Merge branch 'master' into sh_merge_master

.github/workflows/ci.yml

@@ -145,6 +145,24 @@ jobs:
       if: "!(runner.os == 'Windows' && (matrix.python == 3.8 || matrix.python == 3.9 || matrix.python == '3.10-dev'))"
       run: cmake --build build2 --target cpptest
 
+    # Third build - C++17 mode with unstable ABI
+    - name: Configure (unstable ABI)
+      run: >
+        cmake -S . -B build3
+        -DPYBIND11_WERROR=ON
+        -DDOWNLOAD_CATCH=ON
+        -DDOWNLOAD_EIGEN=ON
+        -DCMAKE_CXX_STANDARD=17
+        -DPYBIND11_INTERNALS_VERSION=10000000
+        "-DPYBIND11_TEST_OVERRIDE=test_call_policies.cpp;test_gil_scoped.cpp;test_thread.cpp"
+        ${{ matrix.args }}
+
+    - name: Build (unstable ABI)
+      run: cmake --build build3 -j 2
+
+    - name: Python tests (unstable ABI)
+      run: cmake --build build3 --target pytest
+
     - name: Interface test
       run: cmake --build build2 --target test_cmake_build
 

.pre-commit-config.yaml

@@ -32,7 +32,7 @@ repos:
     exclude: ^noxfile.py$
 
 - repo: https://github.com/asottile/pyupgrade
-  rev: v2.25.0
+  rev: v2.26.0
   hooks:
   - id: pyupgrade
 

CMakeLists.txt

@@ -89,6 +89,9 @@ endif()
 option(PYBIND11_INSTALL "Install pybind11 header files?" ${PYBIND11_MASTER_PROJECT})
 option(PYBIND11_TEST "Build pybind11 test suite?" ${PYBIND11_MASTER_PROJECT})
 option(PYBIND11_NOPYTHON "Disable search for Python" OFF)
+set(PYBIND11_INTERNALS_VERSION
+    ""
+    CACHE STRING "Override the ABI version, may be used to enable the unstable ABI.")
 
 cmake_dependent_option(
   USE_PYTHON_INCLUDE_DIR
@@ -189,6 +192,10 @@ if(NOT TARGET pybind11_headers)
 
   target_compile_features(pybind11_headers INTERFACE cxx_inheriting_constructors cxx_user_literals
                                                      cxx_right_angle_brackets)
+  if(NOT "${PYBIND11_INTERNALS_VERSION}" STREQUAL "")
+    target_compile_definitions(
+      pybind11_headers INTERFACE "PYBIND11_INTERNALS_VERSION=${PYBIND11_INTERNALS_VERSION}")
+  endif()
 else()
   # It is invalid to install a target twice, too.
   set(PYBIND11_INSTALL OFF)

docs/advanced/pycpp/numpy.rst

@@ -171,6 +171,31 @@ template parameter, and it ensures that non-conforming arguments are converted
 into an array satisfying the specified requirements instead of trying the next
 function overload.
 
+There are several methods on arrays; the methods listed below under references
+work, as well as the following functions based on the NumPy API:
+
+- ``.dtype()`` returns the type of the contained values.
+
+- ``.strides()`` returns a pointer to the strides of the array (optionally pass
+  an integer axis to get a number).
+
+- ``.flags()`` returns the flag settings. ``.writable()`` and ``.owndata()``
+  are directly available.
+
+- ``.offset_at()`` returns the offset (optionally pass indices).
+
+- ``.squeeze()`` returns a view with length-1 axes removed.
+
+- ``.view(dtype)`` returns a view of the array with a different dtype.
+
+- ``.reshape({i, j, ...})`` returns a view of the array with a different shape.
+  ``.resize({...})`` is also available.
+
+- ``.index_at(i, j, ...)`` gets the count from the beginning to a given index.
+
+
+There are also several methods for getting references (described below).
+
 Structured types
 ================
 
@@ -345,21 +370,21 @@ The returned proxy object supports some of the same methods as ``py::array`` so
 that it can be used as a drop-in replacement for some existing, index-checked
 uses of ``py::array``:
 
-- ``r.ndim()`` returns the number of dimensions
+- ``.ndim()`` returns the number of dimensions
 
-- ``r.data(1, 2, ...)`` and ``r.mutable_data(1, 2, ...)``` returns a pointer to
+- ``.data(1, 2, ...)`` and ``r.mutable_data(1, 2, ...)``` returns a pointer to
   the ``const T`` or ``T`` data, respectively, at the given indices.  The
   latter is only available to proxies obtained via ``a.mutable_unchecked()``.
 
