Ruby 3.5.0dev (2025-07-19 revision 8df61bfc92463e9896465d73176663c583d8ec0c)
shape.h (8df61bfc92463e9896465d73176663c583d8ec0c)
1#ifndef RUBY_SHAPE_H
2#define RUBY_SHAPE_H
3
4#include "internal/gc.h"
5
6typedef uint16_t attr_index_t;
7typedef uint32_t shape_id_t;
8#define SHAPE_ID_NUM_BITS 32
9#define SHAPE_ID_OFFSET_NUM_BITS 19
10
11STATIC_ASSERT(shape_id_num_bits, SHAPE_ID_NUM_BITS == sizeof(shape_id_t) * CHAR_BIT);
12
13#define SHAPE_BUFFER_SIZE (1 << SHAPE_ID_OFFSET_NUM_BITS)
14#define SHAPE_ID_OFFSET_MASK (SHAPE_BUFFER_SIZE - 1)
15
16#define SHAPE_ID_HEAP_INDEX_BITS 3
17#define SHAPE_ID_HEAP_INDEX_MAX ((1 << SHAPE_ID_HEAP_INDEX_BITS) - 1)
18
19#define SHAPE_ID_FL_USHIFT (SHAPE_ID_OFFSET_NUM_BITS + SHAPE_ID_HEAP_INDEX_BITS)
20#define SHAPE_ID_HEAP_INDEX_OFFSET SHAPE_ID_FL_USHIFT
21
22// shape_id_t bits:
23// 0-18 SHAPE_ID_OFFSET_MASK
24// index in rb_shape_tree.shape_list. Allow to access `rb_shape_t *`.
25// 19-21 SHAPE_ID_HEAP_INDEX_MASK
26// index in rb_shape_tree.capacities. Allow to access slot size.
27// 22 SHAPE_ID_FL_FROZEN
28// Whether the object is frozen or not.
29// 23 SHAPE_ID_FL_HAS_OBJECT_ID
30// Whether the object has an `SHAPE_OBJ_ID` transition.
31// 24 SHAPE_ID_FL_TOO_COMPLEX
32// The object is backed by a `st_table`.
33
34enum shape_id_fl_type {
35#define RBIMPL_SHAPE_ID_FL(n) (1<<(SHAPE_ID_FL_USHIFT+n))
36
37 SHAPE_ID_HEAP_INDEX_MASK = RBIMPL_SHAPE_ID_FL(0) | RBIMPL_SHAPE_ID_FL(1) | RBIMPL_SHAPE_ID_FL(2),
38
39 SHAPE_ID_FL_FROZEN = RBIMPL_SHAPE_ID_FL(3),
40 SHAPE_ID_FL_HAS_OBJECT_ID = RBIMPL_SHAPE_ID_FL(4),
41 SHAPE_ID_FL_TOO_COMPLEX = RBIMPL_SHAPE_ID_FL(5),
42
43 SHAPE_ID_FL_NON_CANONICAL_MASK = SHAPE_ID_FL_FROZEN | SHAPE_ID_FL_HAS_OBJECT_ID,
44 SHAPE_ID_FLAGS_MASK = SHAPE_ID_HEAP_INDEX_MASK | SHAPE_ID_FL_NON_CANONICAL_MASK | SHAPE_ID_FL_TOO_COMPLEX,
45
46#undef RBIMPL_SHAPE_ID_FL
47};
48
49// This masks allows to check if a shape_id contains any ivar.
50// It rely on ROOT_SHAPE_WITH_OBJ_ID==1.
51enum {
52 SHAPE_ID_HAS_IVAR_MASK = SHAPE_ID_FL_TOO_COMPLEX | (SHAPE_ID_OFFSET_MASK - 1),
53};
54
55// The interpreter doesn't care about frozen status or slot size when reading ivars.
56// So we normalize shape_id by clearing these bits to improve cache hits.
57// JITs however might care about it.
