LibJS: Get rid of unnecessary work from canonical_numeric_index_string

The spec version of canonical_numeric_index_string is absurdly complex,
and ends up converting from a string to a number, and then back again
which is both slow and also requires a few allocations and a string
compare.

Instead this patch moves away from using Values to represent canonical
a canonical index. In most cases all we need to know is whether a
PropertyKey is an integer between 0 and 2^^32-2, which we already
compute when we construct a PropertyKey so the existing is_number()
check is sufficient.

The more expensive case is handling strings containing numbers that
don't roundtrip through string conversion. In most cases these turn
into regular string properties, but for TypedArray access these
property names are not treated as normal named properties.
TypedArrays treat these numeric properties as magic indexes that are
ignored on read and are not stored (but are evaluated) on assignment.

For that reason there's now a mode flag on canonical_numeric_index_string
so that only TypedArrays take the cost of the ToString round trip test.
In order to improve the performance of this path this patch includes
some early returns to avoid conversion in cases where we can quickly
know whether a property can round trip.

6 files changed, 114 insertions, 49 deletions

diff --git a/Userland/Libraries/LibJS/Runtime/AbstractOperations.cpp b/Userland/Libraries/LibJS/Runtime/AbstractOperations.cpp
index aaf45b36f2..f275d58c84 100644
--- a/Userland/Libraries/LibJS/Runtime/AbstractOperations.cpp
+++ b/Userland/Libraries/LibJS/Runtime/AbstractOperations.cpp
@@ -1031,7 +1031,7 @@ Object* create_mapped_arguments_object(GlobalObject& global_object, FunctionObje
 }
 
 // 7.1.21 CanonicalNumericIndexString ( argument ), https://tc39.es/ecma262/#sec-canonicalnumericindexstring
-Value canonical_numeric_index_string(GlobalObject& global_object, PropertyKey const& property_key)
+CanonicalIndex canonical_numeric_index_string(PropertyKey const& property_key, CanonicalIndexMode mode)
 {
     // NOTE: If the property name is a number type (An implementation-defined optimized
     // property key type), it can be treated as a string property that has already been
@@ -1040,24 +1040,56 @@ Value canonical_numeric_index_string(GlobalObject& global_object, PropertyKey co
     VERIFY(property_key.is_string() || property_key.is_number());
 
     if (property_key.is_number())
-        return Value(property_key.as_number());
+        return CanonicalIndex(CanonicalIndex::Type::Index, property_key.as_number());
+
+    if (mode != CanonicalIndexMode::DetectNumericRoundtrip)
+        return CanonicalIndex(CanonicalIndex::Type::Undefined, 0);
 
     // 1. Assert: Type(argument) is String.
-    auto argument = Value(js_string(global_object.vm(), property_key.as_string()));
+    auto& argument = property_key.as_string();
+
+    // Handle trivial cases without a full round trip test
+    // We do not need to check for argument == "0" at this point because we
+    // already covered it with the is_number() == true path.
+    if (argument.is_empty())
+        return CanonicalIndex(CanonicalIndex::Type::Undefined, 0);
+    u32 current_index = 0;
+    if (argument.characters()[current_index] == '-') {
+        current_index++;
+        if (current_index == argument.length())
+            return CanonicalIndex(CanonicalIndex::Type::Undefined, 0);
+    }
+    if (argument.characters()[current_index] == '0') {
+        current_index++;
+        if (current_index == argument.length())
+            return CanonicalIndex(CanonicalIndex::Type::Numeric, 0);
+        if (argument.characters()[current_index] != '.')
+            return CanonicalIndex(CanonicalIndex::Type::Undefined, 0);
+        current_index++;
+        if (current_index == argument.length())
+            return CanonicalIndex(CanonicalIndex::Type::Undefined, 0);
+    }
+
+    // Short circuit a few common cases
+    if (argument == "Infinity" || argument == "-Infinity" || argument == "NaN")
+        return CanonicalIndex(CanonicalIndex::Type::Numeric, 0);
 
-    // 2. If argument is "-0", return -0𝔽.
-    if (argument.as_string().string() == "-0")
-        return Value(-0.0);
+    // Short circuit any string that doesn't start with digits
+    if (char first_non_zero = argument.characters()[current_index]; first_non_zero < '0' || first_non_zero > '9')
+        return CanonicalIndex(CanonicalIndex::Type::Undefined, 0);
 
