Merge branch 'main' into jit_float_truediv

Author: eendebakpt

Doc/c-api/interp-lifecycle.rst

@@ -410,6 +410,11 @@ Initializing and finalizing the interpreter
    (zero) if not.  After :c:func:`Py_FinalizeEx` is called, this returns false until
    :c:func:`Py_Initialize` is called again.
 
+   .. versionchanged:: next
+      This function no longer returns true until initialization has fully
+      completed, including import of the :mod:`site` module.  Previously it
+      could return true while :c:func:`Py_Initialize` was still running.
+
 
 .. c:function:: int Py_IsFinalizing()
 

Doc/library/functions.rst

@@ -644,7 +644,7 @@ are always available.  They are listed here in alphabetical order.
    If the given source is a string, then leading and trailing spaces and tabs
    are stripped.
 
-   See :func:`ast.literal_eval` for a function that can safely evaluate strings
+   See :func:`ast.literal_eval` for a function to evaluate strings
    with expressions containing only literals.
 
    .. audit-event:: exec code_object eval

Doc/library/multiprocessing.rst

@@ -932,7 +932,8 @@ For an example of the usage of queues for interprocess communication see
    standard library's :mod:`queue` module are raised to signal timeouts.
 
    :class:`Queue` implements all the methods of :class:`queue.Queue` except for
-   :meth:`~queue.Queue.task_done` and :meth:`~queue.Queue.join`.
+   :meth:`~queue.Queue.task_done`, :meth:`~queue.Queue.join`, and
+   :meth:`~queue.Queue.shutdown`.
 
    .. method:: qsize()
 

Include/internal/pycore_dict.h

@@ -292,7 +292,7 @@ _PyDict_NotifyEvent(PyDict_WatchEvent event,
                     PyObject *value)
 {
     assert(Py_REFCNT((PyObject*)mp) > 0);
-    int watcher_bits = mp->_ma_watcher_tag & DICT_WATCHER_MASK;
+    int watcher_bits = FT_ATOMIC_LOAD_UINT64_RELAXED(mp->_ma_watcher_tag) & DICT_WATCHER_MASK;
     if (watcher_bits) {
         RARE_EVENT_STAT_INC(watched_dict_modification);
         _PyDict_SendEvent(watcher_bits, event, mp, key, value);
@@ -368,7 +368,7 @@ PyDictObject *_PyObject_MaterializeManagedDict_LockHeld(PyObject *);
 static inline Py_ssize_t
 _PyDict_UniqueId(PyDictObject *mp)
 {
-    return (Py_ssize_t)(mp->_ma_watcher_tag >> DICT_UNIQUE_ID_SHIFT);
+    return (Py_ssize_t)(FT_ATOMIC_LOAD_UINT64_RELAXED(mp->_ma_watcher_tag) >> DICT_UNIQUE_ID_SHIFT);
 }
 
 static inline void

Include/internal/pycore_opcode_metadata.h

@@ -424,8 +424,6 @@ extern PyCodeObject *_Py_uop_sym_get_probable_func_code(JitOptRef sym);
 extern PyObject *_Py_uop_sym_get_probable_value(JitOptRef sym);
 extern PyTypeObject *_Py_uop_sym_get_probable_type(JitOptRef sym);
 extern JitOptRef *_Py_uop_sym_set_stack_depth(JitOptContext *ctx, int stack_depth, JitOptRef *current_sp);
-extern uint32_t _Py_uop_sym_get_func_version(JitOptRef ref);
-bool _Py_uop_sym_set_func_version(JitOptContext *ctx, JitOptRef ref, uint32_t version);
 
 extern void _Py_uop_abstractcontext_init(JitOptContext *ctx, _PyBloomFilter *dependencies);
 extern void _Py_uop_abstractcontext_fini(JitOptContext *ctx);

Include/internal/pycore_optimizer.h

@@ -36,7 +36,6 @@ typedef enum _JitSymType {
     JIT_SYM_NON_NULL_TAG = 3,
     JIT_SYM_BOTTOM_TAG = 4,
     JIT_SYM_TYPE_VERSION_TAG = 5,
-    JIT_SYM_FUNC_VERSION_TAG = 6,
     JIT_SYM_KNOWN_CLASS_TAG = 7,
     JIT_SYM_KNOWN_VALUE_TAG = 8,
     JIT_SYM_TUPLE_TAG = 9,

