LibPDF: Store indirect value refs in Value objects

IndirectValueRef is so simple that it can be stored directly in the
Value class instead of being heap allocated.

As the comment in Value says, however, in theory the max bits needed to
store is 48 (16 for the generation index and 32(?) for the object
index), but 32 should be good enough for now. We can increase it to u64
later if necessary.

Userland/Libraries/LibPDF/Document.cpp

@@ -101,20 +101,18 @@ Page Document::get_page(u32 index)
 
 Value Document::resolve(const Value& value)
 {
+    if (value.is_ref()) {
+        // FIXME: Surely indirect PDF objects can't contain another indirect PDF object,
+        // right? Unsure from the spec, but if they can, these return values would have
+        // to be wrapped with another resolve() call.
+        return get_or_load_value(value.as_ref_index());
+    }
+
     if (!value.is_object())
         return value;
 
     auto obj = value.as_object();
 
-    // FIXME: Surely indirect PDF objects can't contain another indirect PDF object,
-    // right? Unsure from the spec, but if they can, these return values would have
-    // to be wrapped with another resolve() call.
-
-    if (obj->is_indirect_value_ref()) {
-        auto object_index = static_cast<NonnullRefPtr<IndirectValueRef>>(obj)->index();
-        return get_or_load_value(object_index);
-    }
-
     if (obj->is_indirect_value())
         return static_cast<NonnullRefPtr<IndirectValue>>(obj)->value();
 
@@ -137,13 +135,13 @@ void Document::add_page_tree_node_to_page_tree(NonnullRefPtr<DictObject> page_tr
         // these pages to the overall page tree
 
         for (auto& value : *kids_array) {
-            auto reference = object_cast<IndirectValueRef>(value.as_object());
-            auto byte_offset = m_xref_table.byte_offset_for_object(reference->index());
+            auto reference_index = value.as_ref_index();
+            auto byte_offset = m_xref_table.byte_offset_for_object(reference_index);
             auto maybe_page_tree_node = m_parser.conditionally_parse_page_tree_node_at_offset(byte_offset);
             if (maybe_page_tree_node) {
                 add_page_tree_node_to_page_tree(maybe_page_tree_node.release_nonnull());
             } else {
-                m_page_object_indices.append(reference->index());
+                m_page_object_indices.append(reference_index);
             }
         }
 
@@ -151,10 +149,8 @@ void Document::add_page_tree_node_to_page_tree(NonnullRefPtr<DictObject> page_tr
     }
 
     // We know all of the kids are leaf nodes
-    for (auto& value : *kids_array) {
-        auto reference = object_cast<IndirectValueRef>(value.as_object());
-        m_page_object_indices.append(reference->index());
-    }
+    for (auto& value : *kids_array)
+        m_page_object_indices.append(value.as_ref_index());
 }
 
 }

Userland/Libraries/LibPDF/Forward.h

@@ -14,18 +14,14 @@ class Object;
 // Note: This macro doesn't care about PlainTextStreamObject and EncodedStreamObject because
 //       we never need to work directly with either of them.
 
-#define ENUMERATE_DIRECT_OBJECT_TYPES(V) \
-    V(StringObject, string)              \
-    V(NameObject, name)                  \
-    V(ArrayObject, array)                \
-    V(DictObject, dict)                  \
-    V(StreamObject, stream)              \
+#define ENUMERATE_OBJECT_TYPES(V) \
+    V(StringObject, string)       \
+    V(NameObject, name)           \
+    V(ArrayObject, array)         \
+    V(DictObject, dict)           \
+    V(StreamObject, stream)       \
     V(IndirectValue, indirect_value)
 
-#define ENUMERATE_OBJECT_TYPES(V)    \
-    ENUMERATE_DIRECT_OBJECT_TYPES(V) \
-    V(IndirectValueRef, indirect_value_ref)
-
 #define FORWARD_DECL(class_name, _) class class_name;
 ENUMERATE_OBJECT_TYPES(FORWARD_DECL)
 #undef FORWARD_DECL

Userland/Libraries/LibPDF/Object.cpp

@@ -30,7 +30,7 @@ NonnullRefPtr<Object> DictObject::get_object(Document* document, const FlyString
     {                                                                                                      \
         return document->resolve_to<class_name>(get(key).value());                                         \
     }
-ENUMERATE_DIRECT_OBJECT_TYPES(DEFINE_ACCESSORS)
+ENUMERATE_OBJECT_TYPES(DEFINE_ACCESSORS)
 #undef DEFINE_INDEXER
 
 static void append_indent(StringBuilder& builder, int indent)
@@ -132,9 +132,4 @@ String IndirectValue::to_string(int indent) const
     return builder.to_string();
 }
 
-String IndirectValueRef::to_string(int) const
-{
-    return String::formatted("{} {} R", index(), generation_index());
-}
-
 }