58#define SHAPE_ID_READ_ONLY_MASK (~(SHAPE_ID_FL_FROZEN | SHAPE_ID_HEAP_INDEX_MASK))
59
60typedef uint32_t redblack_id_t;
61
62#define SHAPE_MAX_FIELDS (attr_index_t)(-1)
63#define SHAPE_FLAG_SHIFT ((SIZEOF_VALUE * CHAR_BIT) - SHAPE_ID_NUM_BITS)
64#define SHAPE_FLAG_MASK (((VALUE)-1) >> SHAPE_ID_NUM_BITS)
65
66#define SHAPE_MAX_VARIATIONS 8
67
68#define INVALID_SHAPE_ID ((shape_id_t)-1)
69#define ATTR_INDEX_NOT_SET ((attr_index_t)-1)
70
71#define ROOT_SHAPE_ID 0x0
72#define ROOT_SHAPE_WITH_OBJ_ID 0x1
73#define ROOT_TOO_COMPLEX_SHAPE_ID (ROOT_SHAPE_ID | SHAPE_ID_FL_TOO_COMPLEX)
74#define ROOT_TOO_COMPLEX_WITH_OBJ_ID (ROOT_SHAPE_WITH_OBJ_ID | SHAPE_ID_FL_TOO_COMPLEX | SHAPE_ID_FL_HAS_OBJECT_ID)
75#define SPECIAL_CONST_SHAPE_ID (ROOT_SHAPE_ID | SHAPE_ID_FL_FROZEN)
76
77typedef struct redblack_node redblack_node_t;
78
79struct rb_shape {
80 VALUE edges; // id_table from ID (ivar) to next shape
81 ID edge_name; // ID (ivar) for transition from parent to rb_shape
82 redblack_node_t *ancestor_index;
83 shape_id_t parent_id;
84 attr_index_t next_field_index; // Fields are either ivars or internal properties like `object_id`
85 attr_index_t capacity; // Total capacity of the object with this shape
86 uint8_t type;
87};
88
89typedef struct rb_shape rb_shape_t;
90
92 ID key;
93 rb_shape_t *value;
94 redblack_id_t l;
95 redblack_id_t r;
96};
97
98enum shape_type {
99 SHAPE_ROOT,
100 SHAPE_IVAR,
101 SHAPE_OBJ_ID,
102};
103
104enum shape_flags {
105 SHAPE_FL_FROZEN = 1 << 0,
106 SHAPE_FL_HAS_OBJECT_ID = 1 << 1,
107 SHAPE_FL_TOO_COMPLEX = 1 << 2,
108
109 SHAPE_FL_NON_CANONICAL_MASK = SHAPE_FL_FROZEN | SHAPE_FL_HAS_OBJECT_ID,
110};
111
112typedef struct {
113 /* object shapes */
114 rb_shape_t *shape_list;
115 rb_shape_t *root_shape;
116 const attr_index_t *capacities;
117 rb_atomic_t next_shape_id;
118
119 redblack_node_t *shape_cache;
120 unsigned int cache_size;
122
123RUBY_SYMBOL_EXPORT_BEGIN
124RUBY_EXTERN rb_shape_tree_t rb_shape_tree;
125RUBY_SYMBOL_EXPORT_END
126
127static inline shape_id_t
128rb_shapes_count(void)
129{
130 return (shape_id_t)RUBY_ATOMIC_LOAD(rb_shape_tree.next_shape_id);
131}
132
134 uint64_t pack;
135 struct {
136 shape_id_t shape_id;
137 attr_index_t index;
138 } unpack;
139};
140
141static inline shape_id_t
142RBASIC_SHAPE_ID(VALUE obj)
143{
145 RUBY_ASSERT(!RB_TYPE_P(obj, T_IMEMO) || IMEMO_TYPE_P(obj, imemo_fields));
146#if RBASIC_SHAPE_ID_FIELD
147 return (shape_id_t)((RBASIC(obj)->shape_id));
148#else
149 return (shape_id_t)((RBASIC(obj)->flags) >> SHAPE_FLAG_SHIFT);
150#endif
151}
152
153// Same as RBASIC_SHAPE_ID but with flags that have no impact
154// on reads removed. e.g. Remove FL_FROZEN.
