[pull] main from python:main

Doc/conf.py

@@ -177,6 +177,7 @@
     ('c:type', '__int64'),
     ('c:type', 'unsigned __int64'),
     ('c:type', 'double'),
+    ('c:type', '_Float16'),
     # Standard C structures
     ('c:struct', 'in6_addr'),
     ('c:struct', 'in_addr'),

Doc/library/struct.rst

@@ -254,7 +254,7 @@ platform-dependent.
 +--------+--------------------------+--------------------+----------------+------------+
 | ``N``  | :c:type:`size_t`         | integer            |                | \(3)       |
 +--------+--------------------------+--------------------+----------------+------------+
-| ``e``  | \(6)                     | float              | 2              | \(4)       |
+| ``e``  | :c:expr:`_Float16`       | float              | 2              | \(4), \(6) |
 +--------+--------------------------+--------------------+----------------+------------+
 | ``f``  | :c:expr:`float`          | float              | 4              | \(4)       |
 +--------+--------------------------+--------------------+----------------+------------+
@@ -328,7 +328,9 @@ Notes:
    revision of the `IEEE 754 standard <ieee 754 standard_>`_. It has a sign
    bit, a 5-bit exponent and 11-bit precision (with 10 bits explicitly stored),
    and can represent numbers between approximately ``6.1e-05`` and ``6.5e+04``
-   at full precision. This type is not widely supported by C compilers: on a
+   at full precision. This type is not widely supported by C compilers:
+   it's available as :c:expr:`_Float16` type, if the compiler supports the Annex H
+   of the C23 standard.  On a
    typical machine, an unsigned short can be used for storage, but not for math
    operations. See the Wikipedia page on the `half-precision floating-point
    format <half precision format_>`_ for more information.

Include/internal/pycore_ceval.h

@@ -249,16 +249,7 @@ static inline void _Py_LeaveRecursiveCallTstate(PyThreadState *tstate) {
 
 PyAPI_FUNC(void) _Py_InitializeRecursionLimits(PyThreadState *tstate);
 
-static inline int _Py_ReachedRecursionLimit(PyThreadState *tstate)  {
-    uintptr_t here_addr = _Py_get_machine_stack_pointer();
-    _PyThreadStateImpl *_tstate = (_PyThreadStateImpl *)tstate;
-    assert(_tstate->c_stack_hard_limit != 0);
-#if _Py_STACK_GROWS_DOWN
-    return here_addr <= _tstate->c_stack_soft_limit;
-#else
-    return here_addr >= _tstate->c_stack_soft_limit;
-#endif
-}
+PyAPI_FUNC(int) _Py_ReachedRecursionLimit(PyThreadState *tstate);
 
 // Export for test_peg_generator
 PyAPI_FUNC(int) _Py_ReachedRecursionLimitWithMargin(

Lib/ensurepip/__init__.py

@@ -16,7 +16,8 @@
 # policies recommend against bundling dependencies. For example, Fedora
 # installs wheel packages in the /usr/share/python-wheels/ directory and don't
 # install the ensurepip._bundled package.
-if (_pkg_dir := sysconfig.get_config_var('WHEEL_PKG_DIR')) is not None:
+_pkg_dir = sysconfig.get_config_var('WHEEL_PKG_DIR')
+if _pkg_dir:
     _WHEEL_PKG_DIR = Path(_pkg_dir).resolve()
 else:
     _WHEEL_PKG_DIR = None

Lib/test/test_ensurepip.py

@@ -7,6 +7,7 @@
 import unittest
 import unittest.mock
 from pathlib import Path
+from test.support import import_helper
 
 import ensurepip
 import ensurepip._uninstall
@@ -36,6 +37,15 @@ def test_version_no_dir(self):
             # when the bundled pip wheel is used, we get _PIP_VERSION
             self.assertEqual(ensurepip._PIP_VERSION, ensurepip.version())
 
+    def test_wheel_pkg_dir_none(self):
+        # gh-146310: empty or None WHEEL_PKG_DIR should not search CWD
+        for value in ('', None):
+            with unittest.mock.patch('sysconfig.get_config_var',
+                                     return_value=value) as get_config_var:
+                module = import_helper.import_fresh_module('ensurepip')
+                self.assertIsNone(module._WHEEL_PKG_DIR)
+                get_config_var.assert_called_once_with('WHEEL_PKG_DIR')
+
     def test_selected_wheel_path_no_dir(self):
         pip_filename = f'pip-{ensurepip._PIP_VERSION}-py3-none-any.whl'
         with unittest.mock.patch.object(ensurepip, '_WHEEL_PKG_DIR', None):

