Merge branch 'master' into stable

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
 ================
 
@@ -233,8 +258,8 @@ by the compiler. The result is returned as a NumPy array of type
 
 .. code-block:: pycon
 
-    >>> x = np.array([[1, 3],[5, 7]])
-    >>> y = np.array([[2, 4],[6, 8]])
+    >>> x = np.array([[1, 3], [5, 7]])
+    >>> y = np.array([[2, 4], [6, 8]])
     >>> z = 3
     >>> result = vectorized_func(x, y, z)
 
@@ -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::
@@ -378,7 +403,7 @@ In Python 2, the syntactic sugar ``...`` is not available, but the singleton
 
 .. code-block:: python
 
-   a = # a NumPy array
+   a = ...  # a NumPy array
    b = a[0, ..., 0]
 
 The function ``py::ellipsis()`` function can be used to perform the same

docs/advanced/pycpp/object.rst

@@ -20,6 +20,47 @@ Available types include :class:`handle`, :class:`object`, :class:`bool_`,
     Be sure to review the :ref:`pytypes_gotchas` before using this heavily in
     your C++ API.
 
+.. _instantiating_compound_types:
+
+Instantiating compound Python types from C++
+============================================
+
+Dictionaries can be initialized in the :class:`dict` constructor:
+
+.. code-block:: cpp
+
+    using namespace pybind11::literals; // to bring in the `_a` literal
+    py::dict d("spam"_a=py::none(), "eggs"_a=42);
+
+A tuple of python objects can be instantiated using :func:`py::make_tuple`:
+
+.. code-block:: cpp
+
+    py::tuple tup = py::make_tuple(42, py::none(), "spam");
+
+Each element is converted to a supported Python type.
+
+A `simple namespace`_ can be instantiated using
+:func:`py::make_simple_namespace`:
+
+.. code-block:: cpp
+
+    using namespace pybind11::literals; // to bring in the `_a` literal
+    py::object ns = py::make_simple_namespace("spam"_a=py::none(), "eggs"_a=42);
+
+Attributes on a namespace can be modified with the :func:`py::delattr`,
+:func:`py::getattr`, and :func:`py::setattr` functions. Simple namespaces can
+be useful as lightweight stand-ins for class instances.
+
+.. note::
+
+    ``make_simple_namespace`` is not available in Python 2.
+
+.. versionchanged:: 2.8
+    ``make_simple_namespace`` added.
+
+.. _simple namespace: https://docs.python.org/3/library/types.html#types.SimpleNamespace
+
 .. _casting_back_and_forth:
 
 Casting back and forth
@@ -30,7 +71,7 @@ types to Python, which can be done using :func:`py::cast`:
 
 .. code-block:: cpp
 
-    MyClass *cls = ..;
+    MyClass *cls = ...;
     py::object obj = py::cast(cls);
 
 The reverse direction uses the following syntax:
@@ -132,6 +173,7 @@ Keyword arguments are also supported. In Python, there is the usual call syntax:
     def f(number, say, to):
         ...  # function code
 
+
     f(1234, say="hello", to=some_instance)  # keyword call in Python
 
 In C++, the same call can be made using:

docs/advanced/pycpp/utilities.rst

@@ -66,7 +66,7 @@ extra type, `py::scoped_estream_redirect <scoped_estream_redirect>`, is identica
 except for defaulting to ``std::cerr`` and ``sys.stderr``; this can be useful with
 `py::call_guard`, which allows multiple items, but uses the default constructor:
 
-.. code-block:: py
+.. code-block:: cpp
 
     // Alternative: Call single function using call guard
     m.def("noisy_func", &call_noisy_function,

docs/advanced/smart_ptrs.rst

@@ -77,6 +77,7 @@ segmentation fault).
 .. code-block:: python
 
    from example import Parent
+
    print(Parent().get_child())
 
 The problem is that ``Parent::get_child()`` returns a pointer to an instance of

docs/benchmark.py

 # -*- coding: utf-8 -*-
-import random
-import os
-import time
 import datetime as dt
+import os
+import random
 
 nfns = 4  # Functions per class
 nargs = 4  # Arguments per function

docs/changelog.rst

@@ -6,8 +6,13 @@ Changelog
 Starting with version 1.8.0, pybind11 releases use a `semantic versioning
 <http://semver.org>`_ policy.
 
-v2.8.0 (WIP)
-------------
+v2.8.0 (Oct 4, 2021)
+--------------------
+
+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
@@ -16,6 +21,158 @@ 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>`_
+
+* Enable defining custom ``__new__`` methods on classes by fixing bug
+  preventing overriding methods if they have non-pybind11 siblings.
+  `#3265 <https://github.com/pybind/pybind11/pull/3265>`_
+
+* Add ``make_value_iterator()``, and fix ``make_key_iterator()`` to return
+  references instead of copies.
+  `#3293 <https://github.com/pybind/pybind11/pull/3293>`_
+
+* Improve the classes generated by ``bind_map``: `#3310 <https://github.com/pybind/pybind11/pull/3310>`_
+    * Change ``.items`` from an iterator to a dictionary view.
+    * Add ``.keys`` and ``.values`` (both dictionary views).
+    * Allow ``__contains__`` to take any object.
+
+* ``pybind11::custom_type_setup`` was added, for customizing the
+  ``PyHeapTypeObject`` corresponding to a class, which may be useful for
+  enabling garbage collection support, among other things.
+  `#3287 <https://github.com/pybind/pybind11/pull/3287>`_
+
+
+Changes:
+
+* Set ``__file__`` constant when running ``eval_file`` in an embedded interpreter.
+  `#3233 <https://github.com/pybind/pybind11/pull/3233>`_
+
+* Python objects and (C++17) ``std::optional`` now accepted in ``py::slice``
+  constructor.
+  `#1101 <https://github.com/pybind/pybind11/pull/1101>`_
+
+* 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>`_
+
+* Avoid evaluating ``PYBIND11_TLS_REPLACE_VALUE`` arguments more than once.
+  `#3290 <https://github.com/pybind/pybind11/pull/3290>`_
+
+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>`_
+
+* Cached Python version information could become stale when CMake was re-run
+  with a different Python version. The build system now detects this and
+  updates this information.
+  `#3299 <https://github.com/pybind/pybind11/pull/3299>`_
+
+* Specified UTF8-encoding in setup.py calls of open().
+  `#3137 <https://github.com/pybind/pybind11/pull/3137>`_
+
+* Fix a harmless warning from CMake 3.21 with the classic Python discovery.
+  `#3220 <https://github.com/pybind/pybind11/pull/3220>`_
+
+* Eigen repo and version can now be specified as cmake options.
+  `#3324 <https://github.com/pybind/pybind11/pull/3324>`_
+
+
+Backend and tidying up:
+
+* Reduced thread-local storage required for keeping alive temporary data for
+  type conversion to one key per ABI version, rather than one key per extension
+  module.  This makes the total thread-local storage required by pybind11 2
+  keys per ABI version.
+  `#3275 <https://github.com/pybind/pybind11/pull/3275>`_
+
+* 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)
 ---------------------
 
