Implement seen_obj_id on object metadata

Author: peterzhu2118

gc/default/default.c

@@ -711,6 +711,7 @@ struct free_slot {
 
 struct object_metadata {
     unsigned char age : RVALUE_AGE_BIT_COUNT;
+    unsigned char seen_obj_id : 1;
 };
 STATIC_ASSERT(SIZEOF_OBJECT_METADATA, sizeof(struct object_metadata) == 1);
 
@@ -833,6 +834,12 @@ RVALUE_AGE_GET(VALUE obj)
     return RVALUE_METADATA(obj)->age;
 }
 
+static bool
+RVALUE_SEEN_OBJ_ID(VALUE obj)
+{
+    return RVALUE_METADATA(obj)->seen_obj_id;
+}
+
 static void
 RVALUE_AGE_SET(VALUE obj, int age)
 {
@@ -1577,13 +1584,13 @@ rb_gc_impl_object_id(void *objspace_ptr, VALUE obj)
     rb_objspace_t *objspace = objspace_ptr;
 
     unsigned int lev = rb_gc_vm_lock();
-    if (FL_TEST(obj, FL_SEEN_OBJ_ID)) {
+    if (RVALUE_SEEN_OBJ_ID(obj)) {
         st_data_t val;
         if (st_lookup(objspace->obj_to_id_tbl, (st_data_t)obj, &val)) {
             id = (VALUE)val;
         }
         else {
-            rb_bug("rb_gc_impl_object_id: FL_SEEN_OBJ_ID flag set but not found in table");
+            rb_bug("rb_gc_impl_object_id: seen_obj_id flag set but not found in table");
         }
     }
     else {
@@ -1594,7 +1601,7 @@ rb_gc_impl_object_id(void *objspace_ptr, VALUE obj)
 
         st_insert(objspace->obj_to_id_tbl, (st_data_t)obj, (st_data_t)id);
         st_insert(objspace->id_to_obj_tbl, (st_data_t)id, (st_data_t)obj);
-        FL_SET(obj, FL_SEEN_OBJ_ID);
+        RVALUE_METADATA(obj)->seen_obj_id = true;
     }
     rb_gc_vm_unlock(lev);
 
@@ -2622,7 +2629,7 @@ rb_gc_impl_pointer_to_heap_p(void *objspace_ptr, const void *ptr)
     return is_pointer_to_heap(objspace_ptr, ptr);
 }
 
-#define ZOMBIE_OBJ_KEPT_FLAGS (FL_SEEN_OBJ_ID | FL_FINALIZE)
+#define ZOMBIE_OBJ_KEPT_FLAGS (FL_FINALIZE)
 
 void
 rb_gc_impl_make_zombie(void *objspace_ptr, VALUE obj, void (*dfree)(void *), void *data)
@@ -2649,8 +2656,8 @@ obj_free_object_id(rb_objspace_t *objspace, VALUE obj)
 {
     st_data_t o = (st_data_t)obj, id;
 
-    GC_ASSERT(BUILTIN_TYPE(obj) == T_NONE || FL_TEST(obj, FL_SEEN_OBJ_ID));
-    FL_UNSET(obj, FL_SEEN_OBJ_ID);
+    GC_ASSERT(BUILTIN_TYPE(obj) == T_NONE || RVALUE_SEEN_OBJ_ID(obj));
+    RVALUE_METADATA(obj)->seen_obj_id = false;
 
     if (st_delete(objspace->obj_to_id_tbl, &o, &id)) {
         GC_ASSERT(id);
@@ -2883,7 +2890,7 @@ rb_gc_impl_copy_finalizer(void *objspace_ptr, VALUE dest, VALUE obj)
 static VALUE
 get_object_id_in_finalizer(rb_objspace_t *objspace, VALUE obj)
 {
-    if (FL_TEST_RAW(obj, FL_SEEN_OBJ_ID)) {
+    if (RVALUE_SEEN_OBJ_ID(obj)) {
         return rb_gc_impl_object_id(objspace, obj);
     }
     else {
@@ -2939,7 +2946,7 @@ finalize_list(rb_objspace_t *objspace, VALUE zombie)
         int lev = rb_gc_vm_lock();
         {
             GC_ASSERT(BUILTIN_TYPE(zombie) == T_ZOMBIE);
-            if (FL_TEST_RAW(zombie, FL_SEEN_OBJ_ID)) {
+            if (RVALUE_SEEN_OBJ_ID(zombie)) {
                 obj_free_object_id(objspace, zombie);
             }
 
@@ -3547,7 +3554,7 @@ gc_sweep_plane(rb_objspace_t *objspace, rb_heap_t *heap, uintptr_t p, bits_t bit
 
                 rb_gc_event_hook(vp, RUBY_INTERNAL_EVENT_FREEOBJ);
 
