LibJS: Keep more string construction in UTF-16

Use a typed RegExp cache key instead of serializing the UTF-16
pattern and flags into a String. This avoids another StringBuilder
path for data that is already naturally represented as UTF-16 plus
flag bits.

Also make legacy unescape operate on UTF-16 code units and build its
result with Utf16StringBuilder. This preserves direct non-ASCII input
and %uXXXX escapes, including lone surrogate code units, without
routing the result through UTF-8 storage.

Libraries/LibJS/Runtime/GlobalObject.cpp

@@ -588,36 +588,43 @@ JS_DEFINE_NATIVE_FUNCTION(GlobalObject::escape)
 JS_DEFINE_NATIVE_FUNCTION(GlobalObject::unescape)
 {
     // 1. Set string to ? ToString(string).
-    auto string = TRY(vm.argument(0).to_byte_string(vm));
+    auto string = TRY(vm.argument(0).to_utf16_string(vm));
 
     // 2. Let length be the length of string.
-    ssize_t length = string.length();
+    ssize_t length = string.length_in_code_units();
 
     // 3. Let R be the empty String.
-    StringBuilder unescaped(length);
+    Utf16StringBuilder unescaped(length);
 
     // 4. Let k be 0.
     // 5. Repeat, while k ≠ length,
     for (auto k = 0; k < length; ++k) {
         // a. Let c be the code unit at index k within string.
-        u32 code_point = string[k];
+        u16 code_unit = string.code_unit_at(k);
 
         // b. If c is the code unit 0x0025 (PERCENT SIGN), then
-        if (code_point == '%') {
+        if (code_unit == '%') {
             // i. Let hexEscape be the empty String.
             // ii. Let skip be 0.
             // iii. If k ≤ length - 6 and the code unit at index k + 1 within string is the code unit 0x0075 (LATIN SMALL LETTER U), then
-            if (k <= length - 6 && string[k + 1] == 'u' && is_ascii_hex_digit(string[k + 2]) && is_ascii_hex_digit(string[k + 3]) && is_ascii_hex_digit(string[k + 4]) && is_ascii_hex_digit(string[k + 5])) {
+            if (k <= length - 6
+                && string.code_unit_at(k + 1) == 'u'
+                && is_ascii_hex_digit(string.code_unit_at(k + 2))
+                && is_ascii_hex_digit(string.code_unit_at(k + 3))
+                && is_ascii_hex_digit(string.code_unit_at(k + 4))
+                && is_ascii_hex_digit(string.code_unit_at(k + 5))) {
                 // 1. Set hexEscape to the substring of string from k + 2 to k + 6.
-                code_point = (parse_ascii_hex_digit(string[k + 2]) << 12) | (parse_ascii_hex_digit(string[k + 3]) << 8) | (parse_ascii_hex_digit(string[k + 4]) << 4) | parse_ascii_hex_digit(string[k + 5]);
+                code_unit = (parse_ascii_hex_digit(string.code_unit_at(k + 2)) << 12) | (parse_ascii_hex_digit(string.code_unit_at(k + 3)) << 8) | (parse_ascii_hex_digit(string.code_unit_at(k + 4)) << 4) | parse_ascii_hex_digit(string.code_unit_at(k + 5));
 
                 // 2. Set skip to 5.
                 k += 5;
             }
             // iv. Else if k ≤ length - 3, then
-            else if (k <= length - 3 && is_ascii_hex_digit(string[k + 1]) && is_ascii_hex_digit(string[k + 2])) {
+            else if (k <= length - 3
+                && is_ascii_hex_digit(string.code_unit_at(k + 1))
+                && is_ascii_hex_digit(string.code_unit_at(k + 2))) {
                 // 1. Set hexEscape to the substring of string from k + 1 to k + 3.
-                code_point = (parse_ascii_hex_digit(string[k + 1]) << 4) | parse_ascii_hex_digit(string[k + 2]);
+                code_unit = (parse_ascii_hex_digit(string.code_unit_at(k + 1)) << 4) | parse_ascii_hex_digit(string.code_unit_at(k + 2));
 