155static inline shape_id_t
156RBASIC_SHAPE_ID_FOR_READ(VALUE obj)
157{
158 return RBASIC_SHAPE_ID(obj) & SHAPE_ID_READ_ONLY_MASK;
159}
160
161#if RUBY_DEBUG
162bool rb_shape_verify_consistency(VALUE obj, shape_id_t shape_id);
163#endif
164
165static inline void
166RBASIC_SET_SHAPE_ID(VALUE obj, shape_id_t shape_id)
167{
169 RUBY_ASSERT(!RB_TYPE_P(obj, T_IMEMO) || IMEMO_TYPE_P(obj, imemo_fields));
170#if RBASIC_SHAPE_ID_FIELD
171 RBASIC(obj)->shape_id = (VALUE)shape_id;
172#else
173 // Object shapes are occupying top bits
174 RBASIC(obj)->flags &= SHAPE_FLAG_MASK;
175 RBASIC(obj)->flags |= ((VALUE)(shape_id) << SHAPE_FLAG_SHIFT);
176#endif
177 RUBY_ASSERT(rb_shape_verify_consistency(obj, shape_id));
178}
179
180void rb_set_namespaced_class_shape_id(VALUE obj, shape_id_t shape_id);
181
182static inline void
183RB_SET_SHAPE_ID(VALUE obj, shape_id_t shape_id)
184{
185 switch (BUILTIN_TYPE(obj)) {
186 case T_CLASS:
187 case T_MODULE:
188 rb_set_namespaced_class_shape_id(obj, shape_id);
189 break;
190 default:
191 RBASIC_SET_SHAPE_ID(obj, shape_id);
192 break;
193 }
194}
195
196static inline rb_shape_t *
197RSHAPE(shape_id_t shape_id)
198{
199 uint32_t offset = (shape_id & SHAPE_ID_OFFSET_MASK);
200 RUBY_ASSERT(offset != INVALID_SHAPE_ID);
201
202 return &rb_shape_tree.shape_list[offset];
203}
204
205int32_t rb_shape_id_offset(void);
206
207RUBY_FUNC_EXPORTED shape_id_t rb_obj_shape_id(VALUE obj);
208shape_id_t rb_shape_get_next_iv_shape(shape_id_t shape_id, ID id);
209bool rb_shape_get_iv_index(shape_id_t shape_id, ID id, attr_index_t *value);
210bool rb_shape_get_iv_index_with_hint(shape_id_t shape_id, ID id, attr_index_t *value, shape_id_t *shape_id_hint);
211bool rb_shape_find_ivar(shape_id_t shape_id, ID id, shape_id_t *ivar_shape);
212
213typedef int rb_shape_foreach_transition_callback(shape_id_t shape_id, void *data);
214bool rb_shape_foreach_field(shape_id_t shape_id, rb_shape_foreach_transition_callback func, void *data);
215
216shape_id_t rb_shape_transition_frozen(VALUE obj);
217shape_id_t rb_shape_transition_complex(VALUE obj);
218shape_id_t rb_shape_transition_remove_ivar(VALUE obj, ID id, shape_id_t *removed_shape_id);
219shape_id_t rb_shape_transition_add_ivar(VALUE obj, ID id);
220shape_id_t rb_shape_transition_add_ivar_no_warnings(VALUE obj, ID id);
221shape_id_t rb_shape_transition_object_id(VALUE obj);
222shape_id_t rb_shape_transition_heap(VALUE obj, size_t heap_index);
223shape_id_t rb_shape_object_id(shape_id_t original_shape_id);
224
225void rb_shape_free_all(void);
226
227shape_id_t rb_shape_rebuild(shape_id_t initial_shape_id, shape_id_t dest_shape_id);
228void rb_shape_copy_fields(VALUE dest, VALUE *dest_buf, shape_id_t dest_shape_id, VALUE *src_buf, shape_id_t src_shape_id);
229void rb_shape_copy_complex_ivars(VALUE dest, VALUE obj, shape_id_t src_shape_id, st_table *fields_table);
230
231static inline bool
232rb_shape_too_complex_p(shape_id_t shape_id)
233{
234 return shape_id & SHAPE_ID_FL_TOO_COMPLEX;
235}
236
237static inline bool
238rb_shape_obj_too_complex_p(VALUE obj)
239{
240 return !RB_SPECIAL_CONST_P(obj) && rb_shape_too_complex_p(RBASIC_SHAPE_ID(obj));
241}
242