-- ``itemsize()`` returns the size of an item in bytes, i.e. ``sizeof(T)``.
+- ``.itemsize()`` returns the size of an item in bytes, i.e. ``sizeof(T)``.
 
-- ``ndim()`` returns the number of dimensions.
+- ``.ndim()`` returns the number of dimensions.
 
-- ``shape(n)`` returns the size of dimension ``n``
+- ``.shape(n)`` returns the size of dimension ``n``
 
-- ``size()`` returns the total number of elements (i.e. the product of the shapes).
+- ``.size()`` returns the total number of elements (i.e. the product of the shapes).
 
-- ``nbytes()`` returns the number of bytes used by the referenced elements
+- ``.nbytes()`` returns the number of bytes used by the referenced elements
   (i.e. ``itemsize()`` times ``size()``).
 
 .. seealso::

docs/changelog.rst

@@ -9,6 +9,11 @@ Starting with version 1.8.0, pybind11 releases use a `semantic versioning
 v2.8.0 (WIP)
 ------------
 
+New features:
+
+* Added ``py::raise_from`` to enable chaining exceptions.
+  `#3215 <https://github.com/pybind/pybind11/pull/3215>`_
+
 * Allow exception translators to be optionally registered local to a module
   instead of applying globally across all pybind11 modules. Use
   ``register_local_exception_translator(ExceptionTranslator&& translator)``
@@ -16,6 +21,115 @@ v2.8.0 (WIP)
   translator)`` to keep your exception remapping code local to the module.
   `#2650 <https://github.com/pybind/pybind11/pull/2650>`_
 
+* Add ``make_simple_namespace`` function for instantiating Python
+  ``SimpleNamespace`` objects.
+  `#2840 <https://github.com/pybind/pybind11/pull/2840>`_
+
+* ``pybind11::scoped_interpreter`` and ``initialize_interpreter`` have new
+  arguments to allow ``sys.argv`` initialization.
+  `#2341 <https://github.com/pybind/pybind11/pull/2341>`_
+
+* Allow Python builtins to be used as callbacks in CPython.
+  `#1413 <https://github.com/pybind/pybind11/pull/1413>`_
+
+* Added ``view`` to view arrays with a different datatype.
+  `#987 <https://github.com/pybind/pybind11/pull/987>`_
+
+* Implemented ``reshape`` on arrays.
+  `#984 <https://github.com/pybind/pybind11/pull/984>`_
+
+
+Changes:
+
+* Set ``__file__`` constant when running ``eval_file`` in an embedded interpreter.
+  `#3233 <https://github.com/pybind/pybind11/pull/3233>`_
+
+* The pybind11 proxy types ``str``, ``bytes``, ``bytearray``, ``tuple``,
+  ``list`` now consistently support passing ``ssize_t`` values for sizes and
+  indexes. Previously, only ``size_t`` was accepted in several interfaces.
+  `#3219 <https://github.com/pybind/pybind11/pull/3219>`_
+
+
+Fixes:
+
+* Bug fix: enum value's ``__int__`` returning non-int when underlying type is bool or of char type.
+  `#1334 <https://github.com/pybind/pybind11/pull/1334>`_
+
+* Fixes bug in setting error state in Capsule's pointer methods.
+  `#3261 <https://github.com/pybind/pybind11/pull/3261>`_
+
+* A long-standing memory leak in ``py::cpp_function::initialize`` was fixed.
+  `#3229 <https://github.com/pybind/pybind11/pull/3229>`_