@@ -1028,6 +1185,7 @@ v2.2.0 (August 31, 2017)
 
       from cpp_module import CppBase1, CppBase2
 
+
       class PyDerived(CppBase1, CppBase2):
           def __init__(self):
               CppBase1.__init__(self)  # C++ bases must be initialized explicitly

docs/classes.rst

@@ -44,12 +44,12 @@ interactive Python session demonstrating this example is shown below:
 
     % python
     >>> import example
-    >>> p = example.Pet('Molly')
+    >>> p = example.Pet("Molly")
     >>> print(p)
     <example.Pet object at 0x10cd98060>
     >>> p.getName()
     u'Molly'
-    >>> p.setName('Charly')
+    >>> p.setName("Charly")
     >>> p.getName()
     u'Charly'
 
@@ -122,10 +122,10 @@ This makes it possible to write
 
 .. code-block:: pycon
 
-    >>> p = example.Pet('Molly')
+    >>> p = example.Pet("Molly")
     >>> p.name
     u'Molly'
-    >>> p.name = 'Charly'
+    >>> p.name = "Charly"
     >>> p.name
     u'Charly'
 
@@ -174,10 +174,10 @@ Native Python classes can pick up new attributes dynamically:
 .. code-block:: pycon
 
     >>> class Pet:
-    ...     name = 'Molly'
+    ...     name = "Molly"
     ...
     >>> p = Pet()
-    >>> p.name = 'Charly'  # overwrite existing
+    >>> p.name = "Charly"  # overwrite existing
     >>> p.age = 2  # dynamically add a new attribute
 
 By default, classes exported from C++ do not support this and the only writable
@@ -195,7 +195,7 @@ Trying to set any other attribute results in an error:
 .. code-block:: pycon
 
     >>> p = example.Pet()
-    >>> p.name = 'Charly'  # OK, attribute defined in C++
+    >>> p.name = "Charly"  # OK, attribute defined in C++
     >>> p.age = 2  # fail
     AttributeError: 'Pet' object has no attribute 'age'
 
@@ -213,7 +213,7 @@ Now everything works as expected:
 .. code-block:: pycon
 
     >>> p = example.Pet()
-    >>> p.name = 'Charly'  # OK, overwrite value in C++
+    >>> p.name = "Charly"  # OK, overwrite value in C++
     >>> p.age = 2  # OK, dynamically add a new attribute
     >>> p.__dict__  # just like a native Python class
     {'age': 2}
@@ -280,7 +280,7 @@ expose fields and methods of both types:
 
 .. code-block:: pycon
 
-    >>> p = example.Dog('Molly')
+    >>> p = example.Dog("Molly")
     >>> p.name
     u'Molly'
     >>> p.bark()
@@ -446,8 +446,7 @@ you can use ``py::detail::overload_cast_impl`` with an additional set of parenth
 Enumerations and internal types
 ===============================
 
-Let's now suppose that the example class contains an internal enumeration type,
-e.g.:
+Let's now suppose that the example class contains internal types like enumerations, e.g.:
 
 .. code-block:: cpp
 
@@ -457,10 +456,15 @@ e.g.:
             Cat
         };
 
+        struct Attributes {
+            float age = 0;
+        };
+
         Pet(const std::string &name, Kind type) : name(name), type(type) { }
 
         std::string name;
         Kind type;
+        Attributes attr;
     };
 
 The binding code for this example looks as follows:
@@ -471,22 +475,28 @@ The binding code for this example looks as follows:
 
     pet.def(py::init<const std::string &, Pet::Kind>())
         .def_readwrite("name", &Pet::name)
-        .def_readwrite("type", &Pet::type);
+        .def_readwrite("type", &Pet::type)
+        .def_readwrite("attr", &Pet::attr);
 
     py::enum_<Pet::Kind>(pet, "Kind")
         .value("Dog", Pet::Kind::Dog)
         .value("Cat", Pet::Kind::Cat)
         .export_values();
 
-To ensure that the ``Kind`` type is created within the scope of ``Pet``, the
-``pet`` :class:`class_` instance must be supplied to the :class:`enum_`.
+    py::class_<Pet::Attributes> attributes(pet, "Attributes")
+        .def(py::init<>())
+        .def_readwrite("age", &Pet::Attributes::age);
+
+
+To ensure that the nested types ``Kind`` and ``Attributes`` are created within the scope of ``Pet``, the
+``pet`` :class:`class_` instance must be supplied to the :class:`enum_` and :class:`class_`
 constructor. The :func:`enum_::export_values` function exports the enum entries
 into the parent scope, which should be skipped for newer C++11-style strongly
 typed enums.
 
