 17}},
+		op{cap: 9, opType: opDeq, value: 12, items: []int{13, 14, 15, 16, 17}},
+		op{cap: 9, opType: opDeq, value: 13, items: []int{14, 15, 16, 17}},
+		op{cap: 9, opType: opDeq, value: 14, items: []int{15, 16, 17}},
+		op{cap: 9, opType: opDeq, value: 15, items: []int{16, 17}},
+		op{cap: 9, opType: opDeq, value: 16, items: []int{17}},
+		op{cap: 9, opType: opDeq, value: 17, items: []int{}},
+		op{cap: 9, opType: opDeq, value: 0, items: []int{}},
+	)
+
+	t.Run("should panic on invalid chunkSize", func(t *testing.T) {
+		defer func() {
+			if r := recover(); r == nil {
+				t.Fatalf("should have panicked")
+			}
+		}()
+		New[int](0)
+	})
+}
+
+const (
+	opEnq = iota // enqueue an item
+	opDeq        // dequeue an item and an item was available
+	opRst        // reset
+)
+
+type ringOp[T comparable] struct {
+	cap    int // expected values
+	opType int // opEnq or opDeq
+	value  T   // value to enqueue or value expected for dequeue; ignored for opRst
+	items  []T // items left
+}
+
+func testRing[T comparable](t *testing.T, r *Ring[T], ops ...ringOp[T]) {
+	for i, op := range ops {
+		testOK := t.Run(fmt.Sprintf("opIndex=%v", i), func(t *testing.T) {
+			testRingOp(t, r, op)
+		})
+		if !testOK {
+			return
+		}
+	}
+}
+
+func testRingOp[T comparable](t *testing.T, r *Ring[T], op ringOp[T]) {
+	var zero T
+	switch op.opType {
+	case opEnq:
+		r.Enqueue(op.value)
+	case opDeq:
+		shouldSucceed := r.Len() > 0
+		v, ok := r.Dequeue()
+		switch {
+		case ok != shouldSucceed:
+			t.Fatalf("should have succeeded: %v", shouldSucceed)
+		case ok && v != op.value:
+			t.Fatalf("expected value: %v; got: %v", op.value, v)
+		case !ok && v != zero:
+			t.Fatalf("expected zero value; got: %v", v)
+		}
+	case opRst:
+		r.Reset()
+	}
+	if c := r.Cap(); c != op.cap {
+		t.Fatalf("expected cap: %v; got: %v", op.cap, c)
+	}
+	if l := r.Len(); l != len(op.items) {
+		t.Errorf("expected Len(): %v; got: %v", len(op.items), l)
+	}
+	var got []T
+	for i := 0; ; i++ {
+		v, ok := r.Nth(i)
+		if !ok {
+			break
+		}
+		got = append(got, v)
+	}
+	if l := len(got); l != len(op.items) {
+		t.Errorf("expected items: %v\ngot items: %v", op.items, got)
+	}
+	for i := range op.items {
+		if op.items[i] != got[i] {
+			t.Fatalf("expected items: %v\ngot items: %v", op.items, got)
+		}
+	}
+	if v, ok := r.Nth(len(op.items)); ok || v != zero {
+		t.Fatalf("expected no more items, got: v=%v; ok=%v", v, ok)
+	}
+}
diff --git a/parser/bench_test.go b/parser/bench_test.go
new file mode 100644
index 000000000..0b9a69276
--- /dev/null
+++ b/parser/bench_test.go
@@ -0,0 +1,20 @@
+package parser
+
+import "testing"
+
+func BenchmarkParser(b *testing.B) {
+	const source = `
+		/*
+			Showing worst case scenario
+		*/
+		let value = trim("contains escapes \n\"\\ \U0001F600 and non ASCII ñ"); // inline comment
+		len(value) == 0x2A
+		// let's introduce an error too
+		whatever
+	`
+	b.ReportAllocs()
+	p := new(Parser)
+	for i := 0; i < b.N; i++ {
+		p.Parse(source, nil)
+	}
+}
diff --git a/parser/lexer/lexer.go b/parser/lexer/lexer.go
index e6b06c09d..f417f02b6 100644
--- a/parser/lexer/lexer.go
+++ b/parser/lexer/lexer.go
@@ -2,152 +2,178 @@ package lexer
 
 import (
 	"fmt"
+	"io"
 	"strings"
+	"unicode/utf8"
 