+
+* Fixes thread safety for some ``pybind11::type_caster`` which require lifetime extension, such as for ``std::string_view``.
+  `#3237 <https://github.com/pybind/pybind11/pull/3237>`_
+
+* Restore compatibility with gcc 4.8.4 as distributed by ubuntu-trusty, linuxmint-17.
+  `#3270 <https://github.com/pybind/pybind11/pull/3270>`_
+
+
+Build system improvements:
+
+* Fix regression in CMake Python package config: improper use of absolute path.
+  `#3144 <https://github.com/pybind/pybind11/pull/3144>`_
+
+* Specified UTF8-encoding in setup.py calls of open().
+  `#3137 <https://github.com/pybind/pybind11/pull/3137>`_
+
+
+Backend and tidying up:
+
+* Optimize NumPy array construction with additional moves.
+  `#3183 <https://github.com/pybind/pybind11/pull/3183>`_
+
+* Conversion to ``std::string`` and ``std::string_view`` now avoids making an
+  extra copy of the data on Python >= 3.3.
+  `#3257 <https://github.com/pybind/pybind11/pull/3257>`_
+
+* Remove const modifier from certain C++ methods on Python collections
+  (``list``, ``set``, ``dict``) such as (``clear()``, ``append()``,
+  ``insert()``, etc...) and annotated them with ``py-non-const``.
+
+* Enable readability ``clang-tidy-const-return`` and remove useless consts.
+  `#3254 <https://github.com/pybind/pybind11/pull/3254>`_
+  `#3194 <https://github.com/pybind/pybind11/pull/3194>`_
+
+* The clang-tidy ``google-explicit-constructor`` option was enabled.
+  `#3250 <https://github.com/pybind/pybind11/pull/3250>`_
+
+* Mark a pytype move constructor as noexcept (perf).
+  `#3236 <https://github.com/pybind/pybind11/pull/3236>`_
+
+* Enable clang-tidy check to guard against inheritance slicing.
+  `#3210 <https://github.com/pybind/pybind11/pull/3210>`_
+
+* Legacy warning suppression pragma were removed from eigen.h. On Unix
+  platforms, please use -isystem for Eigen include directories, to suppress
+  compiler warnings originating from Eigen headers. Note that CMake does this
+  by default. No adjustments are needed for Windows.
+  `#3198 <https://github.com/pybind/pybind11/pull/3198>`_
+
+* Format pybind11 with isort consistent ordering of imports
+  `#3195 <https://github.com/pybind/pybind11/pull/3195>`_
+
+* The warnings-suppression "pragma clamp" at the top/bottom of pybind11 was
+  removed, clearing the path to refactoring and IWYU cleanup.
+  `#3186 <https://github.com/pybind/pybind11/pull/3186>`_
+
+* Enable most bugprone checks in clang-tidy and fix the found potential bugs
+  and poor coding styles.
+  `#3166 <https://github.com/pybind/pybind11/pull/3166>`_
+
+* Add ``clang-tidy-readability`` rules to make boolean casts explicit improving
+  code readability. Also enabled other misc and readability clang-tidy checks.
+  `#3148 <https://github.com/pybind/pybind11/pull/3148>`_
+
+* Move object in ``.pop()`` for list.
+  `#3116 <https://github.com/pybind/pybind11/pull/3116>`_
+
+
+
+
 v2.7.1 (Aug 3, 2021)
 ---------------------
 