 .. code-block:: pycon
 
-    >>> p = Pet('Lucy', Pet.Cat)
+    >>> p = Pet("Lucy", Pet.Cat)
     >>> p.type
     Kind.Cat
     >>> int(p.type)
@@ -508,7 +518,7 @@ The ``name`` property returns the name of the enum value as a unicode string.
 
     .. code-block:: pycon
 
-        >>> p = Pet( "Lucy", Pet.Cat )
+        >>> p = Pet("Lucy", Pet.Cat)
         >>> pet_type = p.type
         >>> pet_type
         Pet.Cat

docs/compiling.rst

@@ -42,10 +42,7 @@ An example of a ``setup.py`` using pybind11's helpers:
         ),
     ]
 
-    setup(
-        ...,
-        ext_modules=ext_modules
-    )
+    setup(..., ext_modules=ext_modules)
 
 If you want to do an automatic search for the highest supported C++ standard,
 that is supported via a ``build_ext`` command override; it will only affect
@@ -64,11 +61,7 @@ that is supported via a ``build_ext`` command override; it will only affect
         ),
     ]
 
-    setup(
-        ...,
-        cmdclass={"build_ext": build_ext},
-        ext_modules=ext_modules
-    )
+    setup(..., cmdclass={"build_ext": build_ext}, ext_modules=ext_modules)
 
 If you have single-file extension modules that are directly stored in the
 Python source tree (``foo.cpp`` in the same directory as where a ``foo.py``
@@ -113,7 +106,7 @@ with the following:
 
     from pybind11.setup_helpers import ParallelCompile, naive_recompile
 
-    SmartCompile("NPY_NUM_BUILD_JOBS", needs_recompile=naive_recompile).install()
+    ParallelCompile("NPY_NUM_BUILD_JOBS", needs_recompile=naive_recompile).install()
 
 
 If you have a more complex build, you can implement a smarter function and pass

docs/conf.py

@@ -13,12 +13,11 @@
 # All configuration values have a default; values that are commented out
 # serve to show the default.
 
-import sys
 import os
-import shlex
+import re
 import subprocess
+import sys
 from pathlib import Path
-import re
 
 DIR = Path(__file__).parent.resolve()
 

docs/faq.rst

@@ -54,7 +54,7 @@ provided by the caller -- in fact, it does nothing at all.
 .. code-block:: python
 
     def increment(i):
-        i += 1 # nope..
+        i += 1  # nope..
 
 pybind11 is also affected by such language-level conventions, which means that
 binding ``increment`` or ``increment_ptr`` will also create Python functions

docs/reference.rst

@@ -63,6 +63,9 @@ Convenience functions converting to Python types
 .. doxygenfunction:: make_key_iterator(Iterator, Sentinel, Extra &&...)
 .. doxygenfunction:: make_key_iterator(Type &, Extra&&...)
 
+.. doxygenfunction:: make_value_iterator(Iterator, Sentinel, Extra &&...)
+.. doxygenfunction:: make_value_iterator(Type &, Extra&&...)
+
 .. _extras:
 
 Passing extra arguments to ``def`` or ``class_``

include/pybind11/attr.h

@@ -12,13 +12,17 @@
 
 #include "cast.h"
 
+#include <functional>
+
 PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 
 /// \addtogroup annotations
 /// @{
 
 /// Annotation for methods
-struct is_method { handle class_; is_method(const handle &c) : class_(c) { } };
+struct is_method { handle class_;
+    explicit is_method(const handle &c) : class_(c) {}
+};
 
 /// Annotation for operators
 struct is_operator { };
@@ -27,16 +31,24 @@ struct is_operator { };
 struct is_final { };
 
 /// Annotation for parent scope
-struct scope { handle value; scope(const handle &s) : value(s) { } };
+struct scope { handle value;
+    explicit scope(const handle &s) : value(s) {}
+};
 
 /// Annotation for documentation
-struct doc { const char *value; doc(const char *value) : value(value) { } };
+struct doc { const char *value;
+    explicit doc(const char *value) : value(value) {}
+};
 
 /// Annotation for function names
-struct name { const char *value; name(const char *value) : value(value) { } };
+struct name { const char *value;
+    explicit name(const char *value) : value(value) {}
+};
 
 /// Annotation indicating that a function is an overload associated with a given "sibling"
-struct sibling { handle value; sibling(const handle &value) : value(value.ptr()) { } };
+struct sibling { handle value;
+    explicit sibling(const handle &value) : value(value.ptr()) {}
+};
 
 /// Annotation indicating that a class derives from another given type
 template <typename T> struct base {
@@ -69,8 +81,27 @@ struct metaclass {
     explicit metaclass(handle value) : value(value) { }
 };
 
+/// Specifies a custom callback with signature `void (PyHeapTypeObject*)` that
+/// may be used to customize the Python type.
+///
+/// The callback is invoked immediately before `PyType_Ready`.
+///
+/// Note: This is an advanced interface, and uses of it may require changes to
+/// work with later versions of pybind11.  You may wish to consult the
+/// implementation of `make_new_python_type` in `detail/classes.h` to understand
+/// the context in which the callback will be run.
+struct custom_type_setup {
+    using callback = std::function<void(PyHeapTypeObject *heap_type)>;
+
+    explicit custom_type_setup(callback value) : value(std::move(value)) {}
+
+    callback value;
+};
+
 /// Annotation that marks a class as local to the module:
-struct module_local { const bool value; constexpr module_local(bool v = true) : value(v) { } };
+struct module_local { const bool value;
+    constexpr explicit module_local(bool v = true) : value(v) {}
+};
 