-                bool has_object_id = FL_TEST_RAW(vp, FL_SEEN_OBJ_ID);
+                bool has_object_id = RVALUE_SEEN_OBJ_ID(vp);
                 rb_gc_obj_free_vm_weak_references(vp);
                 if (rb_gc_obj_free(objspace, vp)) {
                     if (has_object_id) {
@@ -6232,7 +6239,7 @@ rb_gc_impl_object_metadata(void *objspace_ptr, VALUE obj)
     if (RVALUE_MARKING(objspace, obj)) SET_ENTRY(marking, Qtrue);
     if (RVALUE_MARKED(objspace, obj)) SET_ENTRY(marked, Qtrue);
     if (RVALUE_PINNED(objspace, obj)) SET_ENTRY(pinned, Qtrue);
-    if (FL_TEST(obj, FL_SEEN_OBJ_ID)) SET_ENTRY(object_id, rb_obj_id(obj));
+    if (RVALUE_SEEN_OBJ_ID(obj)) SET_ENTRY(object_id, rb_obj_id(obj));
 
     object_metadata_entries[n].name = 0;
     object_metadata_entries[n].val = 0;
@@ -6950,7 +6957,7 @@ gc_move(rb_objspace_t *objspace, VALUE src, VALUE dest, size_t src_slot_size, si
     CLEAR_IN_BITMAP(GET_HEAP_UNCOLLECTIBLE_BITS(src), src);
     CLEAR_IN_BITMAP(GET_HEAP_PAGE(src)->remembered_bits, src);
 
-    if (FL_TEST_RAW(src, FL_SEEN_OBJ_ID)) {
+    if (RVALUE_SEEN_OBJ_ID(src)) {
         /* If the source object's object_id has been seen, we need to update
          * the object to object id mapping. */
         st_data_t srcid = (st_data_t)src, id;

include/ruby/internal/fl_type.h

@@ -61,7 +61,6 @@
 #define FL_TAINT        RBIMPL_CAST((VALUE)RUBY_FL_TAINT)                /**< @old{RUBY_FL_TAINT} */
 #define FL_SHAREABLE    RBIMPL_CAST((VALUE)RUBY_FL_SHAREABLE)            /**< @old{RUBY_FL_SHAREABLE} */
 #define FL_UNTRUSTED    RBIMPL_CAST((VALUE)RUBY_FL_UNTRUSTED)            /**< @old{RUBY_FL_UNTRUSTED} */
-#define FL_SEEN_OBJ_ID  RBIMPL_CAST((VALUE)RUBY_FL_SEEN_OBJ_ID)          /**< @old{RUBY_FL_SEEN_OBJ_ID} */
 #define FL_EXIVAR       RBIMPL_CAST((VALUE)RUBY_FL_EXIVAR)               /**< @old{RUBY_FL_EXIVAR} */
 #define FL_FREEZE       RBIMPL_CAST((VALUE)RUBY_FL_FREEZE)               /**< @old{RUBY_FL_FREEZE} */
 
@@ -268,23 +267,11 @@ ruby_fl_type {
                          = 0,
 
     /**
-     * This flag has something to do with  object IDs.  Unlike in the old days,
-     * an object's object  ID (that a user can  query using `Object#object_id`)
-     * is no longer its physical address represented using Ruby level integers.
-     * It is  now a  monotonic-increasing integer  unrelated to  the underlying
-     * memory arrangement.  Object IDs are assigned when necessary; objects are
-     * born without one,  and will eventually have such  property when queried.
-     * The interpreter has to manage which one is which.  This is the flag that
-     * helps the  management.  Objects  with this  flag set  are the  ones with
-     * object IDs assigned.
+     * This flag is no longer in use.
      *
      * @internal
-     *
-     * But honestly, @shyouhei  doesn't think this flag should  be visible from
-     * 3rd parties.  It must be an implementation detail that they should never
-     * know.  Might better be hidden.
      */
-    RUBY_FL_SEEN_OBJ_ID  = (1<<9),
+    RUBY_FL_UNUSED9  = (1<<9),
 
     /**
      * This flag has something to do with instance variables.  3rd parties need

object.c

@@ -151,8 +151,7 @@ rb_class_allocate_instance(VALUE klass)
 VALUE
 rb_obj_setup(VALUE obj, VALUE klass, VALUE type)
 {
-    VALUE ignored_flags = RUBY_FL_SEEN_OBJ_ID;
-    RBASIC(obj)->flags = (type & ~ignored_flags) | (RBASIC(obj)->flags & ignored_flags);
+    RBASIC(obj)->flags = type | RBASIC(obj)->flags;
     RBASIC_SET_CLASS(obj, klass);
     return obj;
 }

yjit/src/cruby_bindings.inc.rs

@@ -227,7 +227,6 @@ pub const RUBY_FL_FINALIZE: ruby_fl_type = 128;
 pub const RUBY_FL_TAINT: ruby_fl_type = 0;
 pub const RUBY_FL_SHAREABLE: ruby_fl_type = 256;
 pub const RUBY_FL_UNTRUSTED: ruby_fl_type = 0;
-pub const RUBY_FL_SEEN_OBJ_ID: ruby_fl_type = 512;
 pub const RUBY_FL_EXIVAR: ruby_fl_type = 1024;
 pub const RUBY_FL_FREEZE: ruby_fl_type = 2048;
 pub const RUBY_FL_USER0: ruby_fl_type = 4096;