[pull] main from python:main

Include/internal/pycore_jit.h

@@ -25,6 +25,7 @@ typedef _Py_CODEUNIT *(*jit_func)(
 
 int _PyJIT_Compile(_PyExecutorObject *executor, const _PyUOpInstruction *trace, size_t length);
 void _PyJIT_Free(_PyExecutorObject *executor);
+void _PyJIT_Fini(void);
 
 #endif  // _Py_JIT
 

Lib/multiprocessing/spawn.py

@@ -184,7 +184,7 @@ def get_preparation_data(name):
         sys_argv=sys.argv,
         orig_dir=process.ORIGINAL_DIR,
         dir=os.getcwd(),
-        start_method=get_start_method(),
+        start_method=get_start_method(allow_none=True),
         )
 
     # Figure out whether to initialise main in the subprocess as a module

Lib/test/_test_multiprocessing.py

@@ -5967,6 +5967,26 @@ def test_context(self):
             self.assertRaises(ValueError, ctx.set_start_method, None)
             self.check_context(ctx)
 
+    @staticmethod
+    def _dummy_func():
+        pass
+
+    @warnings_helper.ignore_fork_in_thread_deprecation_warnings()
+    def test_spawn_dont_set_context(self):
+        # Run a process with spawn or forkserver context may change
+        # the global start method, see gh-109263.
+        for method in ('fork', 'spawn', 'forkserver'):
+            multiprocessing.set_start_method(None, force=True)
+
+            try:
+                ctx = multiprocessing.get_context(method)
+            except ValueError:
+                continue
+            process = ctx.Process(target=self._dummy_func)
+            process.start()
+            process.join()
+            self.assertIsNone(multiprocessing.get_start_method(allow_none=True))
+
     def test_context_check_module_types(self):
         try:
             ctx = multiprocessing.get_context('forkserver')

Lib/test/test_capi/test_opt.py

@@ -3114,6 +3114,25 @@ def testfunc(n):
         self.assertNotIn("_POP_TOP_INT", uops)
         self.assertIn("_POP_TOP_NOP", uops)
 
+    def test_binary_subscr_tuple_int(self):
+        def testfunc(n):
+            t = (1,)
+            x = 0
+            for _ in range(n):
+                y = t[0]
+                x += y
+            return x
+
+        res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
+        self.assertEqual(res, TIER2_THRESHOLD)
+        self.assertIsNotNone(ex)
+        uops = get_opnames(ex)
+
+        self.assertIn("_BINARY_OP_SUBSCR_TUPLE_INT", uops)
+        self.assertNotIn("_POP_TOP", uops)
+        self.assertNotIn("_POP_TOP_INT", uops)
+        self.assertIn("_POP_TOP_NOP", uops)
+
     def test_143026(self):
         # https://github.com/python/cpython/issues/143026
 

Misc/NEWS.d/next/Library/2025-06-22-18-57-19.gh-issue-109263.f92V95.rst

@@ -0,0 +1,2 @@
+Starting a process from spawn context in :mod:`multiprocessing` no longer
+sets the start method globally.

Python/bytecodes.c

@@ -975,9 +975,9 @@ dummy_func(
         }
 
         macro(BINARY_OP_SUBSCR_TUPLE_INT) =
-            _GUARD_TOS_INT + _GUARD_NOS_TUPLE + unused/5 + _BINARY_OP_SUBSCR_TUPLE_INT;
+            _GUARD_TOS_INT + _GUARD_NOS_TUPLE + unused/5 + _BINARY_OP_SUBSCR_TUPLE_INT + _POP_TOP_INT + POP_TOP;
 
-        op(_BINARY_OP_SUBSCR_TUPLE_INT, (tuple_st, sub_st -- res)) {
+        op(_BINARY_OP_SUBSCR_TUPLE_INT, (tuple_st, sub_st -- res, ts, ss)) {
             PyObject *sub = PyStackRef_AsPyObjectBorrow(sub_st);
             PyObject *tuple = PyStackRef_AsPyObjectBorrow(tuple_st);
 
@@ -991,9 +991,10 @@ dummy_func(
             STAT_INC(BINARY_OP, hit);
             PyObject *res_o = PyTuple_GET_ITEM(tuple, index);
             assert(res_o != NULL);
-            PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);
             res = PyStackRef_FromPyObjectNew(res_o);
-            DECREF_INPUTS();
+            ts = tuple_st;
+            ss = sub_st;
+            INPUTS_DEAD();
         }
 
         op(_GUARD_NOS_DICT, (nos, unused -- nos, unused)) {

Python/jit.c

@@ -151,6 +151,8 @@ typedef struct {
     uintptr_t instruction_starts[UOP_MAX_TRACE_LENGTH];
 } jit_state;
 
