gh-115999: Specialize `STORE_ATTR` in free-threaded builds. (gh-127838)

* Add `_PyDictKeys_StringLookupSplit` which does locking on dict keys and
  use in place of `_PyDictKeys_StringLookup`.

* Change `_PyObject_TryGetInstanceAttribute` to use that function
  in the case of split keys.

* Add `unicodekeys_lookup_split` helper which allows code sharing
  between `_Py_dict_lookup` and `_PyDictKeys_StringLookupSplit`.

* Fix locking for `STORE_ATTR_INSTANCE_VALUE`.  Create
  `_GUARD_TYPE_VERSION_AND_LOCK` uop so that object stays locked and
  `tp_version_tag` cannot change.

* Pass `tp_version_tag` to `specialize_dict_access()`, ensuring
  the version we store on the cache is the correct one (in case of
  it changing during the specalize analysis).

* Split `analyze_descriptor` into `analyze_descriptor_load` and
  `analyze_descriptor_store` since those don't share much logic.
  Add `descriptor_is_class` helper function.

* In `specialize_dict_access`, double check `_PyObject_GetManagedDict()`
  in case we race and dict was materialized before the lock.

* Avoid borrowed references in `_Py_Specialize_StoreAttr()`.

* Use `specialize()` and `unspecialize()` helpers.

* Add unit tests to ensure specializing happens as expected in FT builds.

* Add unit tests to attempt to trigger data races (useful for running under TSAN).

* Add `has_split_table` function to `_testinternalcapi`.

1 files changed, 49 insertions, 16 deletions

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index cf089c368b6..30c12dd4dc9 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -1467,7 +1467,7 @@ dummy_func(
         };
 
         specializing op(_SPECIALIZE_STORE_ATTR, (counter/1, owner -- owner)) {
-            #if ENABLE_SPECIALIZATION
+            #if ENABLE_SPECIALIZATION_FT
             if (ADAPTIVE_COUNTER_TRIGGERS(counter)) {
                 PyObject *name = GETITEM(FRAME_CO_NAMES, oparg);
                 next_instr = this_instr;
@@ -1476,7 +1476,7 @@ dummy_func(
             }
             OPCODE_DEFERRED_INC(STORE_ATTR);
             ADVANCE_ADAPTIVE_COUNTER(this_instr[1].counter);
-            #endif  /* ENABLE_SPECIALIZATION */
+            #endif  /* ENABLE_SPECIALIZATION_FT */
         }
 
         op(_STORE_ATTR, (v, owner --)) {
@@ -2129,7 +2129,18 @@ dummy_func(
         op(_GUARD_TYPE_VERSION, (type_version/2, owner -- owner)) {
             PyTypeObject *tp = Py_TYPE(PyStackRef_AsPyObjectBorrow(owner));
             assert(type_version != 0);
-            EXIT_IF(tp->tp_version_tag != type_version);
+            EXIT_IF(FT_ATOMIC_LOAD_UINT_RELAXED(tp->tp_version_tag) != type_version);
+        }
+
+        op(_GUARD_TYPE_VERSION_AND_LOCK, (type_version/2, owner -- owner)) {
+            PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner);
+            assert(type_version != 0);
+            EXIT_IF(!LOCK_OBJECT(owner_o));
+            PyTypeObject *tp = Py_TYPE(owner_o);
+            if (FT_ATOMIC_LOAD_UINT_RELAXED(tp->tp_version_tag) != type_version) {
+                UNLOCK_OBJECT(owner_o);
+                EXIT_IF(true);
+            }
         }
 
         op(_CHECK_MANAGED_OBJECT_HAS_VALUES, (owner -- owner)) {
@@ -2336,8 +2347,11 @@ dummy_func(
 
             assert(Py_TYPE(owner_o)->tp_dictoffset < 0);
             assert(Py_TYPE(owner_o)->tp_flags & Py_TPFLAGS_INLINE_VALUES);
-            EXIT_IF(_PyObject_GetManagedDict(owner_o));
-            EXIT_IF(_PyObject_InlineValues(owner_o)->valid == 0);
+            if (_PyObject_GetManagedDict(owner_o) ||
+                    !FT_ATOMIC_LOAD_UINT8(_PyObject_InlineValues(owner_o)->valid)) {
+                UNLOCK_OBJECT(owner_o);
+                EXIT_IF(true);
+            }
         }
 