 	"github.com/expr-lang/expr/file"
+	"github.com/expr-lang/expr/internal/ring"
 )
 
+const ringChunkSize = 10
+
+// Lex will buffer and return the tokens of a disposable *[Lexer].
 func Lex(source file.Source) ([]Token, error) {
-	l := &lexer{
-		source: source,
-		tokens: make([]Token, 0),
-		start:  0,
-		end:    0,
+	tokens := make([]Token, 0, ringChunkSize)
+	l := New()
+	l.Reset(source)
+	for {
+		t, err := l.Next()
+		switch err {
+		case nil:
+			tokens = append(tokens, t)
+		case io.EOF:
+			return tokens, nil
+		default:
+			return nil, err
+		}
 	}
-	l.commit()
+}
 
-	for state := root; state != nil; {
-		state = state(l)
+// New returns a reusable lexer.
+func New() *Lexer {
+	return &Lexer{
+		tokens: ring.New[Token](ringChunkSize),
 	}
+}
 
-	if l.err != nil {
-		return nil, l.err.Bind(source)
+type Lexer struct {
+	state      stateFn
+	source     file.Source
+	tokens     *ring.Ring[Token]
+	err        *file.Error
+	start, end struct {
+		byte, rune int
 	}
+	eof bool
+}
 
-	return l.tokens, nil
+func (l *Lexer) Reset(source file.Source) {
+	l.source = source
+	l.tokens.Reset()
+	l.state = root
 }
 
-type lexer struct {
-	source     file.Source
-	tokens     []Token
-	start, end int
-	err        *file.Error
+func (l *Lexer) Next() (Token, error) {
+	for l.state != nil && l.err == nil && l.tokens.Len() == 0 {
+		l.state = l.state(l)
+	}
+	if l.err != nil {
+		return Token{}, l.err.Bind(l.source)
+	}
+	if t, ok := l.tokens.Dequeue(); ok {
+		return t, nil
+	}
+	return Token{}, io.EOF
 }
 
 const eof rune = -1
 
-func (l *lexer) commit() {
+func (l *Lexer) commit() {
 	l.start = l.end
 }
 
-func (l *lexer) next() rune {
-	if l.end >= len(l.source) {
-		l.end++
+func (l *Lexer) next() rune {
+	if l.end.byte >= len(l.source.String()) {
+		l.eof = true
 		return eof
 	}
-	r := l.source[l.end]
-	l.end++
+	r, sz := utf8.DecodeRuneInString(l.source.String()[l.end.byte:])
+	l.end.rune++
+	l.end.byte += sz
 	return r
 }
 
-func (l *lexer) peek() rune {
-	r := l.next()
-	l.backup()
-	return r
+func (l *Lexer) peek() rune {
+	if l.end.byte < len(l.source.String()) {
+		r, _ := utf8.DecodeRuneInString(l.source.String()[l.end.byte:])
+		return r
+	}
+	return eof
 }
 
-func (l *lexer) backup() {
-	l.end--
+func (l *Lexer) peekByte() (byte, bool) {
+	if l.end.byte >= 0 && l.end.byte < len(l.source.String()) {
+		return l.source.String()[l.end.byte], true
+	}
+	return 0, false
 }
 
-func (l *lexer) emit(t Kind) {
+func (l *Lexer) backup() {
+	if l.eof {
+		l.eof = false
+	} else if l.end.rune > 0 {
+		_, sz := utf8.DecodeLastRuneInString(l.source.String()[:l.end.byte])
+		l.end.byte -= sz
+		l.end.rune--
+	}
+}
+
+func (l *Lexer) emit(t Kind) {
 	l.emitValue(t, l.word())
 }
 