Lib/test/test_lazy_import/__init__.py

@@ -484,8 +484,8 @@ def my_filter(name):
         sys.set_lazy_imports_filter(my_filter)
         self.assertIs(sys.get_lazy_imports_filter(), my_filter)
 
-    def test_lazy_modules_attribute_is_set(self):
-        """sys.lazy_modules should be a set per PEP 810."""
+    def test_lazy_modules_attribute_is_dict(self):
+        """sys.lazy_modules should be a dict per PEP 810."""
         self.assertIsInstance(sys.lazy_modules, dict)
 
     @support.requires_subprocess()
@@ -966,9 +966,24 @@ def test_module_added_to_lazy_modules_on_lazy_import(self):
 
     def test_lazy_modules_is_per_interpreter(self):
         """Each interpreter should have independent sys.lazy_modules."""
-        # Basic test that sys.lazy_modules exists and is a set
+        # Basic test that sys.lazy_modules exists and is a dict
         self.assertIsInstance(sys.lazy_modules, dict)
 
+    def test_lazy_module_without_children_is_tracked(self):
+        code = textwrap.dedent("""
+            import sys
+            lazy import json
+            assert "json" in sys.lazy_modules, (
+                f"expected 'json' in sys.lazy_modules, got {set(sys.lazy_modules)}"
+            )
+            assert sys.lazy_modules["json"] == set(), (
+                f"expected empty set for sys.lazy_modules['json'], "
+                f"got {sys.lazy_modules['json']!r}"
+            )
+            print("OK")
+        """)
+        assert_python_ok("-c", code)
+
 
 @support.requires_subprocess()
 class CommandLineAndEnvVarTests(unittest.TestCase):
@@ -1575,7 +1590,7 @@ def access_modules(idx):
         self.assertEqual(result.returncode, 0, f"stdout: {result.stdout}, stderr: {result.stderr}")
         self.assertIn("OK", result.stdout)
 
-    def test_concurrent_lazy_modules_set_updates(self):
+    def test_concurrent_lazy_modules_dict_updates(self):
         """Multiple threads creating lazy imports should safely update sys.lazy_modules."""
         code = textwrap.dedent("""
             import sys

Misc/NEWS.d/next/Core_and_Builtins/2026-03-17-14-20-56.gh-issue-145059.aB3xKm.rst

@@ -0,0 +1 @@
+Fixed ``sys.lazy_modules`` to include lazy modules without submodules. Patch by Bartosz Sławecki.

Misc/NEWS.d/next/Core_and_Builtins/2026-03-27-17-14-18.gh-issue-126910.hooVFQ.rst

@@ -0,0 +1 @@
+Set frame pointers in ``aarch64-unknown-linux-gnu`` JIT code, allowing most native profilers and debuggers to unwind through them. Patch by Diego Russo

Misc/NEWS.d/next/Library/2026-03-24-03-49-50.gh-issue-146310.WhlDir.rst

@@ -0,0 +1,2 @@
+The :mod:`ensurepip` module no longer looks for ``pip-*.whl`` wheel packages
+in the current directory.

Python/ceval.c

@@ -1201,6 +1201,19 @@ _PyEval_GetIter(_PyStackRef iterable, _PyStackRef *index_or_null, int yield_from
     return PyStackRef_FromPyObjectSteal(iter_o);
 }
 
+Py_NO_INLINE int
+_Py_ReachedRecursionLimit(PyThreadState *tstate)  {
+    uintptr_t here_addr = _Py_get_machine_stack_pointer();
+    _PyThreadStateImpl *_tstate = (_PyThreadStateImpl *)tstate;
+    assert(_tstate->c_stack_hard_limit != 0);
+#if _Py_STACK_GROWS_DOWN
+    return here_addr <= _tstate->c_stack_soft_limit;
+#else
+    return here_addr >= _tstate->c_stack_soft_limit;
+#endif
+}
+
+
 #if (defined(__GNUC__) && __GNUC__ >= 10 && !defined(__clang__)) && defined(__x86_64__)
 /*
  * gh-129987: The SLP autovectorizer can cause poor code generation for

Python/import.c

@@ -4377,6 +4377,12 @@ register_lazy_on_parent(PyThreadState *tstate, PyObject *name,
         Py_ssize_t dot = PyUnicode_FindChar(name, '.', 0,
                                             PyUnicode_GET_LENGTH(name), -1);
         if (dot < 0) {
+            PyObject *lazy_submodules = ensure_lazy_submodules(
+                (PyDictObject *)lazy_modules, name);
+            if (lazy_submodules == NULL) {
+                goto done;
+            }
+            Py_DECREF(lazy_submodules);
             ret = 0;
             goto done;
         }