Include/internal/pycore_optimizer_types.h

@@ -49,6 +49,8 @@ extern "C" {
     _Py_atomic_load_uint16_relaxed(&value)
 #define FT_ATOMIC_LOAD_UINT32_RELAXED(value) \
     _Py_atomic_load_uint32_relaxed(&value)
+#define FT_ATOMIC_LOAD_UINT64_RELAXED(value) \
+    _Py_atomic_load_uint64_relaxed(&value)
 #define FT_ATOMIC_LOAD_ULONG_RELAXED(value) \
     _Py_atomic_load_ulong_relaxed(&value)
 #define FT_ATOMIC_STORE_PTR_RELAXED(value, new_value) \
@@ -71,6 +73,12 @@ extern "C" {
     _Py_atomic_store_uint16_relaxed(&value, new_value)
 #define FT_ATOMIC_STORE_UINT32_RELAXED(value, new_value) \
     _Py_atomic_store_uint32_relaxed(&value, new_value)
+#define FT_ATOMIC_AND_UINT64(value, new_value) \
+    (void)_Py_atomic_and_uint64(&value, new_value)
+#define FT_ATOMIC_OR_UINT64(value, new_value) \
+    (void)_Py_atomic_or_uint64(&value, new_value)
+#define FT_ATOMIC_ADD_UINT64(value, new_value) \
+    (void)_Py_atomic_add_uint64(&value, new_value)
 #define FT_ATOMIC_STORE_CHAR_RELAXED(value, new_value) \
     _Py_atomic_store_char_relaxed(&value, new_value)
 #define FT_ATOMIC_LOAD_CHAR_RELAXED(value) \
@@ -146,6 +154,7 @@ extern "C" {
 #define FT_ATOMIC_LOAD_UINT8_RELAXED(value) value
 #define FT_ATOMIC_LOAD_UINT16_RELAXED(value) value
 #define FT_ATOMIC_LOAD_UINT32_RELAXED(value) value
+#define FT_ATOMIC_LOAD_UINT64_RELAXED(value) value
 #define FT_ATOMIC_LOAD_ULONG_RELAXED(value) value
 #define FT_ATOMIC_STORE_PTR_RELAXED(value, new_value) value = new_value
 #define FT_ATOMIC_STORE_PTR_RELEASE(value, new_value) value = new_value
@@ -157,6 +166,9 @@ extern "C" {
 #define FT_ATOMIC_STORE_UINT8_RELAXED(value, new_value) value = new_value
 #define FT_ATOMIC_STORE_UINT16_RELAXED(value, new_value) value = new_value
 #define FT_ATOMIC_STORE_UINT32_RELAXED(value, new_value) value = new_value
+#define FT_ATOMIC_AND_UINT64(value, new_value) (void)(value &= new_value)
+#define FT_ATOMIC_OR_UINT64(value, new_value) (void)(value |= new_value)
+#define FT_ATOMIC_ADD_UINT64(value, new_value) (void)(value += new_value)
 #define FT_ATOMIC_LOAD_CHAR_RELAXED(value) value
 #define FT_ATOMIC_STORE_CHAR_RELAXED(value, new_value) value = new_value
 #define FT_ATOMIC_LOAD_UCHAR_RELAXED(value) value

Include/internal/pycore_pyatomic_ft_wrappers.h

@@ -56,6 +56,29 @@ _PyRuntimeState_SetFinalizing(_PyRuntimeState *runtime, PyThreadState *tstate) {
     }
 }
 
+// Atomic so a thread that reads initialized=1 observes all writes
+// from the initialization sequence (gh-146302).
+
+static inline int
+_PyRuntimeState_GetCoreInitialized(_PyRuntimeState *runtime) {
+    return _Py_atomic_load_int(&runtime->core_initialized);
+}
+
+static inline void
+_PyRuntimeState_SetCoreInitialized(_PyRuntimeState *runtime, int initialized) {
+    _Py_atomic_store_int(&runtime->core_initialized, initialized);
+}
+
+static inline int
+_PyRuntimeState_GetInitialized(_PyRuntimeState *runtime) {
+    return _Py_atomic_load_int(&runtime->initialized);
+}
+
+static inline void
+_PyRuntimeState_SetInitialized(_PyRuntimeState *runtime, int initialized) {
+    _Py_atomic_store_int(&runtime->initialized, initialized);
+}
+
 
 #ifdef __cplusplus
 }

Include/internal/pycore_runtime.h

@@ -158,10 +158,18 @@ struct pyruntimestate {
     /* Is Python preinitialized? Set to 1 by Py_PreInitialize() */
     int preinitialized;
 
-    /* Is Python core initialized? Set to 1 by _Py_InitializeCore() */
+    /* Is Python core initialized? Set to 1 by _Py_InitializeCore().
+
+       Use _PyRuntimeState_GetCoreInitialized() and
+       _PyRuntimeState_SetCoreInitialized() to access it,
+       don't access it directly. */
     int core_initialized;
 
-    /* Is Python fully initialized? Set to 1 by Py_Initialize() */
+    /* Is Python fully initialized? Set to 1 by Py_Initialize().
+
+       Use _PyRuntimeState_GetInitialized() and
+       _PyRuntimeState_SetInitialized() to access it,
+       don't access it directly. */
     int initialized;
 
     /* Set by Py_FinalizeEx(). Only reset to NULL if Py_Initialize()