 /// Annotation to mark enums as an arithmetic type
 struct arithmetic { };
@@ -124,7 +155,7 @@ enum op_id : int;
 enum op_type : int;
 struct undefined_t;
 template <op_id id, op_type ot, typename L = undefined_t, typename R = undefined_t> struct op_;
-inline void keep_alive_impl(size_t Nurse, size_t Patient, function_call &call, handle ret);
+void keep_alive_impl(size_t Nurse, size_t Patient, function_call &call, handle ret);
 
 /// Internal data structure which holds metadata about a keyword argument
 struct argument_record {
@@ -260,6 +291,9 @@ struct type_record {
     /// Custom metaclass (optional)
     handle metaclass;
 
+    /// Custom type setup.
+    custom_type_setup::callback custom_type_setup_callback;
+
     /// Multiple inheritance marker
     bool multiple_inheritance : 1;
 
@@ -464,6 +498,13 @@ struct process_attribute<dynamic_attr> : process_attribute_default<dynamic_attr>
     static void init(const dynamic_attr &, type_record *r) { r->dynamic_attr = true; }
 };
 
+template <>
+struct process_attribute<custom_type_setup> {
+    static void init(const custom_type_setup &value, type_record *r) {
+        r->custom_type_setup_callback = value.value;
+    }
+};
+
 template <>
 struct process_attribute<is_final> : process_attribute_default<is_final> {
     static void init(const is_final &, type_record *r) { r->is_final = true; }
@@ -516,20 +557,31 @@ template <size_t Nurse, size_t Patient> struct process_attribute<keep_alive<Nurs
 /// Recursively iterate over variadic template arguments
 template <typename... Args> struct process_attributes {
     static void init(const Args&... args, function_record *r) {
-        int unused[] = { 0, (process_attribute<typename std::decay<Args>::type>::init(args, r), 0) ... };
-        ignore_unused(unused);
+        PYBIND11_WORKAROUND_INCORRECT_MSVC_C4100(r);
+        PYBIND11_WORKAROUND_INCORRECT_GCC_UNUSED_BUT_SET_PARAMETER(r);
+        using expander = int[];
+        (void) expander{
+            0, ((void) process_attribute<typename std::decay<Args>::type>::init(args, r), 0)...};
     }
     static void init(const Args&... args, type_record *r) {
-        int unused[] = { 0, (process_attribute<typename std::decay<Args>::type>::init(args, r), 0) ... };
-        ignore_unused(unused);
+        PYBIND11_WORKAROUND_INCORRECT_MSVC_C4100(r);
+        PYBIND11_WORKAROUND_INCORRECT_GCC_UNUSED_BUT_SET_PARAMETER(r);
+        using expander = int[];
+        (void) expander{0,
+                        (process_attribute<typename std::decay<Args>::type>::init(args, r), 0)...};
     }
     static void precall(function_call &call) {
-        int unused[] = { 0, (process_attribute<typename std::decay<Args>::type>::precall(call), 0) ... };
-        ignore_unused(unused);
+        PYBIND11_WORKAROUND_INCORRECT_MSVC_C4100(call);
+        using expander = int[];
+        (void) expander{0,
+                        (process_attribute<typename std::decay<Args>::type>::precall(call), 0)...};
     }
     static void postcall(function_call &call, handle fn_ret) {
-        int unused[] = { 0, (process_attribute<typename std::decay<Args>::type>::postcall(call, fn_ret), 0) ... };
-        ignore_unused(unused);
+        PYBIND11_WORKAROUND_INCORRECT_MSVC_C4100(call, fn_ret);
+        PYBIND11_WORKAROUND_INCORRECT_GCC_UNUSED_BUT_SET_PARAMETER(fn_ret);
+        using expander = int[];
+        (void) expander{
+            0, (process_attribute<typename std::decay<Args>::type>::postcall(call, fn_ret), 0)...};
     }
 };
 
@@ -545,6 +597,7 @@ template <typename... Extra,
           size_t named = constexpr_sum(std::is_base_of<arg, Extra>::value...),
           size_t self  = constexpr_sum(std::is_same<is_method, Extra>::value...)>
 constexpr bool expected_num_args(size_t nargs, bool has_args, bool has_kwargs) {
+    PYBIND11_WORKAROUND_INCORRECT_MSVC_C4100(nargs, has_args, has_kwargs);
     return named == 0 || (self + named + size_t(has_args) + size_t(has_kwargs)) == nargs;
 }
 

include/pybind11/cast.h

@@ -82,7 +82,7 @@ public:
         return caster_t::cast(&src.get(), policy, parent);
     }
     template <typename T> using cast_op_type = std::reference_wrapper<type>;
-    operator std::reference_wrapper<type>() { return cast_op<type &>(subcaster); }
+    explicit operator std::reference_wrapper<type>() { return cast_op<type &>(subcaster); }
 };
 
 #define PYBIND11_TYPE_CASTER(type, py_name)                                                       \
@@ -102,9 +102,9 @@ public:
         }                                                                                         \
         return cast(*src, policy, parent);                                                        \
     }                                                                                             \
-    operator type *() { return &value; }                                                          \
-    operator type &() { return value; }                                                           \
-    operator type &&() && { return std::move(value); }                                            \
+    operator type *() { return &value; }               /* NOLINT(bugprone-macro-parentheses) */   \
+    operator type &() { return value; }                /* NOLINT(bugprone-macro-parentheses) */   \
+    operator type &&() && { return std::move(value); } /* NOLINT(bugprone-macro-parentheses) */   \
     template <typename T_>                                                                        \
     using cast_op_type = pybind11::detail::movable_cast_op_type<T_>
 