+static size_t _Py_jit_shim_size = 0;
+
 // Warning! AArch64 requires you to get your hands dirty. These are your gloves:
 
 // value[value_start : value_start + len]
@@ -676,6 +678,7 @@ _PyJIT_Compile(_PyExecutorObject *executor, const _PyUOpInstruction trace[], siz
  * We compile this once only as it effectively a normal
  * function, but we need to use the JIT because it needs
  * to understand the jit-specific calling convention.
+ * Don't forget to call _PyJIT_Fini later!
  */
 static _PyJitEntryFuncPtr
 compile_shim(void)
@@ -717,6 +720,7 @@ compile_shim(void)
         jit_free(memory, total_size);
         return NULL;
     }
+    _Py_jit_shim_size = total_size;
     return (_PyJitEntryFuncPtr)memory;
 }
 
@@ -739,6 +743,7 @@ _Py_LazyJitShim(
     return _Py_jit_entry(executor, frame, stack_pointer, tstate);
 }
 
+// Free executor's memory allocated with _PyJIT_Compile
 void
 _PyJIT_Free(_PyExecutorObject *executor)
 {
@@ -754,4 +759,22 @@ _PyJIT_Free(_PyExecutorObject *executor)
     }
 }
 
+// Free shim memory allocated with compile_shim
+void
+_PyJIT_Fini(void)
+{
+    PyMutex_Lock(&lazy_jit_mutex);
+    unsigned char *memory = (unsigned char *)_Py_jit_entry;
+    size_t size = _Py_jit_shim_size;
+    if (size) {
+        _Py_jit_entry = _Py_LazyJitShim;
+        _Py_jit_shim_size = 0;
+        if (jit_free(memory, size)) {
+            PyErr_FormatUnraisable("Exception ignored while "
+                                   "freeing JIT entry code");
+        }
+    }
+    PyMutex_Unlock(&lazy_jit_mutex);
+}
+
 #endif  // _Py_JIT

Python/optimizer_bytecodes.c

@@ -335,7 +335,7 @@ dummy_func(void) {
         i = sub_st;
     }
 
-    op(_BINARY_OP_SUBSCR_TUPLE_INT, (tuple_st, sub_st -- res)) {
+    op(_BINARY_OP_SUBSCR_TUPLE_INT, (tuple_st, sub_st -- res, ts, ss)) {
         assert(sym_matches_type(tuple_st, &PyTuple_Type));
         if (sym_is_const(ctx, sub_st)) {
             assert(PyLong_CheckExact(sym_get_const(ctx, sub_st)));
@@ -354,6 +354,8 @@ dummy_func(void) {
         else {
             res = sym_new_not_null(ctx);
         }
+        ts = tuple_st;
+        ss = sub_st;
     }
 
     op(_TO_BOOL, (value -- res)) {

Python/pylifecycle.c

@@ -35,6 +35,9 @@
 #include "pycore_uniqueid.h"      // _PyObject_FinalizeUniqueIdPool()
 #include "pycore_warnings.h"      // _PyWarnings_InitState()
 #include "pycore_weakref.h"       // _PyWeakref_GET_REF()
+#ifdef _Py_JIT
+#include "pycore_jit.h"           // _PyJIT_Fini()
+#endif
 
 #include "opcode.h"
 
@@ -2267,6 +2270,7 @@ _Py_Finalize(_PyRuntimeState *runtime)
     /* Print debug stats if any */
     _PyEval_Fini();
 
+
     /* Flush sys.stdout and sys.stderr (again, in case more was printed) */
     if (flush_std_files() < 0) {
         status = -1;
@@ -2346,6 +2350,10 @@ _Py_Finalize(_PyRuntimeState *runtime)
 
     finalize_interp_clear(tstate);
 
+#ifdef _Py_JIT
+    /* Free JIT shim memory */
+    _PyJIT_Fini();
+#endif
 
 #ifdef Py_TRACE_REFS
     /* Display addresses (& refcnts) of all objects still alive.

Tools/tsan/suppressions_free_threading.txt

@@ -26,3 +26,6 @@ thread:pthread_create
 # PyObject_Realloc internally does memcpy which isn't atomic so can race
 # with non-locking reads. See #132070
 race:PyObject_Realloc
+
+# gh-143100: set_swap_bodies in setobject.c calls memcpy, which isn't atomic
+race:set_swap_bodies