     // 3. Let n be ! ToNumber(argument).
-    auto n = MUST(argument.to_number(global_object));
+    char* endptr;
+    auto n = Value(strtod(argument.characters(), &endptr));
+    if (endptr != argument.characters() + argument.length())
+        return CanonicalIndex(CanonicalIndex::Type::Undefined, 0);
 
     // 4. If SameValue(! ToString(n), argument) is false, return undefined.
-    if (!same_value(MUST(n.to_primitive_string(global_object)), argument))
-        return js_undefined();
+    if (n.to_string_without_side_effects() != argument)
+        return CanonicalIndex(CanonicalIndex::Type::Undefined, 0);
 
     // 5. Return n.
-    return n;
+    return CanonicalIndex(CanonicalIndex::Type::Numeric, 0);
 }
 
 // 22.1.3.17.1 GetSubstitution ( matched, str, position, captures, namedCaptures, replacement ), https://tc39.es/ecma262/#sec-getsubstitution
diff --git a/Userland/Libraries/LibJS/Runtime/AbstractOperations.h b/Userland/Libraries/LibJS/Runtime/AbstractOperations.h
index 9e0035baac..1406027da4 100644
--- a/Userland/Libraries/LibJS/Runtime/AbstractOperations.h
+++ b/Userland/Libraries/LibJS/Runtime/AbstractOperations.h
@@ -11,6 +11,7 @@
 #include <LibJS/AST.h>
 #include <LibJS/Forward.h>
 #include <LibJS/Heap/MarkedVector.h>
+#include <LibJS/Runtime/CanonicalIndex.h>
 #include <LibJS/Runtime/FunctionObject.h>
 #include <LibJS/Runtime/GlobalObject.h>
 #include <LibJS/Runtime/PrivateEnvironment.h>
@@ -39,7 +40,12 @@ bool validate_and_apply_property_descriptor(Object*, PropertyKey const&, bool ex
 ThrowCompletionOr<Object*> get_prototype_from_constructor(GlobalObject&, FunctionObject const& constructor, Object* (GlobalObject::*intrinsic_default_prototype)());
 Object* create_unmapped_arguments_object(GlobalObject&, Span<Value> arguments);
 Object* create_mapped_arguments_object(GlobalObject&, FunctionObject&, Vector<FunctionNode::Parameter> const&, Span<Value> arguments, Environment&);
-Value canonical_numeric_index_string(GlobalObject&, PropertyKey const&);
+
+enum class CanonicalIndexMode {
+    DetectNumericRoundtrip,
+    IgnoreNumericRoundtrip,
+};
+CanonicalIndex canonical_numeric_index_string(PropertyKey const&, CanonicalIndexMode needs_numeric);
 ThrowCompletionOr<String> get_substitution(GlobalObject&, Utf16View const& matched, Utf16View const& str, size_t position, Span<Value> captures, Value named_captures, Value replacement);
 
 enum class CallerMode {
diff --git a/Userland/Libraries/LibJS/Runtime/CanonicalIndex.h b/Userland/Libraries/LibJS/Runtime/CanonicalIndex.h
new file mode 100644
index 0000000000..71f81d5155
--- /dev/null
+++ b/Userland/Libraries/LibJS/Runtime/CanonicalIndex.h
@@ -0,0 +1,35 @@
+/*
+ * Copyright (c) 2022, the SerenityOS developers.
+ *
+ * SPDX-License-Identifier: BSD-2-Clause
+ */
+
+#pragma once
+
+class CanonicalIndex {
+public:
+    enum class Type {
+        Index,
+        Numeric,
+        Undefined,
+    };
+
+    CanonicalIndex(Type type, u32 index)
+        : m_type(type)
+        , m_index(index)
+    {
+    }
+
+    u32 as_index() const
+    {
+        VERIFY(is_index());
+        return m_index;
+    }
+
+    bool is_index() const { return m_type == Type::Index; }
+    bool is_undefined() const { return m_type == Type::Undefined; }
+
+private:
+    Type m_type;
+    u32 m_index;
+};
diff --git a/Userland/Libraries/LibJS/Runtime/PrimitiveString.cpp b/Userland/Libraries/LibJS/Runtime/PrimitiveString.cpp
index a113dc8f76..d518073def 100644
--- a/Userland/Libraries/LibJS/Runtime/PrimitiveString.cpp
+++ b/Userland/Libraries/LibJS/Runtime/PrimitiveString.cpp
@@ -65,16 +65,14 @@ Optional<Value> PrimitiveString::get(GlobalObject& global_object, PropertyKey co
             return Value(static_cast<double>(length));
         }
     }
-    auto index = canonical_numeric_index_string(global_object, property_key);
-    if (index.type() != JS::Value::Type::Int32)
-        return {};
-    if (index.as_i32() < 0)
+    auto index = canonical_numeric_index_string(property_key, CanonicalIndexMode::IgnoreNumericRoundtrip);
+    if (!index.is_index())
         return {};
     auto str = utf16_string_view();
     auto length = str.length_in_code_units();
-    if (static_cast<i32>(length) <= index.as_i32())
+    if (length <= index.as_index())
         return {};
-    return js_string(vm(), str.substring_view(index.as_i32(), 1));
+    return js_string(vm(), str.substring_view(index.as_index(), 1));
 }
 