                 // 2. Set skip to 2.
                 k += 2;
@@ -632,13 +639,13 @@ JS_DEFINE_NATIVE_FUNCTION(GlobalObject::unescape)
         }
 
         // c. Set R to the string-concatenation of R and c.
-        unescaped.append_code_point(code_point);
+        unescaped.append_code_unit(code_unit);
 
         // d. Set k to k + 1.
     }
 
     // 6. Return R.
-    return PrimitiveString::create(vm, unescaped.to_byte_string());
+    return PrimitiveString::create(vm, unescaped.to_string());
 }
 
 }

Libraries/LibJS/Runtime/RegExpPrototype.cpp

@@ -78,9 +78,23 @@ static ThrowCompletionOr<void> increment_last_index(VM& vm, Object& regexp_objec
 }
 
 // FIXME: Add an eviction policy to bound the size of this cache.
+struct RegexCacheKey {
+    Utf16String pattern;
+    RegExpObject::Flags flags;
+
+    bool operator==(RegexCacheKey const&) const = default;
+};
+
+struct RegexCacheKeyTraits : public Traits<RegexCacheKey> {
+    static unsigned hash(RegexCacheKey const& key)
+    {
+        return pair_int_hash(key.pattern.hash(), to_underlying(key.flags));
+    }
+};
+
 static auto& regex_cache()
 {
-    static NeverDestroyed<HashMap<String, NonnullOwnPtr<regex::ECMAScriptRegex>>> cache;
+    static NeverDestroyed<HashMap<RegexCacheKey, NonnullOwnPtr<regex::ECMAScriptRegex>, RegexCacheKeyTraits>> cache;
     return *cache;
 }
 
@@ -93,13 +107,7 @@ static regex::ECMAScriptRegex const* get_or_compile_regex(RegExpObject& regexp_o
     auto const& pattern = regexp_object.pattern();
     auto flag_bits = regexp_object.flag_bits();
 
-    // Build a cache key from pattern + flag bits.
+    RegexCacheKey cache_key { pattern, flag_bits };
-    StringBuilder key_builder;
-    key_builder.append('/');
-    key_builder.append(pattern.utf16_view());
-    key_builder.append('/');
-    key_builder.append_code_point(static_cast<u8>(flag_bits));
-    auto cache_key = key_builder.to_string_without_validation();
 
     if (auto it = regex_cache().find(cache_key); it != regex_cache().end()) {
         auto* ptr = it->value.ptr();
@@ -110,7 +118,7 @@ static regex::ECMAScriptRegex const* get_or_compile_regex(RegExpObject& regexp_o
     bool unicode = has_flag(flag_bits, RegExpObject::Flags::Unicode);
     bool unicode_sets = has_flag(flag_bits, RegExpObject::Flags::UnicodeSets);
 
-    // Parse the pattern from UTF-16 source to UTF-8 with escape normalization.
+    // Normalize non-ASCII code units to ASCII escapes before compiling the pattern.
     auto parsed_pattern = parse_regex_pattern(pattern.utf16_view(), unicode, unicode_sets);
     if (parsed_pattern.is_error())
         return nullptr;
@@ -132,7 +140,7 @@ static regex::ECMAScriptRegex const* get_or_compile_regex(RegExpObject& regexp_o
 
     auto owned = make<regex::ECMAScriptRegex>(compiled.release_value());
     auto* ptr = owned.ptr();
-    regex_cache().set(cache_key, move(owned));
+    regex_cache().set(move(cache_key), move(owned));
     regexp_object.set_cached_regex(ptr);
     return ptr;
 }

Tests/LibJS/Runtime/builtins/functions/escapeUnescape.js

@@ -16,6 +16,9 @@ test("unescape", () => {
         ["%E4%F6%FC", "äöü"],
         ["%u0107", "ć"],
         ["@*_+-./", "@*_+-./"],
+        ["äöü", "äöü"],
+        ["%uD834%uDF06", "\ud834\udf06"],
+        ["%uD800", "\ud800"],
     ].forEach(test => {
         expect(unescape(test[0])).toBe(test[1]);
     });