243static inline bool
244rb_shape_has_object_id(shape_id_t shape_id)
245{
246 return shape_id & SHAPE_ID_FL_HAS_OBJECT_ID;
247}
248
249static inline bool
250rb_shape_canonical_p(shape_id_t shape_id)
251{
252 return !(shape_id & SHAPE_ID_FL_NON_CANONICAL_MASK);
253}
254
255static inline uint8_t
256rb_shape_heap_index(shape_id_t shape_id)
257{
258 return (uint8_t)((shape_id & SHAPE_ID_HEAP_INDEX_MASK) >> SHAPE_ID_HEAP_INDEX_OFFSET);
259}
260
261static inline shape_id_t
262rb_shape_root(size_t heap_id)
263{
264 shape_id_t heap_index = (shape_id_t)(heap_id + 1);
265 shape_id_t heap_flags = heap_index << SHAPE_ID_HEAP_INDEX_OFFSET;
266
267 RUBY_ASSERT((heap_flags & SHAPE_ID_HEAP_INDEX_MASK) == heap_flags);
268 RUBY_ASSERT(rb_shape_heap_index(heap_flags) == heap_index);
269
270 return ROOT_SHAPE_ID | heap_flags;
271}
272
273static inline shape_id_t
274RSHAPE_PARENT(shape_id_t shape_id)
275{
276 return RSHAPE(shape_id)->parent_id;
277}
278
279static inline enum shape_type
280RSHAPE_TYPE(shape_id_t shape_id)
281{
282 return RSHAPE(shape_id)->type;
283}
284
285static inline bool
286RSHAPE_TYPE_P(shape_id_t shape_id, enum shape_type type)
287{
288 return RSHAPE_TYPE(shape_id) == type;
289}
290
291static inline attr_index_t
292RSHAPE_EMBEDDED_CAPACITY(shape_id_t shape_id)
293{
294 uint8_t heap_index = rb_shape_heap_index(shape_id);
295 if (heap_index) {
296 return rb_shape_tree.capacities[heap_index - 1];
297 }
298 return 0;
299}
300
301static inline attr_index_t
302RSHAPE_CAPACITY(shape_id_t shape_id)
303{
304 attr_index_t embedded_capacity = RSHAPE_EMBEDDED_CAPACITY(shape_id);
305
306 if (embedded_capacity > RSHAPE(shape_id)->capacity) {
307 return embedded_capacity;
308 }
309 else {
310 return RSHAPE(shape_id)->capacity;
311 }
312}
313
314static inline attr_index_t
315RSHAPE_LEN(shape_id_t shape_id)
316{
317 return RSHAPE(shape_id)->next_field_index;
318}
319
320static inline attr_index_t
321RSHAPE_INDEX(shape_id_t shape_id)
322{
323 return RSHAPE_LEN(shape_id) - 1;
324}
325
326static inline ID
327RSHAPE_EDGE_NAME(shape_id_t shape_id)
328{
329 return RSHAPE(shape_id)->edge_name;
330}
331
332static inline uint32_t
333ROBJECT_FIELDS_CAPACITY(VALUE obj)
334{
335 RBIMPL_ASSERT_TYPE(obj, RUBY_T_OBJECT);
336 // Asking for capacity doesn't make sense when the object is using
337 // a hash table for storing instance variables
338 RUBY_ASSERT(!rb_shape_obj_too_complex_p(obj));
339 return RSHAPE_CAPACITY(RBASIC_SHAPE_ID(obj));
340}
341
342static inline st_table *
343ROBJECT_FIELDS_HASH(VALUE obj)
344{
345 RBIMPL_ASSERT_TYPE(obj, RUBY_T_OBJECT);
346 RUBY_ASSERT(rb_shape_obj_too_complex_p(obj));
347 return (st_table *)ROBJECT(obj)->as.heap.fields;
348}
349
350static inline void
351ROBJECT_SET_FIELDS_HASH(VALUE obj, const st_table *tbl)
352{
353 RBIMPL_ASSERT_TYPE(obj, RUBY_T_OBJECT);
354 RUBY_ASSERT(rb_shape_obj_too_complex_p(obj));
355 ROBJECT(obj)->as.heap.fields = (VALUE *)tbl;
356}
357
358static inline uint32_t
359ROBJECT_FIELDS_COUNT(VALUE obj)
360{
361 if (rb_shape_obj_too_complex_p(obj)) {
362 return (uint32_t)rb_st_table_size(ROBJECT_FIELDS_HASH(obj));
363 }
364 else {
365 RBIMPL_ASSERT_TYPE(obj, RUBY_T_OBJECT);