 PrimitiveString* js_string(Heap& heap, Utf16View const& view)
diff --git a/Userland/Libraries/LibJS/Runtime/StringObject.cpp b/Userland/Libraries/LibJS/Runtime/StringObject.cpp
index f92a65eb43..4434ee1e40 100644
--- a/Userland/Libraries/LibJS/Runtime/StringObject.cpp
+++ b/Userland/Libraries/LibJS/Runtime/StringObject.cpp
@@ -44,7 +44,7 @@ void StringObject::visit_edges(Cell::Visitor& visitor)
 }
 
 // 10.4.3.5 StringGetOwnProperty ( S, P ), https://tc39.es/ecma262/#sec-stringgetownproperty
-static Optional<PropertyDescriptor> string_get_own_property(GlobalObject& global_object, StringObject const& string, PropertyKey const& property_key)
+static Optional<PropertyDescriptor> string_get_own_property(StringObject const& string, PropertyKey const& property_key)
 {
     // 1. Assert: S is an Object that has a [[StringData]] internal slot.
     // 2. Assert: IsPropertyKey(P) is true.
@@ -57,15 +57,12 @@ static Optional<PropertyDescriptor> string_get_own_property(GlobalObject& global
         return {};
 
     // 4. Let index be ! CanonicalNumericIndexString(P).
-    auto index = canonical_numeric_index_string(global_object, property_key);
+    auto index = canonical_numeric_index_string(property_key, CanonicalIndexMode::IgnoreNumericRoundtrip);
+
     // 5. If index is undefined, return undefined.
-    if (index.is_undefined())
-        return {};
     // 6. If IsIntegralNumber(index) is false, return undefined.
-    if (!index.is_integral_number())
-        return {};
     // 7. If index is -0𝔽, return undefined.
-    if (index.is_negative_zero())
+    if (!index.is_index())
         return {};
 
     // 8. Let str be S.[[StringData]].
@@ -76,11 +73,11 @@ static Optional<PropertyDescriptor> string_get_own_property(GlobalObject& global
     auto length = str.length_in_code_units();
 
     // 11. If ℝ(index) < 0 or len ≤ ℝ(index), return undefined.
-    if (index.as_double() < 0 || length <= index.as_double())
+    if (length <= index.as_index())
         return {};
 
     // 12. Let resultStr be the String value of length 1, containing one code unit from str, specifically the code unit at index ℝ(index).
-    auto result_str = js_string(string.vm(), str.substring_view(index.as_double(), 1));
+    auto result_str = js_string(string.vm(), str.substring_view(index.as_index(), 1));
 
     // 13. Return the PropertyDescriptor { [[Value]]: resultStr, [[Writable]]: false, [[Enumerable]]: true, [[Configurable]]: false }.
     return PropertyDescriptor {
@@ -104,7 +101,7 @@ ThrowCompletionOr<Optional<PropertyDescriptor>> StringObject::internal_get_own_p
         return descriptor;
 
     // 4. Return ! StringGetOwnProperty(S, P).
-    return string_get_own_property(global_object(), *this, property_key);
+    return string_get_own_property(*this, property_key);
 }
 
 // 10.4.3.2 [[DefineOwnProperty]] ( P, Desc ), https://tc39.es/ecma262/#sec-string-exotic-objects-defineownproperty-p-desc
@@ -114,7 +111,7 @@ ThrowCompletionOr<bool> StringObject::internal_define_own_property(PropertyKey c
     VERIFY(property_key.is_valid());
 