@@ -145,9 +145,8 @@ public:
                 py_value = (py_type) PyFloat_AsDouble(src.ptr());
             else
                 return false;
-        } else if (PyFloat_Check(src.ptr())) {
-            return false;
-        } else if (!convert && !PYBIND11_LONG_CHECK(src.ptr()) && !index_check(src.ptr())) {
+        } else if (PyFloat_Check(src.ptr())
+                   || (!convert && !PYBIND11_LONG_CHECK(src.ptr()) && !index_check(src.ptr()))) {
             return false;
         } else {
             handle src_or_index = src;
@@ -280,7 +279,7 @@ public:
     }
 
     template <typename T> using cast_op_type = void*&;
-    operator void *&() { return value; }
+    explicit operator void *&() { return value; }
     static constexpr auto name = _("capsule");
 private:
     void *value = nullptr;
@@ -378,13 +377,33 @@ template <typename StringType, bool IsView = false> struct string_caster {
 #endif
         }
 
+#if PY_VERSION_HEX >= 0x03030000
+        // On Python >= 3.3, for UTF-8 we avoid the need for a temporary `bytes`
+        // object by using `PyUnicode_AsUTF8AndSize`.
+        if (PYBIND11_SILENCE_MSVC_C4127(UTF_N == 8)) {
+            Py_ssize_t size = -1;
+            const auto *buffer
+                = reinterpret_cast<const CharT *>(PyUnicode_AsUTF8AndSize(load_src.ptr(), &size));
+            if (!buffer) {
+                PyErr_Clear();
+                return false;
+            }
+            value = StringType(buffer, static_cast<size_t>(size));
+            return true;
+        }
+#endif
+
         auto utfNbytes = reinterpret_steal<object>(PyUnicode_AsEncodedString(
             load_src.ptr(), UTF_N == 8 ? "utf-8" : UTF_N == 16 ? "utf-16" : "utf-32", nullptr));
         if (!utfNbytes) { PyErr_Clear(); return false; }
 
         const auto *buffer = reinterpret_cast<const CharT *>(PYBIND11_BYTES_AS_STRING(utfNbytes.ptr()));
         size_t length = (size_t) PYBIND11_BYTES_SIZE(utfNbytes.ptr()) / sizeof(CharT);
-        if (UTF_N > 8) { buffer++; length--; } // Skip BOM for UTF-16/32
+        // Skip BOM for UTF-16/32
+        if (PYBIND11_SILENCE_MSVC_C4127(UTF_N > 8)) {
+            buffer++;
+            length--;
+        }
         value = StringType(buffer, length);
 
         // If we're loading a string_view we need to keep the encoded Python object alive:
@@ -484,8 +503,10 @@ public:
         return StringCaster::cast(StringType(1, src), policy, parent);
     }
 
-    operator CharT*() { return none ? nullptr : const_cast<CharT *>(static_cast<StringType &>(str_caster).c_str()); }
-    operator CharT&() {
+    explicit operator CharT *() {
+        return none ? nullptr : const_cast<CharT *>(static_cast<StringType &>(str_caster).c_str());
+    }
+    explicit operator CharT &() {
         if (none)
             throw value_error("Cannot convert None to a character");
 
@@ -499,7 +520,7 @@ public:
         // out how long the first encoded character is in bytes to distinguish between these two
         // errors.  We also allow want to allow unicode characters U+0080 through U+00FF, as those
         // can fit into a single char value.
-        if (StringCaster::UTF_N == 8 && str_len > 1 && str_len <= 4) {
+        if (PYBIND11_SILENCE_MSVC_C4127(StringCaster::UTF_N == 8) && str_len > 1 && str_len <= 4) {
             auto v0 = static_cast<unsigned char>(value[0]);
             // low bits only: 0-127
             // 0b110xxxxx - start of 2-byte sequence
@@ -524,7 +545,7 @@ public:
         // UTF-16 is much easier: we can only have a surrogate pair for values above U+FFFF, thus a
         // surrogate pair with total length 2 instantly indicates a range error (but not a "your
         // string was too long" error).
-        else if (StringCaster::UTF_N == 16 && str_len == 2) {
+        else if (PYBIND11_SILENCE_MSVC_C4127(StringCaster::UTF_N == 16) && str_len == 2) {
             one_char = static_cast<CharT>(value[0]);
             if (one_char >= 0xD800 && one_char < 0xE000)
                 throw value_error("Character code point not in range(0x10000)");
@@ -578,8 +599,8 @@ public:
 
     template <typename T> using cast_op_type = type;
 
-    operator type() & { return implicit_cast(indices{}); }
-    operator type() && { return std::move(*this).implicit_cast(indices{}); }
+    explicit operator type() & { return implicit_cast(indices{}); }
+    explicit operator type() && { return std::move(*this).implicit_cast(indices{}); }
 
 protected:
     template <size_t... Is>
@@ -605,6 +626,8 @@ protected:
     /* Implementation: Convert a C++ tuple into a Python tuple */
     template <typename T, size_t... Is>
     static handle cast_impl(T &&src, return_value_policy policy, handle parent, index_sequence<Is...>) {
+        PYBIND11_WORKAROUND_INCORRECT_MSVC_C4100(src, policy, parent);
+        PYBIND11_WORKAROUND_INCORRECT_GCC_UNUSED_BUT_SET_PARAMETER(policy, parent);
         std::array<object, size> entries{{
             reinterpret_steal<object>(make_caster<Ts>::cast(std::get<Is>(std::forward<T>(src)), policy, parent))...
         }};
@@ -777,7 +800,7 @@ struct pyobject_caster {
         // For Python 2, without this implicit conversion, Python code would
         // need to be cluttered with six.ensure_text() or similar, only to be
         // un-cluttered later after Python 2 support is dropped.
-        if (std::is_same<T, str>::value && isinstance<bytes>(src)) {
+        if (PYBIND11_SILENCE_MSVC_C4127(std::is_same<T, str>::value) && isinstance<bytes>(src)) {
             PyObject *str_from_bytes = PyUnicode_FromEncodedObject(src.ptr(), "utf-8", nullptr);
             if (!str_from_bytes) throw error_already_set();
             value = reinterpret_steal<type>(str_from_bytes);
@@ -1013,6 +1036,16 @@ template <return_value_policy policy = return_value_policy::automatic_reference,
     return result;
 }
 