-func (l *lexer) emitValue(t Kind, value string) {
-	l.tokens = append(l.tokens, Token{
-		Location: file.Location{From: l.start, To: l.end},
+func (l *Lexer) emitValue(t Kind, value string) {
+	l.tokens.Enqueue(Token{
+		Location: file.Location{From: l.start.rune, To: l.end.rune},
 		Kind:     t,
 		Value:    value,
 	})
 	l.commit()
 }
 
-func (l *lexer) emitEOF() {
-	from := l.end - 2
+func (l *Lexer) emitEOF() {
+	from := l.end.rune - 1
 	if from < 0 {
 		from = 0
 	}
-	to := l.end - 1
+	to := l.end.rune - 0
 	if to < 0 {
 		to = 0
 	}
-	l.tokens = append(l.tokens, Token{
+	l.tokens.Enqueue(Token{
 		Location: file.Location{From: from, To: to},
 		Kind:     EOF,
 	})
 	l.commit()
 }
 
-func (l *lexer) skip() {
+func (l *Lexer) skip() {
 	l.commit()
 }
 
-func (l *lexer) word() string {
-	// TODO: boundary check is NOT needed here, but for some reason CI fuzz tests are failing.
-	if l.start > len(l.source) || l.end > len(l.source) {
-		return "__invalid__"
-	}
-	return string(l.source[l.start:l.end])
+func (l *Lexer) word() string {
+	return l.source.String()[l.start.byte:l.end.byte]
 }
 
-func (l *lexer) accept(valid string) bool {
-	if strings.ContainsRune(valid, l.next()) {
+func (l *Lexer) accept(valid string) bool {
+	if strings.ContainsRune(valid, l.peek()) {
+		l.next()
 		return true
 	}
-	l.backup()
 	return false
 }
 
-func (l *lexer) acceptRun(valid string) {
-	for strings.ContainsRune(valid, l.next()) {
+func (l *Lexer) acceptRun(valid string) {
+	for l.accept(valid) {
 	}
-	l.backup()
 }
 
-func (l *lexer) skipSpaces() {
-	r := l.peek()
-	for ; r == ' '; r = l.peek() {
-		l.next()
-	}
+func (l *Lexer) skipSpaces() {
+	l.acceptRun(" ")
 	l.skip()
 }
 
-func (l *lexer) acceptWord(word string) bool {
-	pos := l.end
-
-	l.skipSpaces()
-
-	for _, ch := range word {
-		if l.next() != ch {
-			l.end = pos
-			return false
-		}
-	}
-	if r := l.peek(); r != ' ' && r != eof {
-		l.end = pos
-		return false
-	}
-
-	return true
-}
-
-func (l *lexer) error(format string, args ...any) stateFn {
+func (l *Lexer) error(format string, args ...any) stateFn {
 	if l.err == nil { // show first error
+		end := l.end.rune
+		if l.eof {
+			end++
+		}
 		l.err = &file.Error{
 			Location: file.Location{
-				From: l.end - 1,
-				To:   l.end,
+				From: end - 1,
+				To:   end,
 			},
 			Message: fmt.Sprintf(format, args...),
 		}
@@ -167,7 +193,7 @@ func digitVal(ch rune) int {
 
 func lower(ch rune) rune { return ('a' - 'A') | ch } // returns lower-case ch iff ch is ASCII letter
 
-func (l *lexer) scanDigits(ch rune, base, n int) rune {
+func (l *Lexer) scanDigits(ch rune, base, n int) rune {
 	for n > 0 && digitVal(ch) < base {
 		ch = l.next()
 		n--
@@ -178,7 +204,7 @@ func (l *lexer) scanDigits(ch rune, base, n int) rune {
 	return ch
 }
 
-func (l *lexer) scanEscape(quote rune) rune {
+func (l *Lexer) scanEscape(quote rune) rune {
 	ch := l.next() // read character after '/'
 	switch ch {
 	case 'a', 'b', 'f', 'n', 'r', 't', 'v', '\\', quote:
@@ -198,7 +224,7 @@ func (l *lexer) scanEscape(quote rune) rune {
 	return ch
 }
 