     // 2. Let stringDesc be ! StringGetOwnProperty(S, P).
-    auto string_descriptor = string_get_own_property(global_object(), *this, property_key);
+    auto string_descriptor = string_get_own_property(*this, property_key);
 
     // 3. If stringDesc is not undefined, then
     if (string_descriptor.has_value()) {
diff --git a/Userland/Libraries/LibJS/Runtime/TypedArray.h b/Userland/Libraries/LibJS/Runtime/TypedArray.h
index 640e516457..4e81d0eb38 100644
--- a/Userland/Libraries/LibJS/Runtime/TypedArray.h
+++ b/Userland/Libraries/LibJS/Runtime/TypedArray.h
@@ -74,31 +74,28 @@ private:
 };
 
 // 10.4.5.9 IsValidIntegerIndex ( O, index ), https://tc39.es/ecma262/#sec-isvalidintegerindex
-inline bool is_valid_integer_index(TypedArrayBase const& typed_array, Value property_index)
+inline bool is_valid_integer_index(TypedArrayBase const& typed_array, CanonicalIndex property_index)
 {
+    // 1. If IsDetachedBuffer(O.[[ViewedArrayBuffer]]) is true, return false.
     if (typed_array.viewed_array_buffer()->is_detached())
         return false;
 
-    // TODO: This can be optimized by skipping the following 3 out of 4 checks if property_index
-    //  came from a number-type PropertyKey instead of a canonicalized string-type PropertyKey
-
-    // If ! IsIntegralNumber(index) is false, return false.
-    if (!property_index.is_integral_number())
-        return false;
-    // If index is -0𝔽, return false.
-    if (property_index.is_negative_zero())
+    // 2. If ! IsIntegralNumber(index) is false, return false.
+    // 3. If index is -0𝔽, return false.
+    if (!property_index.is_index())
         return false;
 
-    // If ℝ(index) < 0 or ℝ(index) ≥ O.[[ArrayLength]], return false.
-    if (property_index.as_double() < 0 || property_index.as_double() >= typed_array.array_length())
+    // 4. If ℝ(index) < 0 or ℝ(index) ≥ O.[[ArrayLength]], return false.
+    if (property_index.as_index() >= typed_array.array_length())
         return false;
 
+    // 5. Return true.
     return true;
 }
 
 // 10.4.5.10 IntegerIndexedElementGet ( O, index ), https://tc39.es/ecma262/#sec-integerindexedelementget
 template<typename T>
-inline Value integer_indexed_element_get(TypedArrayBase const& typed_array, Value property_index)
+inline Value integer_indexed_element_get(TypedArrayBase const& typed_array, CanonicalIndex property_index)
 {
     // 1. Assert: O is an Integer-Indexed exotic object.
 