Userland/Libraries/LibPDF/Object.h

@@ -222,26 +222,6 @@ private:
     Value m_value;
 };
 
-class IndirectValueRef final : public Object {
-public:
-    IndirectValueRef(u32 index, u32 generation_index)
-        : m_index(index)
-    {
-        set_generation_index(generation_index);
-    }
-
-    ~IndirectValueRef() override = default;
-
-    [[nodiscard]] ALWAYS_INLINE u32 index() const { return m_index; }
-
-    ALWAYS_INLINE bool is_indirect_value_ref() const override { return true; }
-    ALWAYS_INLINE const char* type_name() const override { return "indirect_object_ref"; }
-    String to_string(int indent) const override;
-
-private:
-    u32 m_index;
-};
-
 template<IsObject To, IsObject From>
 [[nodiscard]] ALWAYS_INLINE static NonnullRefPtr<To> object_cast(NonnullRefPtr<From> obj
 #ifdef PDF_DEBUG

Userland/Libraries/LibPDF/Parser.cpp

@@ -321,7 +321,7 @@ Value Parser::parse_possible_indirect_value_or_ref()
         m_reader.discard();
         consume();
         consume_whitespace();
-        return make_object<IndirectValueRef>(first_number.as_int(), second_number.as_int());
+        return Value(first_number.as_int(), second_number.as_int());
     }
 
     if (m_reader.matches("obj")) {

Userland/Libraries/LibPDF/Value.cpp

@@ -30,6 +30,9 @@ Value& Value::operator=(const Value& other)
     case Type::Float:
         m_as_float = other.m_as_float;
         break;
+    case Type::Ref:
+        m_as_ref = other.m_as_ref;
+        break;
     case Type::Object:
         m_as_object = other.m_as_object;
         if (m_as_object)
@@ -50,6 +53,8 @@ String Value::to_string(int indent) const
         return String::number(as_int());
     case Type::Float:
         return String::number(as_float());
+    case Type::Ref:
+        return String::formatted("{} {} R", as_ref_index(), as_ref_generation_index());
     case Type::Object:
         return as_object()->to_string(indent);
     }

Userland/Libraries/LibPDF/Value.h

@@ -14,6 +14,14 @@ class Object;
 
 class Value {
 public:
+    // We store refs as u32, with 18 bits for the index and 14 bits for the
+    // generation index. The generation index is stored in the higher bits.
+    // This may need to be rethought later, as the max generation index is
+    // 2^16 and the max for the object index is probably 2^32 (I don't know
+    // exactly)
+    static constexpr auto max_ref_index = (1 << 19) - 1;            // 2 ^ 18 - 1
+    static constexpr auto max_ref_generation_index = (1 << 15) - 1; // 2 ^ 14 - 1
+
     Value()
         : m_type(Type::Null)
     {
@@ -37,6 +45,14 @@ public:
         m_as_float = f;
     }
 
+    Value(u32 index, u32 generation_index)
+        : m_type(Type::Ref)
+    {
+        VERIFY(index < max_ref_index);
+        VERIFY(generation_index < max_ref_generation_index);
+        m_as_ref = (generation_index << 14) | index;
+    }
+
     template<IsObject T>
     Value(NonnullRefPtr<T> obj)
         : m_type(Type::Object)
@@ -59,7 +75,7 @@ public:
     [[nodiscard]] ALWAYS_INLINE bool is_int() const { return m_type == Type::Int; }
     [[nodiscard]] ALWAYS_INLINE bool is_float() const { return m_type == Type::Float; }
     [[nodiscard]] ALWAYS_INLINE bool is_number() const { return is_int() || is_float(); }
-
+    [[nodiscard]] ALWAYS_INLINE bool is_ref() const { return m_type == Type::Ref; }
     [[nodiscard]] ALWAYS_INLINE bool is_object() const { return m_type == Type::Object; }
 
     [[nodiscard]] ALWAYS_INLINE bool as_bool() const
@@ -94,6 +110,18 @@ public:
         return static_cast<float>(as_int());
     }
 
+    [[nodiscard]] ALWAYS_INLINE u32 as_ref_index() const
+    {
+        VERIFY(is_ref());
+        return m_as_ref & 0x3ffff;
+    }
+
+    [[nodiscard]] ALWAYS_INLINE u32 as_ref_generation_index() const
+    {
+        VERIFY(is_ref());
+        return m_as_ref >> 18;
+    }
+
     [[nodiscard]] ALWAYS_INLINE NonnullRefPtr<Object> as_object() const { return *m_as_object; }
 
     [[nodiscard]] ALWAYS_INLINE explicit operator bool() const { return !is_null(); }
@@ -106,12 +134,14 @@ private:
         Bool,
         Int,
         Float,
+        Ref,
         Object,
     };
 
     union {
         bool m_as_bool;
         int m_as_int;
+        u32 m_as_ref;
         float m_as_float;
         Object* m_as_object;
     };