-func (l *lexer) scanString(quote rune) (n int) {
+func (l *Lexer) scanString(quote rune) (n int) {
 	ch := l.next() // read character after quote
 	for ch != quote {
 		if ch == '\n' || ch == eof {
@@ -215,7 +241,7 @@ func (l *lexer) scanString(quote rune) (n int) {
 	return
 }
 
-func (l *lexer) scanRawString(quote rune) (n int) {
+func (l *Lexer) scanRawString(quote rune) (n int) {
 	ch := l.next() // read character after back tick
 	for ch != quote {
 		if ch == eof {
@@ -225,6 +251,6 @@ func (l *lexer) scanRawString(quote rune) (n int) {
 		ch = l.next()
 		n++
 	}
-	l.emitValue(String, string(l.source[l.start+1:l.end-1]))
+	l.emitValue(String, l.source.String()[l.start.byte+1:l.end.byte-1])
 	return
 }
diff --git a/parser/lexer/lexer_test.go b/parser/lexer/lexer_test.go
index db02d2acf..5171f4255 100644
--- a/parser/lexer/lexer_test.go
+++ b/parser/lexer/lexer_test.go
@@ -335,6 +335,7 @@ literal not terminated (1:10)
 früh ♥︎
 unrecognized character: U+2665 '♥' (1:6)
  | früh ♥︎
+ | .....^
 `
 
 func TestLex_error(t *testing.T) {
diff --git a/parser/lexer/state.go b/parser/lexer/state.go
index c694a2ca0..e5ad45bcd 100644
--- a/parser/lexer/state.go
+++ b/parser/lexer/state.go
@@ -6,9 +6,9 @@ import (
 	"github.com/expr-lang/expr/parser/utils"
 )
 
-type stateFn func(*lexer) stateFn
+type stateFn func(*Lexer) stateFn
 
-func root(l *lexer) stateFn {
+func root(l *Lexer) stateFn {
 	switch r := l.next(); {
 	case r == eof:
 		l.emitEOF()
@@ -61,7 +61,7 @@ func root(l *lexer) stateFn {
 	return root
 }
 
-func number(l *lexer) stateFn {
+func number(l *Lexer) stateFn {
 	if !l.scanNumber() {
 		return l.error("bad number syntax: %q", l.word())
 	}
@@ -69,7 +69,7 @@ func number(l *lexer) stateFn {
 	return root
 }
 
-func (l *lexer) scanNumber() bool {
+func (l *Lexer) scanNumber() bool {
 	digits := "0123456789_"
 	// Is it hex?
 	if l.accept("0") {
@@ -107,7 +107,7 @@ func (l *lexer) scanNumber() bool {
 	return true
 }
 
-func dot(l *lexer) stateFn {
+func dot(l *Lexer) stateFn {
 	l.next()
 	if l.accept("0123456789") {
 		l.backup()
@@ -118,7 +118,7 @@ func dot(l *lexer) stateFn {
 	return root
 }
 
-func identifier(l *lexer) stateFn {
+func identifier(l *Lexer) stateFn {
 loop:
 	for {
 		switch r := l.next(); {
@@ -140,7 +140,7 @@ loop:
 	return root
 }
 
-func not(l *lexer) stateFn {
+func not(l *Lexer) stateFn {
 	l.emit(Operator)
 
 	l.skipSpaces()
@@ -167,13 +167,13 @@ func not(l *lexer) stateFn {
 	return root
 }
 
-func questionMark(l *lexer) stateFn {
+func questionMark(l *Lexer) stateFn {
 	l.accept(".?")
 	l.emit(Operator)
 	return root
 }
 
-func slash(l *lexer) stateFn {
+func slash(l *Lexer) stateFn {
 	if l.accept("/") {
 		return singleLineComment
 	}
@@ -184,7 +184,7 @@ func slash(l *lexer) stateFn {
 	return root
 }
 