include/pybind11/detail/internals.h

@@ -12,6 +12,24 @@
 #include "../pytypes.h"
 #include "smart_holder_sfinae_hooks_only.h"
 
+/// Tracks the `internals` and `type_info` ABI version independent of the main library version.
+///
+/// Some portions of the code use an ABI that is conditional depending on this
+/// version number.  That allows ABI-breaking changes to be "pre-implemented".
+/// Once the default version number is incremented, the conditional logic that
+/// no longer applies can be removed.  Additionally, users that need not
+/// maintain ABI compatibility can increase the version number in order to take
+/// advantage of any functionality/efficiency improvements that depend on the
+/// newer ABI.
+///
+/// WARNING: If you choose to manually increase the ABI version, note that
+/// pybind11 may not be tested as thoroughly with a non-default ABI version, and
+/// further ABI-incompatible changes may be made before the ABI is officially
+/// changed to the new version.
+#ifndef PYBIND11_INTERNALS_VERSION
+#    define PYBIND11_INTERNALS_VERSION 4
+#endif
+
 PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 
 using ExceptionTranslator = void (*)(std::exception_ptr);
@@ -26,30 +44,58 @@ inline PyObject *make_object_base_type(PyTypeObject *metaclass);
 // The old Python Thread Local Storage (TLS) API is deprecated in Python 3.7 in favor of the new
 // Thread Specific Storage (TSS) API.
 #if PY_VERSION_HEX >= 0x03070000
-#    define PYBIND11_TLS_KEY_INIT(var) Py_tss_t *var = nullptr
+// Avoid unnecessary allocation of `Py_tss_t`, since we cannot use
-#    define PYBIND11_TLS_GET_VALUE(key) PyThread_tss_get((key))
+// `Py_LIMITED_API` anyway.
-#    define PYBIND11_TLS_REPLACE_VALUE(key, value) PyThread_tss_set((key), (value))
+#    if PYBIND11_INTERNALS_VERSION > 4
-#    define PYBIND11_TLS_DELETE_VALUE(key) PyThread_tss_set((key), nullptr)
+#        define PYBIND11_TLS_KEY_REF Py_tss_t &
-#    define PYBIND11_TLS_FREE(key) PyThread_tss_free(key)
+#        ifdef __GNUC__
-#else
+// Clang on macOS warns due to `Py_tss_NEEDS_INIT` not specifying an initializer
-    // Usually an int but a long on Cygwin64 with Python 3.x
+// for every field.
-#    define PYBIND11_TLS_KEY_INIT(var) decltype(PyThread_create_key()) var = 0
+#            define PYBIND11_TLS_KEY_INIT(var)                                                    \
-#    define PYBIND11_TLS_GET_VALUE(key) PyThread_get_key_value((key))
+                _Pragma("GCC diagnostic push")                                         /**/       \
-#    if PY_MAJOR_VERSION < 3
+                    _Pragma("GCC diagnostic ignored \"-Wmissing-field-initializers\"") /**/       \
-#        define PYBIND11_TLS_DELETE_VALUE(key)                               \
+                    Py_tss_t var                                                                  \
-             PyThread_delete_key_value(key)
+                    = Py_tss_NEEDS_INIT;                                                          \
-#        define PYBIND11_TLS_REPLACE_VALUE(key, value)                       \
+                _Pragma("GCC diagnostic pop")
-             do {                                                            \
+#        else
-                 PyThread_delete_key_value((key));                           \
+#            define PYBIND11_TLS_KEY_INIT(var) Py_tss_t var = Py_tss_NEEDS_INIT;
-                 PyThread_set_key_value((key), (value));                     \
+#        endif
-             } while (false)
+#        define PYBIND11_TLS_KEY_CREATE(var) (PyThread_tss_create(&(var)) == 0)
+#        define PYBIND11_TLS_GET_VALUE(key) PyThread_tss_get(&(key))
+#        define PYBIND11_TLS_REPLACE_VALUE(key, value) PyThread_tss_set(&(key), (value))
+#        define PYBIND11_TLS_DELETE_VALUE(key) PyThread_tss_set(&(key), nullptr)
+#        define PYBIND11_TLS_FREE(key) PyThread_tss_delete(&(key))
 #    else
-#        define PYBIND11_TLS_DELETE_VALUE(key)                               \
+#        define PYBIND11_TLS_KEY_REF Py_tss_t *
-             PyThread_set_key_value((key), nullptr)
+#        define PYBIND11_TLS_KEY_INIT(var) Py_tss_t *var = nullptr;
-#        define PYBIND11_TLS_REPLACE_VALUE(key, value)                       \
+#        define PYBIND11_TLS_KEY_CREATE(var)                                                      \
-             PyThread_set_key_value((key), (value))
+            (((var) = PyThread_tss_alloc()) != nullptr && (PyThread_tss_create((var)) == 0))
+#        define PYBIND11_TLS_GET_VALUE(key) PyThread_tss_get((key))
+#        define PYBIND11_TLS_REPLACE_VALUE(key, value) PyThread_tss_set((key), (value))
+#        define PYBIND11_TLS_DELETE_VALUE(key) PyThread_tss_set((key), nullptr)
+#        define PYBIND11_TLS_FREE(key) PyThread_tss_free(key)
 #    endif
-#    define PYBIND11_TLS_FREE(key) (void)key
+#else
+// Usually an int but a long on Cygwin64 with Python 3.x
+#    define PYBIND11_TLS_KEY_REF decltype(PyThread_create_key())
+#    define PYBIND11_TLS_KEY_INIT(var) PYBIND11_TLS_KEY_REF var = 0;
+#    define PYBIND11_TLS_KEY_CREATE(var) (((var) = PyThread_create_key()) != -1)
+#    define PYBIND11_TLS_GET_VALUE(key) PyThread_get_key_value((key))
+#    if PY_MAJOR_VERSION < 3 || defined(PYPY_VERSION)
+// On CPython < 3.4 and on PyPy, `PyThread_set_key_value` strangely does not set
+// the value if it has already been set.  Instead, it must first be deleted and
+// then set again.
+#        define PYBIND11_TLS_DELETE_VALUE(key) PyThread_delete_key_value(key)
+#        define PYBIND11_TLS_REPLACE_VALUE(key, value)                                            \
+            do {                                                                                  \
+                PyThread_delete_key_value((key));                                                 \
+                PyThread_set_key_value((key), (value));                                           \
+            } while (false)
+#    else
+#        define PYBIND11_TLS_DELETE_VALUE(key) PyThread_set_key_value((key), nullptr)
+#        define PYBIND11_TLS_REPLACE_VALUE(key, value) PyThread_set_key_value((key), (value))
+#    endif
+#    define PYBIND11_TLS_FREE(key) (void) key
 #endif
 
