Revert "Revert "This commit implements the Object Shapes technique in…

… CRuby.""

This reverts commit 9a6803c.

bootstraptest/test_attr.rb

@@ -34,3 +34,19 @@ class A
     print "ok"
   end
 }, '[ruby-core:15120]'
+
+assert_equal %{ok}, %{
+  class Big
+    attr_reader :foo
+    def initialize
+      @foo = "ok"
+    end
+  end
+
+  obj = Big.new
+  100.times do |i|
+    obj.instance_variable_set(:"@ivar_\#{i}", i)
+  end
+
+  Big.new.foo
+}

compile.c

@@ -2058,20 +2058,7 @@ cdhash_set_label_i(VALUE key, VALUE val, VALUE ptr)
 static inline VALUE
 get_ivar_ic_value(rb_iseq_t *iseq,ID id)
 {
-    VALUE val;
-    struct rb_id_table *tbl = ISEQ_COMPILE_DATA(iseq)->ivar_cache_table;
-    if (tbl) {
-        if (rb_id_table_lookup(tbl,id,&val)) {
-            return val;
-        }
-    }
-    else {
-        tbl = rb_id_table_create(1);
-        ISEQ_COMPILE_DATA(iseq)->ivar_cache_table = tbl;
-    }
-    val = INT2FIX(ISEQ_BODY(iseq)->ivc_size++);
-    rb_id_table_insert(tbl,id,val);
-    return val;
+    return INT2FIX(ISEQ_BODY(iseq)->ivc_size++);
 }
 
 static inline VALUE
@@ -2472,9 +2459,13 @@ iseq_set_sequence(rb_iseq_t *iseq, LINK_ANCHOR *const anchor)
                             generated_iseq[code_index + 1 + j] = (VALUE)ic;
                         }
                         break;
+                      case TS_IVC: /* inline ivar cache */
+                        {
+                            unsigned int ic_index = FIX2UINT(operands[j]);
+                            vm_ic_attr_index_initialize(((IVC)&body->is_entries[ic_index]), INVALID_SHAPE_ID);
+                        }
                       case TS_ISE: /* inline storage entry: `once` insn */
                       case TS_ICVARC: /* inline cvar cache */
-                      case TS_IVC: /* inline ivar cache */
                         {
                             unsigned int ic_index = FIX2UINT(operands[j]);
                             IC ic = &ISEQ_IS_ENTRY_START(body, type)[ic_index].ic_cache;
@@ -11514,6 +11505,11 @@ ibf_load_code(const struct ibf_load *load, rb_iseq_t *iseq, ibf_offset_t bytecod
 
                     ISE ic = ISEQ_IS_ENTRY_START(load_body, operand_type) + op;
                     code[code_index] = (VALUE)ic;
+
+                    if (operand_type == TS_IVC) {
+                        vm_ic_attr_index_initialize(((IVC)code[code_index]), INVALID_SHAPE_ID);
+                    }
+
                 }
                 break;
               case TS_CALLDATA:

debug_counter.h

@@ -130,7 +130,6 @@ RB_DEBUG_COUNTER(frame_C2R)
 /* instance variable counts
  *
  * * ivar_get_ic_hit/miss: ivar_get inline cache (ic) hit/miss counts (VM insn)
- * * ivar_get_ic_miss_serial: ivar_get ic miss reason by serial (VM insn)
  * * ivar_get_ic_miss_unset:                      ... by unset (VM insn)
  * * ivar_get_ic_miss_noobject:                   ... by "not T_OBJECT" (VM insn)
  * * ivar_set_...: same counts with ivar_set (VM insn)
@@ -140,17 +139,17 @@ RB_DEBUG_COUNTER(frame_C2R)
  */
 RB_DEBUG_COUNTER(ivar_get_ic_hit)
 RB_DEBUG_COUNTER(ivar_get_ic_miss)
-RB_DEBUG_COUNTER(ivar_get_ic_miss_serial)
-RB_DEBUG_COUNTER(ivar_get_ic_miss_unset)
 RB_DEBUG_COUNTER(ivar_get_ic_miss_noobject)
 RB_DEBUG_COUNTER(ivar_set_ic_hit)
 RB_DEBUG_COUNTER(ivar_set_ic_miss)
-RB_DEBUG_COUNTER(ivar_set_ic_miss_serial)
-RB_DEBUG_COUNTER(ivar_set_ic_miss_unset)
 RB_DEBUG_COUNTER(ivar_set_ic_miss_iv_hit)
 RB_DEBUG_COUNTER(ivar_set_ic_miss_noobject)
 RB_DEBUG_COUNTER(ivar_get_base)
 RB_DEBUG_COUNTER(ivar_set_base)
+RB_DEBUG_COUNTER(ivar_get_ic_miss_set)
+RB_DEBUG_COUNTER(ivar_get_cc_miss_set)
+RB_DEBUG_COUNTER(ivar_get_ic_miss_unset)
+RB_DEBUG_COUNTER(ivar_get_cc_miss_unset)
 