-func singleLineComment(l *lexer) stateFn {
+func singleLineComment(l *Lexer) stateFn {
 	for {
 		r := l.next()
 		if r == eof || r == '\n' {
@@ -195,7 +195,7 @@ func singleLineComment(l *lexer) stateFn {
 	return root
 }
 
-func multiLineComment(l *lexer) stateFn {
+func multiLineComment(l *Lexer) stateFn {
 	for {
 		r := l.next()
 		if r == eof {
@@ -209,7 +209,7 @@ func multiLineComment(l *lexer) stateFn {
 	return root
 }
 
-func pointer(l *lexer) stateFn {
+func pointer(l *Lexer) stateFn {
 	l.accept("#")
 	l.emit(Operator)
 	for {
diff --git a/parser/lexer/token.go b/parser/lexer/token.go
index 459fa6905..c809c690e 100644
--- a/parser/lexer/token.go
+++ b/parser/lexer/token.go
@@ -31,17 +31,13 @@ func (t Token) String() string {
 }
 
 func (t Token) Is(kind Kind, values ...string) bool {
-	if len(values) == 0 {
-		return kind == t.Kind
+	if kind != t.Kind {
+		return false
 	}
-
 	for _, v := range values {
 		if v == t.Value {
-			goto found
+			return true
 		}
 	}
-	return false
-
-found:
-	return kind == t.Kind
+	return len(values) == 0
 }
diff --git a/parser/lexer/utils.go b/parser/lexer/utils.go
index 5c9e6b59d..fdb8beaa1 100644
--- a/parser/lexer/utils.go
+++ b/parser/lexer/utils.go
@@ -36,7 +36,8 @@ func unescape(value string) (string, error) {
 	if size >= math.MaxInt {
 		return "", fmt.Errorf("too large string")
 	}
-	buf := make([]byte, 0, size)
+	buf := new(strings.Builder)
+	buf.Grow(int(size))
 	for len(value) > 0 {
 		c, multibyte, rest, err := unescapeChar(value)
 		if err != nil {
@@ -44,13 +45,13 @@ func unescape(value string) (string, error) {
 		}
 		value = rest
 		if c < utf8.RuneSelf || !multibyte {
-			buf = append(buf, byte(c))
+			buf.WriteByte(byte(c))
 		} else {
 			n := utf8.EncodeRune(runeTmp[:], c)
-			buf = append(buf, runeTmp[:n]...)
+			buf.Write(runeTmp[:n])
 		}
 	}
-	return string(buf), nil
+	return buf.String(), nil
 }
 
 // unescapeChar takes a string input and returns the following info:
diff --git a/parser/parser.go b/parser/parser.go
index 0a463fed5..e1dd111fc 100644
--- a/parser/parser.go
+++ b/parser/parser.go
@@ -1,7 +1,9 @@
 package parser
 
 import (
+	"errors"
 	"fmt"
+	"io"
 	"math"
 	"strconv"
 	"strings"
@@ -10,6 +12,7 @@ import (
 	"github.com/expr-lang/expr/builtin"
 	"github.com/expr-lang/expr/conf"
 	"github.com/expr-lang/expr/file"
+	"github.com/expr-lang/expr/parser/lexer"
 	. "github.com/expr-lang/expr/parser/lexer"
 	"github.com/expr-lang/expr/parser/operator"
 	"github.com/expr-lang/expr/parser/utils"
@@ -44,17 +47,50 @@ var predicates = map[string]struct {
 	"reduce":        {[]arg{expr, predicate, expr | optional}},
 }
 
-type parser struct {
-	tokens    []Token
-	current   Token
-	pos       int
-	err       *file.Error
-	config    *conf.Config
-	depth     int  // predicate call depth
-	nodeCount uint // tracks number of AST nodes created
+// Parser is a reusable parser. The zero value is ready for use.
+type Parser struct {
+	lexer            *lexer.Lexer
+	current, stashed Token
+	hasStash         bool
+	err              *file.Error
+	config           *conf.Config
+	depth            int  // predicate call depth
+	nodeCount        uint // tracks number of AST nodes created
 }
 