         op(_STORE_ATTR_INSTANCE_VALUE, (offset/1, value, owner --)) {
@@ -2347,21 +2361,20 @@ dummy_func(
             assert(_PyObject_GetManagedDict(owner_o) == NULL);
             PyObject **value_ptr = (PyObject**)(((char *)owner_o) + offset);
             PyObject *old_value = *value_ptr;
-            *value_ptr = PyStackRef_AsPyObjectSteal(value);
+            FT_ATOMIC_STORE_PTR_RELEASE(*value_ptr, PyStackRef_AsPyObjectSteal(value));
             if (old_value == NULL) {
                 PyDictValues *values = _PyObject_InlineValues(owner_o);
                 Py_ssize_t index = value_ptr - values->values;
                 _PyDictValues_AddToInsertionOrder(values, index);
             }
-            else {
-                Py_DECREF(old_value);
-            }
+            UNLOCK_OBJECT(owner_o);
+            Py_XDECREF(old_value);
             PyStackRef_CLOSE(owner);
         }
 
         macro(STORE_ATTR_INSTANCE_VALUE) =
             unused/1 +
-            _GUARD_TYPE_VERSION +
+            _GUARD_TYPE_VERSION_AND_LOCK +
             _GUARD_DORV_NO_DICT +
             _STORE_ATTR_INSTANCE_VALUE;
 
@@ -2370,16 +2383,34 @@ dummy_func(
             assert(Py_TYPE(owner_o)->tp_flags & Py_TPFLAGS_MANAGED_DICT);
             PyDictObject *dict = _PyObject_GetManagedDict(owner_o);
             DEOPT_IF(dict == NULL);
+            DEOPT_IF(!LOCK_OBJECT(dict));
+            #ifdef Py_GIL_DISABLED
+            if (dict != _PyObject_GetManagedDict(owner_o)) {
+                UNLOCK_OBJECT(dict);
+                DEOPT_IF(true);
+            }
+            #endif
             assert(PyDict_CheckExact((PyObject *)dict));
             PyObject *name = GETITEM(FRAME_CO_NAMES, oparg);
-            DEOPT_IF(hint >= (size_t)dict->ma_keys->dk_nentries);
-            DEOPT_IF(!DK_IS_UNICODE(dict->ma_keys));
+            if (hint >= (size_t)dict->ma_keys->dk_nentries ||
+                    !DK_IS_UNICODE(dict->ma_keys)) {
+                UNLOCK_OBJECT(dict);
+                DEOPT_IF(true);
+            }
             PyDictUnicodeEntry *ep = DK_UNICODE_ENTRIES(dict->ma_keys) + hint;
-            DEOPT_IF(ep->me_key != name);
+            if (ep->me_key != name) {
+                UNLOCK_OBJECT(dict);
+                DEOPT_IF(true);
+            }
             PyObject *old_value = ep->me_value;
-            DEOPT_IF(old_value == NULL);
+            if (old_value == NULL) {
+                UNLOCK_OBJECT(dict);
+                DEOPT_IF(true);
+            }
             _PyDict_NotifyEvent(tstate->interp, PyDict_EVENT_MODIFIED, dict, name, PyStackRef_AsPyObjectBorrow(value));
-            ep->me_value = PyStackRef_AsPyObjectSteal(value);
+            FT_ATOMIC_STORE_PTR_RELEASE(ep->me_value, PyStackRef_AsPyObjectSteal(value));
+            UNLOCK_OBJECT(dict);
+
             // old_value should be DECREFed after GC track checking is done, if not, it could raise a segmentation fault,
             // when dict only holds the strong reference to value in ep->me_value.
             Py_XDECREF(old_value);
@@ -2395,10 +2426,12 @@ dummy_func(
         op(_STORE_ATTR_SLOT, (index/1, value, owner --)) {
             PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner);
 
+            DEOPT_IF(!LOCK_OBJECT(owner_o));
             char *addr = (char *)owner_o + index;
             STAT_INC(STORE_ATTR, hit);
             PyObject *old_value = *(PyObject **)addr;
-            *(PyObject **)addr = PyStackRef_AsPyObjectSteal(value);
+            FT_ATOMIC_STORE_PTR_RELEASE(*(PyObject **)addr, PyStackRef_AsPyObjectSteal(value));
+            UNLOCK_OBJECT(owner_o);
             Py_XDECREF(old_value);
             PyStackRef_CLOSE(owner);
         }