+#if PY_VERSION_HEX >= 0x03030000
+template <typename... Args,
+          typename = detail::enable_if_t<args_are_all_keyword_or_ds<Args...>()>>
+object make_simple_namespace(Args&&... args_) {
+    PyObject *ns = _PyNamespace_New(dict(std::forward<Args>(args_)...).ptr());
+    if (!ns) throw error_already_set();
+    return reinterpret_steal<object>(ns);
+}
+#endif
+
 /// \ingroup annotations
 /// Annotation for arguments
 struct arg {
@@ -1161,13 +1194,14 @@ public:
     }
 
     template <typename Return, typename Guard, typename Func>
+    // NOLINTNEXTLINE(readability-const-return-type)
     enable_if_t<!std::is_void<Return>::value, Return> call(Func &&f) && {
-        return std::move(*this).template call_impl<Return>(std::forward<Func>(f), indices{}, Guard{});
+        return std::move(*this).template call_impl<remove_cv_t<Return>>(std::forward<Func>(f), indices{}, Guard{});
     }
 
     template <typename Return, typename Guard, typename Func>
     enable_if_t<std::is_void<Return>::value, void_type> call(Func &&f) && {
-        std::move(*this).template call_impl<Return>(std::forward<Func>(f), indices{}, Guard{});
+        std::move(*this).template call_impl<remove_cv_t<Return>>(std::forward<Func>(f), indices{}, Guard{});
         return void_type();
     }
 
@@ -1231,8 +1265,8 @@ public:
         // Tuples aren't (easily) resizable so a list is needed for collection,
         // but the actual function call strictly requires a tuple.
         auto args_list = list();
-        int _[] = { 0, (process(args_list, std::forward<Ts>(values)), 0)... };
-        ignore_unused(_);
+        using expander = int[];
+        (void) expander{0, (process(args_list, std::forward<Ts>(values)), 0)...};
 
         m_args = std::move(args_list);
     }

include/pybind11/detail/class.h

@@ -210,7 +210,7 @@ extern "C" inline void pybind11_meta_dealloc(PyObject *obj) {
         internals.direct_conversions.erase(tindex);
 
         if (tinfo->module_local)
-            registered_local_types_cpp().erase(tindex);
+            get_local_internals().registered_types_cpp.erase(tindex);
         else
             internals.registered_types_cpp.erase(tindex);
         internals.registered_types_py.erase(tinfo->type);
@@ -683,11 +683,13 @@ inline PyObject* make_new_python_type(const type_record &rec) {
     if (rec.buffer_protocol)
         enable_buffer_protocol(heap_type);
 
+    if (rec.custom_type_setup_callback)
+        rec.custom_type_setup_callback(heap_type);
+
     if (PyType_Ready(type) < 0)
         pybind11_fail(std::string(rec.name) + ": PyType_Ready failed (" + error_string() + ")!");
 
-    assert(rec.dynamic_attr ? PyType_HasFeature(type, Py_TPFLAGS_HAVE_GC)
-                            : !PyType_HasFeature(type, Py_TPFLAGS_HAVE_GC));
+    assert(!rec.dynamic_attr || PyType_HasFeature(type, Py_TPFLAGS_HAVE_GC));
 
     /* Register type with the parent scope */
     if (rec.scope)

include/pybind11/detail/common.h

@@ -10,12 +10,12 @@
 #pragma once
 
 #define PYBIND11_VERSION_MAJOR 2
-#define PYBIND11_VERSION_MINOR 7
-#define PYBIND11_VERSION_PATCH 1
+#define PYBIND11_VERSION_MINOR 8
+#define PYBIND11_VERSION_PATCH 0
 
 // Similar to Python's convention: https://docs.python.org/3/c-api/apiabiversion.html
 // Additional convention: 0xD = dev
-#define PYBIND11_VERSION_HEX 0x02070100
+#define PYBIND11_VERSION_HEX 0x02080000
 
 #define PYBIND11_NAMESPACE_BEGIN(name) namespace name {
 #define PYBIND11_NAMESPACE_END(name) }
@@ -99,10 +99,26 @@
 #  endif
 #endif
 
-#if defined(_MSC_VER)
-#  define PYBIND11_NOINLINE __declspec(noinline)
+// For CUDA, GCC7, GCC8:
+// PYBIND11_NOINLINE_FORCED is incompatible with `-Wattributes -Werror`.
+// When defining PYBIND11_NOINLINE_FORCED, it is best to also use `-Wno-attributes`.
+// However, the measured shared-library size saving when using noinline are only
+// 1.7% for CUDA, -0.2% for GCC7, and 0.0% for GCC8 (using -DCMAKE_BUILD_TYPE=MinSizeRel,
+// the default under pybind11/tests).
+#if !defined(PYBIND11_NOINLINE_FORCED) && \
+    (defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8)))
+#  define PYBIND11_NOINLINE_DISABLED
+#endif
+
+// The PYBIND11_NOINLINE macro is for function DEFINITIONS.
+// In contrast, FORWARD DECLARATIONS should never use this macro:
+// https://stackoverflow.com/questions/9317473/forward-declaration-of-inline-functions
+#if defined(PYBIND11_NOINLINE_DISABLED) // Option for maximum portability and experimentation.
+#  define PYBIND11_NOINLINE inline
+#elif defined(_MSC_VER)
+#  define PYBIND11_NOINLINE __declspec(noinline) inline
 #else
-#  define PYBIND11_NOINLINE __attribute__ ((noinline))
+#  define PYBIND11_NOINLINE __attribute__ ((noinline)) inline
 #endif
 