-func (p *parser) checkNodeLimit() error {
+func (p *Parser) Parse(input string, config *conf.Config) (*Tree, error) {
+	if p.lexer == nil {
+		p.lexer = lexer.New()
+	}
+	p.config = config
+	source := file.NewSource(input)
+	p.lexer.Reset(source)
+	p.next()
+	node := p.parseSequenceExpression()
+
+	if !p.current.Is(EOF) {
+		p.error("unexpected token %v", p.current)
+	}
+
+	tree := &Tree{
+		Node:   node,
+		Source: source,
+	}
+	err := p.err
+
+	// cleanup non-reusable pointer values and reset state
+	p.err = nil
+	p.config = nil
+	p.lexer.Reset(file.Source{})
+
+	if err != nil {
+		return tree, err.Bind(source)
+	}
+
+	return tree, nil
+}
+
+func (p *Parser) checkNodeLimit() error {
 	p.nodeCount++
 	if p.config == nil {
 		if p.nodeCount > conf.DefaultMaxNodes {
@@ -70,7 +106,7 @@ func (p *parser) checkNodeLimit() error {
 	return nil
 }
 
-func (p *parser) createNode(n Node, loc file.Location) Node {
+func (p *Parser) createNode(n Node, loc file.Location) Node {
 	if err := p.checkNodeLimit(); err != nil {
 		return nil
 	}
@@ -81,7 +117,7 @@ func (p *parser) createNode(n Node, loc file.Location) Node {
 	return n
 }
 
-func (p *parser) createMemberNode(n *MemberNode, loc file.Location) *MemberNode {
+func (p *Parser) createMemberNode(n *MemberNode, loc file.Location) *MemberNode {
 	if err := p.checkNodeLimit(); err != nil {
 		return nil
 	}
@@ -102,42 +138,14 @@ func Parse(input string) (*Tree, error) {
 }
 
 func ParseWithConfig(input string, config *conf.Config) (*Tree, error) {
-	source := file.NewSource(input)
-
-	tokens, err := Lex(source)
-	if err != nil {
-		return nil, err
-	}
-
-	p := &parser{
-		tokens:  tokens,
-		current: tokens[0],
-		config:  config,
-	}
-
-	node := p.parseSequenceExpression()
-
-	if !p.current.Is(EOF) {
-		p.error("unexpected token %v", p.current)
-	}
-
-	tree := &Tree{
-		Node:   node,
-		Source: source,
-	}
-
-	if p.err != nil {
-		return tree, p.err.Bind(source)
-	}
-
-	return tree, nil
+	return new(Parser).Parse(input, config)
 }
 
-func (p *parser) error(format string, args ...any) {
+func (p *Parser) error(format string, args ...any) {
 	p.errorAt(p.current, format, args...)
 }
 
-func (p *parser) errorAt(token Token, format string, args ...any) {
+func (p *Parser) errorAt(token Token, format string, args ...any) {
 	if p.err == nil { // show first error
 		p.err = &file.Error{
 			Location: token.Location,
@@ -146,16 +154,32 @@ func (p *parser) errorAt(token Token, format string, args ...any) {
 	}
 }
 
-func (p *parser) next() {
-	p.pos++
-	if p.pos >= len(p.tokens) {
-		p.error("unexpected end of expression")
+func (p *Parser) next() {
+	if p.hasStash {
+		p.current = p.stashed
+		p.hasStash = false
 		return
 	}
-	p.current = p.tokens[p.pos]
+
+	token, err := p.lexer.Next()
+	var e *file.Error
+	switch {
+	case err == nil:
+		p.current = token
+	case errors.Is(err, io.EOF):
+		p.error("unexpected end of expression")
+	case errors.As(err, &e):
+		p.err = e
+	default:
+		p.err = &file.Error{
+			Location: p.current.Location,
+			Message:  "unknown lexing error",
+			Prev:     err,
+		}
+	}
 }
 