 // Python loads modules by default with dlopen with the RTLD_LOCAL flag; under libc++ and possibly
@@ -107,22 +153,31 @@ struct internals {
     std::unordered_map<const PyObject *, std::vector<PyObject *>> patients;
     std::forward_list<ExceptionTranslator> registered_exception_translators;
     std::unordered_map<std::string, void *> shared_data; // Custom data to be shared across extensions
+#if PYBIND11_INTERNALS_VERSION == 4
     std::vector<PyObject *> unused_loader_patient_stack_remove_at_v5;
+#endif
     std::forward_list<std::string> static_strings; // Stores the std::strings backing detail::c_str()
     PyTypeObject *static_property_type;
     PyTypeObject *default_metaclass;
     PyObject *instance_base;
 #if defined(WITH_THREAD)
-    PYBIND11_TLS_KEY_INIT(tstate);
+    PYBIND11_TLS_KEY_INIT(tstate)
+#    if PYBIND11_INTERNALS_VERSION > 4
+    PYBIND11_TLS_KEY_INIT(loader_life_support_tls_key)
+#    endif // PYBIND11_INTERNALS_VERSION > 4
     PyInterpreterState *istate = nullptr;
     ~internals() {
+#    if PYBIND11_INTERNALS_VERSION > 4
+        PYBIND11_TLS_FREE(loader_life_support_tls_key);
+#    endif // PYBIND11_INTERNALS_VERSION > 4
+
         // This destructor is called *after* Py_Finalize() in finalize_interpreter().
-        // That *SHOULD BE* fine. The following details what happens when PyThread_tss_free is called.
+        // That *SHOULD BE* fine. The following details what happens when PyThread_tss_free is
-        // PYBIND11_TLS_FREE is PyThread_tss_free on python 3.7+. On older python, it does nothing.
+        // called. PYBIND11_TLS_FREE is PyThread_tss_free on python 3.7+. On older python, it does
-        // PyThread_tss_free calls PyThread_tss_delete and PyMem_RawFree.
+        // nothing. PyThread_tss_free calls PyThread_tss_delete and PyMem_RawFree.
-        // PyThread_tss_delete just calls TlsFree (on Windows) or pthread_key_delete (on *NIX). Neither
+        // PyThread_tss_delete just calls TlsFree (on Windows) or pthread_key_delete (on *NIX).
-        // of those have anything to do with CPython internals.
+        // Neither of those have anything to do with CPython internals. PyMem_RawFree *requires*
-        // PyMem_RawFree *requires* that the `tstate` be allocated with the CPython allocator.
+        // that the `tstate` be allocated with the CPython allocator.
         PYBIND11_TLS_FREE(tstate);
     }
 #endif
@@ -154,9 +209,6 @@ struct type_info {
     bool module_local : 1;
 };
 
-/// Tracks the `internals` and `type_info` ABI version independent of the main library version
-#define PYBIND11_INTERNALS_VERSION 4
-
 /// On MSVC, debug and release builds are not ABI-compatible!
 #if defined(_MSC_VER) && defined(_DEBUG)
 #  define PYBIND11_BUILD_TYPE "_debug"
@@ -302,21 +354,21 @@ PYBIND11_NOINLINE internals &get_internals() {
         internals_ptr = new internals();
 #if defined(WITH_THREAD)
 