366 RUBY_ASSERT(!rb_shape_obj_too_complex_p(obj));
367 return RSHAPE(RBASIC_SHAPE_ID(obj))->next_field_index;
368 }
369}
370
371static inline uint32_t
372RBASIC_FIELDS_COUNT(VALUE obj)
373{
374 return RSHAPE(rb_obj_shape_id(obj))->next_field_index;
375}
376
377bool rb_obj_set_shape_id(VALUE obj, shape_id_t shape_id);
378
379static inline bool
380rb_shape_obj_has_id(VALUE obj)
381{
382 return rb_shape_has_object_id(RBASIC_SHAPE_ID(obj));
383}
384
385static inline bool
386rb_shape_has_ivars(shape_id_t shape_id)
387{
388 return shape_id & SHAPE_ID_HAS_IVAR_MASK;
389}
390
391static inline bool
392rb_shape_obj_has_ivars(VALUE obj)
393{
394 return rb_shape_has_ivars(RBASIC_SHAPE_ID(obj));
395}
396
397static inline bool
398rb_shape_has_fields(shape_id_t shape_id)
399{
400 return shape_id & (SHAPE_ID_OFFSET_MASK | SHAPE_ID_FL_TOO_COMPLEX);
401}
402
403static inline bool
404rb_shape_obj_has_fields(VALUE obj)
405{
406 return rb_shape_has_fields(RBASIC_SHAPE_ID(obj));
407}
408
409static inline bool
410rb_obj_exivar_p(VALUE obj)
411{
412 switch (TYPE(obj)) {
413 case T_NONE:
414 case T_OBJECT:
415 case T_CLASS:
416 case T_MODULE:
417 case T_IMEMO:
418 return false;
419 default:
420 break;
421 }
422 return rb_shape_obj_has_fields(obj);
423}
424
425// For ext/objspace
426RUBY_SYMBOL_EXPORT_BEGIN
427typedef void each_shape_callback(shape_id_t shape_id, void *data);
428void rb_shape_each_shape_id(each_shape_callback callback, void *data);
429size_t rb_shape_memsize(shape_id_t shape);
430size_t rb_shape_edges_count(shape_id_t shape_id);
431size_t rb_shape_depth(shape_id_t shape_id);
432RUBY_SYMBOL_EXPORT_END
433
434#endif
#define RUBY_ASSERT(...)
Asserts that the given expression is truthy if and only if RUBY_DEBUG is truthy.
Definition assert.h:219
std::atomic< unsigned > rb_atomic_t
Type that is eligible for atomic operations.
Definition atomic.h:69
#define RUBY_ATOMIC_LOAD(var)
Atomic load.
Definition atomic.h:153
#define RUBY_EXTERN
Declaration of externally visible global variables.
Definition dllexport.h:45
#define TYPE(_)
Old name of rb_type.
Definition value_type.h:108
#define T_IMEMO
Old name of RUBY_T_IMEMO.
Definition value_type.h:67
#define T_NONE
Old name of RUBY_T_NONE.
Definition value_type.h:74
#define T_MODULE
Old name of RUBY_T_MODULE.
Definition value_type.h:70
#define T_OBJECT
Old name of RUBY_T_OBJECT.
Definition value_type.h:75
#define T_CLASS
Old name of RUBY_T_CLASS.
Definition value_type.h:58
#define BUILTIN_TYPE
Old name of RB_BUILTIN_TYPE.
Definition value_type.h:85
VALUE type(ANYARGS)
ANYARGS-ed function type.
#define RBASIC(obj)
Convenient casting macro.
Definition rbasic.h:40
#define ROBJECT(obj)
Convenient casting macro.
Definition robject.h:43
static bool RB_SPECIAL_CONST_P(VALUE obj)
Checks if the given object is of enum ruby_special_consts.
Definition st.h:79
uintptr_t ID
Type that represents a Ruby identifier such as a variable name.
Definition value.h:52
uintptr_t VALUE
Type that represents a Ruby object.
Definition value.h:40
static bool RB_TYPE_P(VALUE obj, enum ruby_value_type t)
Queries if the given object is of given type.
Definition value_type.h:376
@ RUBY_T_OBJECT
Definition value_type.h:116