-func (p *parser) expect(kind Kind, values ...string) {
+func (p *Parser) expect(kind Kind, values ...string) {
 	if p.current.Is(kind, values...) {
 		p.next()
 		return
@@ -165,7 +189,7 @@ func (p *parser) expect(kind Kind, values ...string) {
 
 // parse functions
 
-func (p *parser) parseSequenceExpression() Node {
+func (p *Parser) parseSequenceExpression() Node {
 	nodes := []Node{p.parseExpression(0)}
 
 	for p.current.Is(Operator, ";") && p.err == nil {
@@ -186,7 +210,7 @@ func (p *parser) parseSequenceExpression() Node {
 	}, nodes[0].Location())
 }
 
-func (p *parser) parseExpression(precedence int) Node {
+func (p *Parser) parseExpression(precedence int) Node {
 	if p.err != nil {
 		return nil
 	}
@@ -209,15 +233,16 @@ func (p *parser) parseExpression(precedence int) Node {
 
 		// Handle "not *" operator, like "not in" or "not contains".
 		if negate {
-			currentPos := p.pos
+			tokenBackup := p.current
 			p.next()
 			if operator.AllowedNegateSuffix(p.current.Value) {
 				if op, ok := operator.Binary[p.current.Value]; ok && op.Precedence >= precedence {
 					notToken = p.current
 					opToken = p.current
 				} else {
-					p.pos = currentPos
-					p.current = opToken
+					p.hasStash = true
+					p.stashed = p.current
+					p.current = tokenBackup
 					break
 				}
 			} else {
@@ -288,7 +313,7 @@ func (p *parser) parseExpression(precedence int) Node {
 	return nodeLeft
 }
 
-func (p *parser) parseVariableDeclaration() Node {
+func (p *Parser) parseVariableDeclaration() Node {
 	p.expect(Operator, "let")
 	variableName := p.current
 	p.expect(Identifier)
@@ -303,7 +328,7 @@ func (p *parser) parseVariableDeclaration() Node {
 	}, variableName.Location)
 }
 
-func (p *parser) parseConditionalIf() Node {
+func (p *Parser) parseConditionalIf() Node {
 	p.next()
 	nodeCondition := p.parseExpression(0)
 	p.expect(Bracket, "{")
@@ -322,7 +347,7 @@ func (p *parser) parseConditionalIf() Node {
 
 }
 
-func (p *parser) parseConditional(node Node) Node {
+func (p *Parser) parseConditional(node Node) Node {
 	var expr1, expr2 Node
 	for p.current.Is(Operator, "?") && p.err == nil {
 		p.next()
@@ -349,7 +374,7 @@ func (p *parser) parseConditional(node Node) Node {
 	return node
 }
 
-func (p *parser) parsePrimary() Node {
+func (p *Parser) parsePrimary() Node {
 	token := p.current
 
 	if token.Is(Operator) {
@@ -402,7 +427,7 @@ func (p *parser) parsePrimary() Node {
 	return p.parseSecondary()
 }
 
-func (p *parser) parseSecondary() Node {
+func (p *Parser) parseSecondary() Node {
 	var node Node
 	token := p.current
 
@@ -501,7 +526,7 @@ func (p *parser) parseSecondary() Node {
 	return p.parsePostfixExpression(node)
 }
 
-func (p *parser) toIntegerNode(number int64) Node {
+func (p *Parser) toIntegerNode(number int64) Node {
 	if number > math.MaxInt {
 		p.error("integer literal is too large")
 		return nil
@@ -509,7 +534,7 @@ func (p *parser) toIntegerNode(number int64) Node {
 	return p.createNode(&IntegerNode{Value: int(number)}, p.current.Location)
 }
 
-func (p *parser) toFloatNode(number float64) Node {
+func (p *Parser) toFloatNode(number float64) Node {
 	if number > math.MaxFloat64 {
 		p.error("float literal is too large")
 		return nil
@@ -517,7 +542,7 @@ func (p *parser) toFloatNode(number float64) Node {
 	return p.createNode(&FloatNode{Value: number}, p.current.Location)
 }
 