@@ -112,7 +109,7 @@ inline Value integer_indexed_element_get(TypedArrayBase const& typed_array, Valu
     // 4. Let arrayTypeName be the String value of O.[[TypedArrayName]].
     // 5. Let elementSize be the Element Size value specified in Table 64 for arrayTypeName.
     // 6. Let indexedPosition be (ℝ(index) × elementSize) + offset.
-    Checked<size_t> indexed_position = (i64)property_index.as_double();
+    Checked<size_t> indexed_position = property_index.as_index();
     indexed_position *= typed_array.element_size();
     indexed_position += offset;
     // FIXME: Not exactly sure what we should do when overflow occurs.
@@ -130,7 +127,7 @@ inline Value integer_indexed_element_get(TypedArrayBase const& typed_array, Valu
 // 10.4.5.11 IntegerIndexedElementSet ( O, index, value ), https://tc39.es/ecma262/#sec-integerindexedelementset
 // NOTE: In error cases, the function will return as if it succeeded.
 template<typename T>
-inline ThrowCompletionOr<void> integer_indexed_element_set(TypedArrayBase& typed_array, Value property_index, Value value)
+inline ThrowCompletionOr<void> integer_indexed_element_set(TypedArrayBase& typed_array, CanonicalIndex property_index, Value value)
 {
     VERIFY(!value.is_empty());
     auto& global_object = typed_array.global_object();
@@ -157,7 +154,7 @@ inline ThrowCompletionOr<void> integer_indexed_element_set(TypedArrayBase& typed
     // b. Let arrayTypeName be the String value of O.[[TypedArrayName]].
     // c. Let elementSize be the Element Size value specified in Table 64 for arrayTypeName.
     // d. Let indexedPosition be (ℝ(index) × elementSize) + offset.
-    Checked<size_t> indexed_position = (i64)property_index.as_double();
+    Checked<size_t> indexed_position = property_index.as_index();
     indexed_position *= typed_array.element_size();
     indexed_position += offset;
     // FIXME: Not exactly sure what we should do when overflow occurs.
@@ -198,7 +195,7 @@ public:
         // NOTE: This includes an implementation-defined optimization, see note above!
         if (property_key.is_string() || property_key.is_number()) {
             // a. Let numericIndex be ! CanonicalNumericIndexString(P).
-            auto numeric_index = canonical_numeric_index_string(global_object(), property_key);
+            auto numeric_index = canonical_numeric_index_string(property_key, CanonicalIndexMode::DetectNumericRoundtrip);
             // b. If numericIndex is not undefined, then
             if (!numeric_index.is_undefined()) {
                 // i. Let value be ! IntegerIndexedElementGet(O, numericIndex).
@@ -238,7 +235,7 @@ public:
         // NOTE: This includes an implementation-defined optimization, see note above!
         if (property_key.is_string() || property_key.is_number()) {
             // a. Let numericIndex be ! CanonicalNumericIndexString(P).
-            auto numeric_index = canonical_numeric_index_string(global_object(), property_key);
+            auto numeric_index = canonical_numeric_index_string(property_key, CanonicalIndexMode::DetectNumericRoundtrip);
             // b. If numericIndex is not undefined, return ! IsValidIntegerIndex(O, numericIndex).
             if (!numeric_index.is_undefined())
                 return is_valid_integer_index(*this, numeric_index);
@@ -264,7 +261,7 @@ public:
         // NOTE: This includes an implementation-defined optimization, see note above!
         if (property_key.is_string() || property_key.is_number()) {
             // a. Let numericIndex be ! CanonicalNumericIndexString(P).
-            auto numeric_index = canonical_numeric_index_string(global_object(), property_key);
+            auto numeric_index = canonical_numeric_index_string(property_key, CanonicalIndexMode::DetectNumericRoundtrip);
             // b. If numericIndex is not undefined, then
             if (!numeric_index.is_undefined()) {
                 // i. If ! IsValidIntegerIndex(O, numericIndex) is false, return false.
@@ -315,7 +312,7 @@ public:
         // NOTE: This includes an implementation-defined optimization, see note above!
         if (property_key.is_string() || property_key.is_number()) {
             // a. Let numericIndex be ! CanonicalNumericIndexString(P).
-            auto numeric_index = canonical_numeric_index_string(global_object(), property_key);
+            auto numeric_index = canonical_numeric_index_string(property_key, CanonicalIndexMode::DetectNumericRoundtrip);
             // b. If numericIndex is not undefined, then
             if (!numeric_index.is_undefined()) {
                 // i. Return ! IntegerIndexedElementGet(O, numericIndex).
@@ -343,7 +340,7 @@ public:
         // NOTE: This includes an implementation-defined optimization, see note above!
         if (property_key.is_string() || property_key.is_number()) {
             // a. Let numericIndex be ! CanonicalNumericIndexString(P).
-            auto numeric_index = canonical_numeric_index_string(global_object(), property_key);
+            auto numeric_index = canonical_numeric_index_string(property_key, CanonicalIndexMode::DetectNumericRoundtrip);
             // b. If numericIndex is not undefined, then
             if (!numeric_index.is_undefined()) {
                 // i. Perform ? IntegerIndexedElementSet(O, numericIndex, V).
@@ -373,7 +370,7 @@ public:
         // NOTE: This includes an implementation-defined optimization, see note above!
         if (property_key.is_string() || property_key.is_number()) {
             // a. Let numericIndex be ! CanonicalNumericIndexString(P).
-            auto numeric_index = canonical_numeric_index_string(global_object(), property_key);
+            auto numeric_index = canonical_numeric_index_string(property_key, CanonicalIndexMode::DetectNumericRoundtrip);
             // b. If numericIndex is not undefined, then
             if (!numeric_index.is_undefined()) {
                 // i. If ! IsValidIntegerIndex(O, numericIndex) is false, return true; else return false.