Tools/jit/_optimizers.py

@@ -162,6 +162,7 @@ class Optimizer:
     label_prefix: str
     symbol_prefix: str
     re_global: re.Pattern[str]
+    frame_pointers: bool
     # The first block in the linked list:
     _root: _Block = dataclasses.field(init=False, default_factory=_Block)
     _labels: dict[str, _Block] = dataclasses.field(init=False, default_factory=dict)
@@ -193,6 +194,7 @@ class Optimizer:
     _re_small_const_1 = _RE_NEVER_MATCH
     _re_small_const_2 = _RE_NEVER_MATCH
     const_reloc = "<Not supported>"
+    _frame_pointer_modify: typing.ClassVar[re.Pattern[str]] = _RE_NEVER_MATCH
 
     def __post_init__(self) -> None:
         # Split the code into a linked list of basic blocks. A basic block is an
@@ -553,6 +555,16 @@ def _small_const_2(self, inst: Instruction) -> tuple[str, Instruction | None]:
     def _small_consts_match(self, inst1: Instruction, inst2: Instruction) -> bool:
         raise NotImplementedError()
 
+    def _validate(self) -> None:
+        for block in self._blocks():
+            if not block.instructions:
+                continue
+            for inst in block.instructions:
+                if self.frame_pointers:
+                    assert (
+                        self._frame_pointer_modify.match(inst.text) is None
+                    ), "Frame pointer should not be modified"
+
     def run(self) -> None:
         """Run this optimizer."""
         self._insert_continue_label()
@@ -565,6 +577,7 @@ def run(self) -> None:
             self._remove_unreachable()
         self._fixup_external_labels()
         self._fixup_constants()
+        self._validate()
         self.path.write_text(self._body())
 
 
@@ -595,6 +608,7 @@ class OptimizerAArch64(Optimizer):  # pylint: disable = too-few-public-methods
         r"\s*(?P<instruction>ldr)\s+.*(?P<value>_JIT_OP(ARG|ERAND(0|1))_(16|32)).*"
     )
     const_reloc = "CUSTOM_AARCH64_CONST"
+    _frame_pointer_modify = re.compile(r"\s*stp\s+x29.*")
 
     def _get_reg(self, inst: Instruction) -> str:
         _, rest = inst.text.split(inst.name)
@@ -649,4 +663,5 @@ class OptimizerX86(Optimizer):  # pylint: disable = too-few-public-methods
     # https://www.felixcloutier.com/x86/jmp
     _re_jump = re.compile(r"\s*jmp\s+(?P<target>[\w.]+)")
     # https://www.felixcloutier.com/x86/ret
-    _re_return = re.compile(r"\s*ret\b")
+    _re_return = re.compile(r"\s*retq?\b")
+    _frame_pointer_modify = re.compile(r"\s*movq?\s+%(\w+),\s+%rbp.*")

Tools/jit/_targets.py

@@ -176,21 +176,24 @@ async def _compile(
             f"{s}",
             f"{c}",
         ]
+        is_shim = opname == "shim"
         if self.frame_pointers:
-            frame_pointer = "all" if opname == "shim" else "reserved"
+            frame_pointer = "all" if is_shim else "reserved"
             args_s += ["-Xclang", f"-mframe-pointer={frame_pointer}"]
         args_s += self.args
         # Allow user-provided CFLAGS to override any defaults
         args_s += shlex.split(self.cflags)
         await _llvm.run(
             "clang", args_s, echo=self.verbose, llvm_version=self.llvm_version
         )
-        self.optimizer(
-            s,
-            label_prefix=self.label_prefix,
-            symbol_prefix=self.symbol_prefix,
-            re_global=self.re_global,
-        ).run()
+        if not is_shim:
+            self.optimizer(
+                s,
+                label_prefix=self.label_prefix,
+                symbol_prefix=self.symbol_prefix,
+                re_global=self.re_global,
+                frame_pointers=self.frame_pointers,
+            ).run()
         args_o = [f"--target={self.triple}", "-c", "-o", f"{o}", f"{s}"]
         await _llvm.run(
             "clang", args_o, echo=self.verbose, llvm_version=self.llvm_version
@@ -593,7 +596,9 @@ def get_target(host: str) -> _COFF32 | _COFF64 | _ELF | _MachO:
         # -mno-outline-atomics: Keep intrinsics from being emitted.
         args = ["-fpic", "-mno-outline-atomics", "-fno-plt"]
         optimizer = _optimizers.OptimizerAArch64
-        target = _ELF(host, condition, args=args, optimizer=optimizer)
+        target = _ELF(
+            host, condition, args=args, optimizer=optimizer, frame_pointers=True
+        )
     elif re.fullmatch(r"i686-pc-windows-msvc", host):
         host = "i686-pc-windows-msvc"
         condition = "defined(_M_IX86)"