-func (p *parser) parseCall(token Token, arguments []Node, checkOverrides bool) Node {
+func (p *Parser) parseCall(token Token, arguments []Node, checkOverrides bool) Node {
 	var node Node
 
 	isOverridden := false
@@ -595,7 +620,7 @@ func (p *parser) parseCall(token Token, arguments []Node, checkOverrides bool) N
 	return node
 }
 
-func (p *parser) parseArguments(arguments []Node) []Node {
+func (p *Parser) parseArguments(arguments []Node) []Node {
 	// If pipe operator is used, the first argument is the left-hand side
 	// of the operator, so we do not parse it as an argument inside brackets.
 	offset := len(arguments)
@@ -616,7 +641,7 @@ func (p *parser) parseArguments(arguments []Node) []Node {
 	return arguments
 }
 
-func (p *parser) parsePredicate() Node {
+func (p *Parser) parsePredicate() Node {
 	startToken := p.current
 	withBrackets := false
 	if p.current.Is(Bracket, "{") {
@@ -648,7 +673,7 @@ func (p *parser) parsePredicate() Node {
 	return predicateNode
 }
 
-func (p *parser) parseArrayExpression(token Token) Node {
+func (p *Parser) parseArrayExpression(token Token) Node {
 	nodes := make([]Node, 0)
 
 	p.expect(Bracket, "[")
@@ -672,7 +697,7 @@ end:
 	return node
 }
 
-func (p *parser) parseMapExpression(token Token) Node {
+func (p *Parser) parseMapExpression(token Token) Node {
 	p.expect(Bracket, "{")
 
 	nodes := make([]Node, 0)
@@ -725,7 +750,7 @@ end:
 	return node
 }
 
-func (p *parser) parsePostfixExpression(node Node) Node {
+func (p *Parser) parsePostfixExpression(node Node) Node {
 	postfixToken := p.current
 	for (postfixToken.Is(Operator) || postfixToken.Is(Bracket)) && p.err == nil {
 		optional := postfixToken.Value == "?."
@@ -855,7 +880,7 @@ func (p *parser) parsePostfixExpression(node Node) Node {
 	}
 	return node
 }
-func (p *parser) parseComparison(left Node, token Token, precedence int) Node {
+func (p *Parser) parseComparison(left Node, token Token, precedence int) Node {
 	var rootNode Node
 	for {
 		comparator := p.parseExpression(precedence + 1)
diff --git a/vm/vm_test.go b/vm/vm_test.go
index 91752a419..817fc6cc2 100644
--- a/vm/vm_test.go
+++ b/vm/vm_test.go
@@ -8,6 +8,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/expr-lang/expr/file"
 	"github.com/expr-lang/expr/internal/testify/require"
 
 	"github.com/expr-lang/expr"
@@ -609,10 +610,10 @@ func TestVM_DirectCallOpcodes(t *testing.T) {
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			program := vm.NewProgram(
-				nil, // source
-				nil, // node
-				nil, // locations
-				0,   // variables
+				file.Source{}, // source
+				nil,           // node
+				nil,           // locations
+				0,             // variables
 				tt.consts,
 				tt.bytecode,
 				tt.args,
@@ -735,10 +736,10 @@ func TestVM_IndexAndCountOperations(t *testing.T) {
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			program := vm.NewProgram(
-				nil, // source
-				nil, // node
-				nil, // locations
-				0,   // variables
+				file.Source{}, // source
+				nil,           // node
+				nil,           // locations
+				0,             // variables
 				tt.consts,
 				tt.bytecode,
 				tt.args,
@@ -1176,10 +1177,10 @@ func TestVM_DirectBasicOpcodes(t *testing.T) {
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			program := vm.NewProgram(
-				nil, // source
-				nil, // node
-				nil, // locations
-				0,   // variables
+				file.Source{}, // source
+				nil,           // node
+				nil,           // locations
+				0,             // variables
 				tt.consts,
 				tt.bytecode,
 				tt.args,