 /* local variable counts
  *

ext/coverage/depend

@@ -165,7 +165,9 @@ coverage.o: $(top_srcdir)/ccan/check_type/check_type.h
 coverage.o: $(top_srcdir)/ccan/container_of/container_of.h
 coverage.o: $(top_srcdir)/ccan/list/list.h
 coverage.o: $(top_srcdir)/ccan/str/str.h
+coverage.o: $(top_srcdir)/constant.h
 coverage.o: $(top_srcdir)/gc.h
+coverage.o: $(top_srcdir)/id_table.h
 coverage.o: $(top_srcdir)/internal.h
 coverage.o: $(top_srcdir)/internal/array.h
 coverage.o: $(top_srcdir)/internal/compilers.h
@@ -176,12 +178,14 @@ coverage.o: $(top_srcdir)/internal/sanitizers.h
 coverage.o: $(top_srcdir)/internal/serial.h
 coverage.o: $(top_srcdir)/internal/static_assert.h
 coverage.o: $(top_srcdir)/internal/thread.h
+coverage.o: $(top_srcdir)/internal/variable.h
 coverage.o: $(top_srcdir)/internal/vm.h
 coverage.o: $(top_srcdir)/internal/warnings.h
 coverage.o: $(top_srcdir)/method.h
 coverage.o: $(top_srcdir)/node.h
 coverage.o: $(top_srcdir)/ruby_assert.h
 coverage.o: $(top_srcdir)/ruby_atomic.h
+coverage.o: $(top_srcdir)/shape.h
 coverage.o: $(top_srcdir)/thread_pthread.h
 coverage.o: $(top_srcdir)/vm_core.h
 coverage.o: $(top_srcdir)/vm_opts.h

ext/objspace/depend

@@ -350,6 +350,7 @@ objspace.o: $(top_srcdir)/internal/serial.h
 objspace.o: $(top_srcdir)/internal/static_assert.h
 objspace.o: $(top_srcdir)/internal/warnings.h
 objspace.o: $(top_srcdir)/node.h
+objspace.o: $(top_srcdir)/shape.h
 objspace.o: $(top_srcdir)/symbol.h
 objspace.o: objspace.c
 objspace.o: {$(VPATH)}id.h
@@ -533,7 +534,9 @@ objspace_dump.o: $(top_srcdir)/ccan/check_type/check_type.h
 objspace_dump.o: $(top_srcdir)/ccan/container_of/container_of.h
 objspace_dump.o: $(top_srcdir)/ccan/list/list.h
 objspace_dump.o: $(top_srcdir)/ccan/str/str.h
+objspace_dump.o: $(top_srcdir)/constant.h
 objspace_dump.o: $(top_srcdir)/gc.h
+objspace_dump.o: $(top_srcdir)/id_table.h
 objspace_dump.o: $(top_srcdir)/internal.h
 objspace_dump.o: $(top_srcdir)/internal/array.h
 objspace_dump.o: $(top_srcdir)/internal/compilers.h
@@ -544,12 +547,14 @@ objspace_dump.o: $(top_srcdir)/internal/sanitizers.h
 objspace_dump.o: $(top_srcdir)/internal/serial.h
 objspace_dump.o: $(top_srcdir)/internal/static_assert.h
 objspace_dump.o: $(top_srcdir)/internal/string.h
+objspace_dump.o: $(top_srcdir)/internal/variable.h
 objspace_dump.o: $(top_srcdir)/internal/vm.h
 objspace_dump.o: $(top_srcdir)/internal/warnings.h
 objspace_dump.o: $(top_srcdir)/method.h
 objspace_dump.o: $(top_srcdir)/node.h
 objspace_dump.o: $(top_srcdir)/ruby_assert.h
 objspace_dump.o: $(top_srcdir)/ruby_atomic.h
+objspace_dump.o: $(top_srcdir)/shape.h
 objspace_dump.o: $(top_srcdir)/thread_pthread.h
 objspace_dump.o: $(top_srcdir)/vm_core.h
 objspace_dump.o: $(top_srcdir)/vm_opts.h

gc.c

@@ -2895,8 +2895,7 @@ rb_class_instance_allocate_internal(VALUE klass, VALUE flags, bool wb_protected)
     GC_ASSERT((flags & RUBY_T_MASK) == T_OBJECT);
     GC_ASSERT(flags & ROBJECT_EMBED);
 