-        #if PY_VERSION_HEX < 0x03090000
+#    if PY_VERSION_HEX < 0x03090000
-                PyEval_InitThreads();
+        PyEval_InitThreads();
-        #endif
+#    endif
         PyThreadState *tstate = PyThreadState_Get();
-        #if PY_VERSION_HEX >= 0x03070000
+        if (!PYBIND11_TLS_KEY_CREATE(internals_ptr->tstate)) {
-            internals_ptr->tstate = PyThread_tss_alloc();
+            pybind11_fail("get_internals: could not successfully initialize the tstate TSS key!");
-            if (!internals_ptr->tstate || (PyThread_tss_create(internals_ptr->tstate) != 0))
+        }
-                pybind11_fail("get_internals: could not successfully initialize the tstate TSS key!");
+        PYBIND11_TLS_REPLACE_VALUE(internals_ptr->tstate, tstate);
-            PyThread_tss_set(internals_ptr->tstate, tstate);
+
-        #else
+#    if PYBIND11_INTERNALS_VERSION > 4
-            internals_ptr->tstate = PyThread_create_key();
+        if (!PYBIND11_TLS_KEY_CREATE(internals_ptr->loader_life_support_tls_key)) {
-            if (internals_ptr->tstate == -1)
+            pybind11_fail("get_internals: could not successfully initialize the "
-                pybind11_fail("get_internals: could not successfully initialize the tstate TLS key!");
+                          "loader_life_support TSS key!");
-            PyThread_set_key_value(internals_ptr->tstate, tstate);
+        }
-        #endif
+#    endif
         internals_ptr->istate = tstate->interp;
 #endif
         builtins[id] = capsule(internals_pp);
@@ -328,7 +380,6 @@ PYBIND11_NOINLINE internals &get_internals() {
     return **internals_pp;
 }
 
-
 // the internals struct (above) is shared between all the modules. local_internals are only
 // for a single module. Any changes made to internals may require an update to
 // PYBIND11_INTERNALS_VERSION, breaking backwards compatibility. local_internals is, by design,
@@ -336,8 +387,41 @@ PYBIND11_NOINLINE internals &get_internals() {
 // impact any other modules, because the only things accessing the local internals is the
 // module that contains them.
 struct local_internals {
-  type_map<type_info *> registered_types_cpp;
+    type_map<type_info *> registered_types_cpp;
-  std::forward_list<ExceptionTranslator> registered_exception_translators;
+    std::forward_list<ExceptionTranslator> registered_exception_translators;
+#if defined(WITH_THREAD) && PYBIND11_INTERNALS_VERSION == 4
+
+    // For ABI compatibility, we can't store the loader_life_support TLS key in
+    // the `internals` struct directly.  Instead, we store it in `shared_data` and
+    // cache a copy in `local_internals`.  If we allocated a separate TLS key for
+    // each instance of `local_internals`, we could end up allocating hundreds of
+    // TLS keys if hundreds of different pybind11 modules are loaded (which is a
+    // plausible number).
+    PYBIND11_TLS_KEY_INIT(loader_life_support_tls_key)
+
+    // Holds the shared TLS key for the loader_life_support stack.
+    struct shared_loader_life_support_data {
+        PYBIND11_TLS_KEY_INIT(loader_life_support_tls_key)
+        shared_loader_life_support_data() {
+            if (!PYBIND11_TLS_KEY_CREATE(loader_life_support_tls_key)) {
+                pybind11_fail("local_internals: could not successfully initialize the "
+                              "loader_life_support TLS key!");
+            }
+        }
+        // We can't help but leak the TLS key, because Python never unloads extension modules.
+    };
+
+    local_internals() {
+        auto &internals = get_internals();
+        // Get or create the `loader_life_support_stack_key`.
+        auto &ptr = internals.shared_data["_life_support"];
+        if (!ptr) {
+            ptr = new shared_loader_life_support_data;
+        }
+        loader_life_support_tls_key
+            = static_cast<shared_loader_life_support_data *>(ptr)->loader_life_support_tls_key;
+    }
+#endif //  defined(WITH_THREAD) && PYBIND11_INTERNALS_VERSION == 4
 };
 
 /// Works like `get_internals`, but for things which are locally registered.

include/pybind11/detail/type_caster_base.h

@@ -35,30 +35,43 @@ private:
     loader_life_support* parent = nullptr;
     std::unordered_set<PyObject *> keep_alive;
 