-    st_table *index_tbl = RCLASS_IV_INDEX_TBL(klass);
-    uint32_t index_tbl_num_entries = index_tbl == NULL ? 0 : (uint32_t)index_tbl->num_entries;
+    uint32_t index_tbl_num_entries = RCLASS_EXT(klass)->max_iv_count;
 
     size_t size;
     bool embed = true;
@@ -2931,7 +2930,7 @@ rb_class_instance_allocate_internal(VALUE klass, VALUE flags, bool wb_protected)
 #endif
     }
     else {
-        rb_init_iv_list(obj);
+        rb_ensure_iv_list_size(obj, 0, index_tbl_num_entries);
     }
 
     return obj;
@@ -3206,20 +3205,6 @@ rb_free_const_table(struct rb_id_table *tbl)
     rb_id_table_free(tbl);
 }
 
-static int
-free_iv_index_tbl_free_i(st_data_t key, st_data_t value, st_data_t data)
-{
-    xfree((void *)value);
-    return ST_CONTINUE;
-}
-
-static void
-iv_index_tbl_free(struct st_table *tbl)
-{
-    st_foreach(tbl, free_iv_index_tbl_free_i, 0);
-    st_free_table(tbl);
-}
-
 // alive: if false, target pointers can be freed already.
 //        To check it, we need objspace parameter.
 static void
@@ -3435,6 +3420,16 @@ obj_free(rb_objspace_t *objspace, VALUE obj)
             RB_DEBUG_COUNTER_INC(obj_obj_transient);
         }
         else {
+            rb_shape_t *shape = rb_shape_get_shape_by_id(ROBJECT_SHAPE_ID(obj));
+            if (shape) {
+                VALUE klass = RBASIC_CLASS(obj);
+
+                // Increment max_iv_count if applicable, used to determine size pool allocation
+                uint32_t num_of_ivs = shape->iv_count;
+                if (RCLASS_EXT(klass)->max_iv_count < num_of_ivs) {
+                    RCLASS_EXT(klass)->max_iv_count = num_of_ivs;
+                }
+            }
             xfree(RANY(obj)->as.object.as.heap.ivptr);
             RB_DEBUG_COUNTER_INC(obj_obj_ptr);
         }
@@ -3449,9 +3444,6 @@ obj_free(rb_objspace_t *objspace, VALUE obj)
         if (RCLASS_CONST_TBL(obj)) {
             rb_free_const_table(RCLASS_CONST_TBL(obj));
         }
-        if (RCLASS_IV_INDEX_TBL(obj)) {
-            iv_index_tbl_free(RCLASS_IV_INDEX_TBL(obj));
-        }
         if (RCLASS_CVC_TBL(obj)) {
             rb_id_table_foreach_values(RCLASS_CVC_TBL(obj), cvar_table_free_i, NULL);
             rb_id_table_free(RCLASS_CVC_TBL(obj));
@@ -4873,10 +4865,6 @@ obj_memsize_of(VALUE obj, int use_all_types)
             if (RCLASS_CVC_TBL(obj)) {
                 size += rb_id_table_memsize(RCLASS_CVC_TBL(obj));
             }
-            if (RCLASS_IV_INDEX_TBL(obj)) {
-                // TODO: more correct value
-                size += st_memsize(RCLASS_IV_INDEX_TBL(obj));
-            }
             if (RCLASS_EXT(obj)->iv_tbl) {
                 size += st_memsize(RCLASS_EXT(obj)->iv_tbl);
             }
@@ -10408,27 +10396,13 @@ update_subclass_entries(rb_objspace_t *objspace, rb_subclass_entry_t *entry)
     }
 }
 
-static int
-update_iv_index_tbl_i(st_data_t key, st_data_t value, st_data_t arg)
-{
-    rb_objspace_t *objspace = (rb_objspace_t *)arg;
-    struct rb_iv_index_tbl_entry *ent = (struct rb_iv_index_tbl_entry *)value;
-    UPDATE_IF_MOVED(objspace, ent->class_value);
-    return ST_CONTINUE;
-}
-
 static void
 update_class_ext(rb_objspace_t *objspace, rb_classext_t *ext)
 {
     UPDATE_IF_MOVED(objspace, ext->origin_);
     UPDATE_IF_MOVED(objspace, ext->includer);
     UPDATE_IF_MOVED(objspace, ext->refined_class);
     update_subclass_entries(objspace, ext->subclasses);
-
-    // ext->iv_index_tbl
-    if (ext->iv_index_tbl) {
-        st_foreach(ext->iv_index_tbl, update_iv_index_tbl_i, (st_data_t)objspace);
-    }
 }
 
 static void

include/ruby/internal/core/robject.h

@@ -46,7 +46,6 @@
 #define ROBJECT_EMBED         ROBJECT_EMBED
 #define ROBJECT_NUMIV         ROBJECT_NUMIV
 #define ROBJECT_IVPTR         ROBJECT_IVPTR
-#define ROBJECT_IV_INDEX_TBL  ROBJECT_IV_INDEX_TBL
 /** @endcond */
 