-    static loader_life_support** get_stack_pp() {
 #if defined(WITH_THREAD)
-        thread_local static loader_life_support* per_thread_stack = nullptr;
+    // Store stack pointer in thread-local storage.
-        return &per_thread_stack;
+    static PYBIND11_TLS_KEY_REF get_stack_tls_key() {
-#else
+#    if PYBIND11_INTERNALS_VERSION == 4
-        static loader_life_support* global_stack = nullptr;
+        return get_local_internals().loader_life_support_tls_key;
-        return &global_stack;
+#    else
-#endif
+        return get_internals().loader_life_support_tls_key;
+#    endif
     }
+    static loader_life_support *get_stack_top() {
+        return static_cast<loader_life_support *>(PYBIND11_TLS_GET_VALUE(get_stack_tls_key()));
+    }
+    static void set_stack_top(loader_life_support *value) {
+        PYBIND11_TLS_REPLACE_VALUE(get_stack_tls_key(), value);
+    }
+#else
+    // Use single global variable for stack.
+    static loader_life_support **get_stack_pp() {
+        static loader_life_support *global_stack = nullptr;
+        return global_stack;
+    }
+    static loader_life_support *get_stack_top() { return *get_stack_pp(); }
+    static void set_stack_top(loader_life_support *value) { *get_stack_pp() = value; }
+#endif
 
 public:
     /// A new patient frame is created when a function is entered
     loader_life_support() {
-        loader_life_support** stack = get_stack_pp();
+        parent = get_stack_top();
-        parent = *stack;
+        set_stack_top(this);
-        *stack = this;
     }
 
     /// ... and destroyed after it returns
     ~loader_life_support() {
-        loader_life_support** stack = get_stack_pp();
+        if (get_stack_top() != this)
-        if (*stack != this)
             pybind11_fail("loader_life_support: internal error");
-        *stack = parent;
+        set_stack_top(parent);
         for (auto* item : keep_alive)
             Py_DECREF(item);
     }
@@ -66,7 +79,7 @@ public:
     /// This can only be used inside a pybind11-bound function, either by `argument_loader`
     /// at argument preparation time or by `py::cast()` at execution time.
     PYBIND11_NOINLINE static void add_patient(handle h) {
-        loader_life_support* frame =  *get_stack_pp();
+        loader_life_support *frame = get_stack_top();
         if (!frame) {
             // NOTE: It would be nice to include the stack frames here, as this indicates
             // use of pybind11::cast<> outside the normal call framework, finding such

include/pybind11/gil.h

@@ -50,7 +50,7 @@ PYBIND11_NAMESPACE_END(detail)
 class gil_scoped_acquire {
 public:
     PYBIND11_NOINLINE gil_scoped_acquire() {
-        auto const &internals = detail::get_internals();
+        auto &internals = detail::get_internals();
         tstate = (PyThreadState *) PYBIND11_TLS_GET_VALUE(internals.tstate);
 
         if (!tstate) {
@@ -132,7 +132,7 @@ public:
         // `get_internals()` must be called here unconditionally in order to initialize
         // `internals.tstate` for subsequent `gil_scoped_acquire` calls. Otherwise, an
         // initialization race could occur as multiple threads try `gil_scoped_acquire`.
-        const auto &internals = detail::get_internals();
+        auto &internals = detail::get_internals();
         tstate = PyEval_SaveThread();
         if (disassoc) {
             auto key = internals.tstate;

include/pybind11/numpy.h

@@ -322,7 +322,7 @@ template <typename T> using remove_all_extents_t = typename array_info<T>::type;
 
 template <typename T> using is_pod_struct = all_of<
     std::is_standard_layout<T>,     // since we're accessing directly in memory we need a standard layout type
-#if defined(__GLIBCXX__) && (__GLIBCXX__ < 20150422 || __GLIBCXX__ == 20150623 || __GLIBCXX__ == 20150626 || __GLIBCXX__ == 20160803)
+#if defined(__GLIBCXX__) && (__GLIBCXX__ < 20150422 || __GLIBCXX__ == 20150426 || __GLIBCXX__ == 20150623 || __GLIBCXX__ == 20150626 || __GLIBCXX__ == 20160803)
     // libstdc++ < 5 (including versions 4.8.5, 4.9.3 and 4.9.4 which were released after 5)
     // don't implement is_trivially_copyable, so approximate it
     std::is_trivially_destructible<T>,

tests/test_pytypes.py

@@ -467,7 +467,8 @@ def test_issue2361():
     assert m.issue2361_str_implicit_copy_none() == "None"
     with pytest.raises(TypeError) as excinfo:
         assert m.issue2361_dict_implicit_copy_none()
-    assert "'NoneType' object is not iterable" in str(excinfo.value)
+    assert "NoneType" in str(excinfo.value)
+    assert "iterable" in str(excinfo.value)
 
 
 @pytest.mark.parametrize(