 /**
@@ -132,7 +131,7 @@ struct RObject {
              *
              * This is a shortcut for `RCLASS_IV_INDEX_TBL(rb_obj_class(obj))`.
              */
-            struct st_table *iv_index_tbl;
+            struct rb_id_table *iv_index_tbl;
         } heap;
 
 #if USE_RVARGC

include/ruby/internal/fl_type.h

@@ -941,21 +941,8 @@ RB_OBJ_FREEZE_RAW(VALUE obj)
     RB_FL_SET_RAW(obj, RUBY_FL_FREEZE);
 }
 
-/**
- * Prevents further modifications to the given object.  ::rb_eFrozenError shall
- * be raised if modification is attempted.
- *
- * @param[out]  x  Object in question.
- */
-static inline void
-rb_obj_freeze_inline(VALUE x)
-{
-    if (RB_FL_ABLE(x)) {
-        RB_OBJ_FREEZE_RAW(x);
-        if (RBASIC_CLASS(x) && !(RBASIC(x)->flags & RUBY_FL_SINGLETON)) {
-            rb_freeze_singleton_class(x);
-        }
-    }
-}
+RUBY_SYMBOL_EXPORT_BEGIN
+void rb_obj_freeze_inline(VALUE obj);
+RUBY_SYMBOL_EXPORT_END
 
 #endif /* RBIMPL_FL_TYPE_H */

inits.c

@@ -77,6 +77,7 @@ rb_call_inits(void)
     CALL(vm_stack_canary);
     CALL(ast);
     CALL(gc_stress);
+    CALL(shape);
 
     // enable builtin loading
     CALL(builtin);

internal.h

@@ -48,9 +48,6 @@
 #undef RHASH_TBL
 #undef RHASH_EMPTY_P
 
-/* internal/object.h */
-#undef ROBJECT_IV_INDEX_TBL
-
 /* internal/struct.h */
 #undef RSTRUCT_LEN
 #undef RSTRUCT_PTR

internal/class.h

@@ -14,6 +14,7 @@
 #include "ruby/internal/stdbool.h"     /* for bool */
 #include "ruby/intern.h"        /* for rb_alloc_func_t */
 #include "ruby/ruby.h"          /* for struct RBasic */
+#include "shape.h"
 
 #ifdef RCLASS_SUPER
 # undef RCLASS_SUPER
@@ -27,8 +28,8 @@ struct rb_subclass_entry {
 
 struct rb_iv_index_tbl_entry {
     uint32_t index;
-    rb_serial_t class_serial;
-    VALUE class_value;
+    shape_id_t source_shape_id;
+    shape_id_t dest_shape_id;
 };
 
 struct rb_cvar_class_tbl_entry {
@@ -38,7 +39,6 @@ struct rb_cvar_class_tbl_entry {
 };
 
 struct rb_classext_struct {
-    struct st_table *iv_index_tbl; // ID -> struct rb_iv_index_tbl_entry
     struct st_table *iv_tbl;
 #if SIZEOF_SERIAL_T == SIZEOF_VALUE /* otherwise m_tbl is in struct RClass */
     struct rb_id_table *m_tbl;
@@ -64,6 +64,10 @@ struct rb_classext_struct {
     const VALUE refined_class;
     rb_alloc_func_t allocator;
     const VALUE includer;
+    uint32_t max_iv_count;
+#if !SHAPE_IN_BASIC_FLAGS
+    shape_id_t shape_id;
+#endif
 };
 
 struct RClass {
@@ -102,7 +106,6 @@ typedef struct rb_classext_struct rb_classext_t;
 #define RCLASS_CALLABLE_M_TBL(c) (RCLASS_EXT(c)->callable_m_tbl)
 #define RCLASS_CC_TBL(c) (RCLASS_EXT(c)->cc_tbl)
 #define RCLASS_CVC_TBL(c) (RCLASS_EXT(c)->cvc_tbl)
-#define RCLASS_IV_INDEX_TBL(c) (RCLASS_EXT(c)->iv_index_tbl)
 #define RCLASS_ORIGIN(c) (RCLASS_EXT(c)->origin_)
 #define RCLASS_REFINED_CLASS(c) (RCLASS_EXT(c)->refined_class)
 #if SIZEOF_SERIAL_T == SIZEOF_VALUE