 #if defined(__MINGW32__)
@@ -145,7 +161,26 @@
 
 // https://en.cppreference.com/w/c/chrono/localtime
 #if defined(__STDC_LIB_EXT1__) && !defined(__STDC_WANT_LIB_EXT1__)
-#    define __STDC_WANT_LIB_EXT1__
+#  define __STDC_WANT_LIB_EXT1__
+#endif
+
+#ifdef __has_include
+// std::optional (but including it in c++14 mode isn't allowed)
+#  if defined(PYBIND11_CPP17) && __has_include(<optional>)
+#    define PYBIND11_HAS_OPTIONAL 1
+#  endif
+// std::experimental::optional (but not allowed in c++11 mode)
+#  if defined(PYBIND11_CPP14) && (__has_include(<experimental/optional>) && \
+                                 !__has_include(<optional>))
+#    define PYBIND11_HAS_EXP_OPTIONAL 1
+#  endif
+// std::variant
+#  if defined(PYBIND11_CPP17) && __has_include(<variant>)
+#    define PYBIND11_HAS_VARIANT 1
+#  endif
+#elif defined(_MSC_VER) && defined(PYBIND11_CPP17)
+#  define PYBIND11_HAS_OPTIONAL 1
+#  define PYBIND11_HAS_VARIANT 1
 #endif
 
 #include <Python.h>
@@ -220,8 +255,8 @@
 #define PYBIND11_BYTES_SIZE PyBytes_Size
 #define PYBIND11_LONG_CHECK(o) PyLong_Check(o)
 #define PYBIND11_LONG_AS_LONGLONG(o) PyLong_AsLongLong(o)
-#define PYBIND11_LONG_FROM_SIGNED(o) PyLong_FromSsize_t((ssize_t) o)
-#define PYBIND11_LONG_FROM_UNSIGNED(o) PyLong_FromSize_t((size_t) o)
+#define PYBIND11_LONG_FROM_SIGNED(o) PyLong_FromSsize_t((ssize_t) (o))
+#define PYBIND11_LONG_FROM_UNSIGNED(o) PyLong_FromSize_t((size_t) (o))
 #define PYBIND11_BYTES_NAME "bytes"
 #define PYBIND11_STRING_NAME "str"
 #define PYBIND11_SLICE_OBJECT PyObject
@@ -299,6 +334,19 @@ extern "C" {
         }                                                                      \
     }
 
+#if PY_VERSION_HEX >= 0x03030000
+
+#define PYBIND11_CATCH_INIT_EXCEPTIONS \
+        catch (pybind11::error_already_set &e) {                                 \
+            pybind11::raise_from(e, PyExc_ImportError, "initialization failed"); \
+            return nullptr;                                                      \
+        } catch (const std::exception &e) {                                      \
+            PyErr_SetString(PyExc_ImportError, e.what());                        \
+            return nullptr;                                                      \
+        }                                                                        \
+
+#else
+
 #define PYBIND11_CATCH_INIT_EXCEPTIONS \
         catch (pybind11::error_already_set &e) {                               \
             PyErr_SetString(PyExc_ImportError, e.what());                      \
@@ -308,6 +356,8 @@ extern "C" {
             return nullptr;                                                    \
         }                                                                      \
 
+#endif
+
 /** \rst
     ***Deprecated in favor of PYBIND11_MODULE***
 
@@ -356,30 +406,35 @@ extern "C" {
             });
         }
 \endrst */
-#define PYBIND11_MODULE(name, variable)                                        \
-    static ::pybind11::module_::module_def                                     \
-        PYBIND11_CONCAT(pybind11_module_def_, name) PYBIND11_MAYBE_UNUSED;     \
-    PYBIND11_MAYBE_UNUSED                                                      \
-    static void PYBIND11_CONCAT(pybind11_init_, name)(::pybind11::module_ &);  \
-    PYBIND11_PLUGIN_IMPL(name) {                                               \
-        PYBIND11_CHECK_PYTHON_VERSION                                          \
-        PYBIND11_ENSURE_INTERNALS_READY                                        \
-        auto m = ::pybind11::module_::create_extension_module(                 \
-            PYBIND11_TOSTRING(name), nullptr,                                  \
-            &PYBIND11_CONCAT(pybind11_module_def_, name));                     \
-        try {                                                                  \
-            PYBIND11_CONCAT(pybind11_init_, name)(m);                          \
-            return m.ptr();                                                    \
-        } PYBIND11_CATCH_INIT_EXCEPTIONS                                       \
-    }                                                                          \
-    void PYBIND11_CONCAT(pybind11_init_, name)(::pybind11::module_ &variable)
-
+#define PYBIND11_MODULE(name, variable)                                                           \
+    static ::pybind11::module_::module_def PYBIND11_CONCAT(pybind11_module_def_, name)            \
+        PYBIND11_MAYBE_UNUSED;                                                                    \
+    PYBIND11_MAYBE_UNUSED                                                                         \
+    static void PYBIND11_CONCAT(pybind11_init_, name)(::pybind11::module_ &);                     \
+    PYBIND11_PLUGIN_IMPL(name) {                                                                  \
+        PYBIND11_CHECK_PYTHON_VERSION                                                             \
+        PYBIND11_ENSURE_INTERNALS_READY                                                           \
+        auto m = ::pybind11::module_::create_extension_module(                                    \
+            PYBIND11_TOSTRING(name), nullptr, &PYBIND11_CONCAT(pybind11_module_def_, name));      \
+        try {                                                                                     \
+            PYBIND11_CONCAT(pybind11_init_, name)(m);                                             \
+            return m.ptr();                                                                       \
+        }                                                                                         \
+        PYBIND11_CATCH_INIT_EXCEPTIONS                                                            \
+    }                                                                                             \
+    void PYBIND11_CONCAT(pybind11_init_, name)(::pybind11::module_ & (variable))
 
 PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 
 using ssize_t = Py_ssize_t;
 using size_t  = std::size_t;
 
+template <typename IntType>
+inline ssize_t ssize_t_cast(const IntType &val) {
+    static_assert(sizeof(IntType) <= sizeof(ssize_t), "Implicit narrowing is not permitted.");
+    return static_cast<ssize_t>(val);
+}
+
 /// Approach used to cast a previously unknown C++ instance into a Python object
 enum class return_value_policy : uint8_t {
     /** This is the default return value policy, which falls back to the policy
@@ -507,7 +562,7 @@ struct instance {
     void allocate_layout();
 
     /// Destroys/deallocates all of the above
-    void deallocate_layout() const;
+    void deallocate_layout();
 
     /// Returns the value_and_holder wrapper for the given type (or the first, if `find_type`
     /// omitted).  Returns a default-constructed (with `.inst = nullptr`) object on failure if
@@ -735,9 +790,6 @@ using function_signature_t = conditional_t<
 template <typename T> using is_lambda = satisfies_none_of<remove_reference_t<T>,
         std::is_function, std::is_pointer, std::is_member_pointer>;
 
-/// Ignore that a variable is unused in compiler warnings
-inline void ignore_unused(const int *) { }
-
 // [workaround(intel)] Internal error on fold expression
 /// Apply a function over each element of a parameter pack
 #if defined(__cpp_fold_expressions) && !defined(__INTEL_COMPILER)
@@ -782,8 +834,8 @@ PYBIND11_RUNTIME_EXCEPTION(import_error, PyExc_ImportError)
 PYBIND11_RUNTIME_EXCEPTION(cast_error, PyExc_RuntimeError) /// Thrown when pybind11::cast or handle::call fail due to a type casting error
 PYBIND11_RUNTIME_EXCEPTION(reference_cast_error, PyExc_RuntimeError) /// Used internally
 
-[[noreturn]] PYBIND11_NOINLINE inline void pybind11_fail(const char *reason) { throw std::runtime_error(reason); }
-[[noreturn]] PYBIND11_NOINLINE inline void pybind11_fail(const std::string &reason) { throw std::runtime_error(reason); }
+[[noreturn]] PYBIND11_NOINLINE void pybind11_fail(const char *reason) { throw std::runtime_error(reason); }
+[[noreturn]] PYBIND11_NOINLINE void pybind11_fail(const std::string &reason) { throw std::runtime_error(reason); }
 
 template <typename T, typename SFINAE = void> struct format_descriptor { };
 
@@ -885,6 +937,7 @@ public:
 
     // Implicit conversion constructor from any arbitrary container type with values convertible to T
     template <typename Container, typename = enable_if_t<std::is_convertible<decltype(*std::begin(std::declval<const Container &>())), T>::value>>
+    // NOLINTNEXTLINE(google-explicit-constructor)
     any_container(const Container &c) : any_container(std::begin(c), std::end(c)) { }
 
     // initializer_list's aren't deducible, so don't get matched by the above template; we need this
@@ -893,9 +946,11 @@ public:
     any_container(const std::initializer_list<TIn> &c) : any_container(c.begin(), c.end()) { }
 
     // Avoid copying if given an rvalue vector of the correct type.
+    // NOLINTNEXTLINE(google-explicit-constructor)
     any_container(std::vector<T> &&v) : v(std::move(v)) { }
 
     // Moves the vector out of an rvalue any_container
+    // NOLINTNEXTLINE(google-explicit-constructor)
     operator std::vector<T> &&() && { return std::move(v); }
 
     // Dereferencing obtains a reference to the underlying vector
@@ -928,5 +983,39 @@ inline static std::shared_ptr<T> try_get_shared_from_this(std::enable_shared_fro
 #endif
 }
 
+// For silencing "unused" compiler warnings in special situations.
+template <typename... Args>
+#if defined(_MSC_VER) && _MSC_VER >= 1910 && _MSC_VER < 1920 // MSVC 2017
+constexpr
+#endif
+inline void silence_unused_warnings(Args &&...) {}
+
+// MSVC warning C4100: Unreferenced formal parameter
+#if defined(_MSC_VER) && _MSC_VER <= 1916
+#    define PYBIND11_WORKAROUND_INCORRECT_MSVC_C4100(...)                                         \
+        detail::silence_unused_warnings(__VA_ARGS__)
+#else
+#    define PYBIND11_WORKAROUND_INCORRECT_MSVC_C4100(...)
+#endif
+
+// GCC -Wunused-but-set-parameter  All GCC versions (as of July 2021).
+#if defined(__GNUG__) && !defined(__clang__) && !defined(__INTEL_COMPILER)
+#    define PYBIND11_WORKAROUND_INCORRECT_GCC_UNUSED_BUT_SET_PARAMETER(...)                   \
+        detail::silence_unused_warnings(__VA_ARGS__)
+#else
+#    define PYBIND11_WORKAROUND_INCORRECT_GCC_UNUSED_BUT_SET_PARAMETER(...)
+#endif
+
+#if defined(_MSC_VER) // All versions (as of July 2021).
+
+// warning C4127: Conditional expression is constant
+constexpr inline bool silence_msvc_c4127(bool cond) { return cond; }
+
+#    define PYBIND11_SILENCE_MSVC_C4127(...) ::pybind11::detail::silence_msvc_c4127(__VA_ARGS__)
+
+#else
+#    define PYBIND11_SILENCE_MSVC_C4127(...) __VA_ARGS__
+#endif
+
 PYBIND11_NAMESPACE_END(detail)
 PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)

include/pybind11/detail/descr.h

@@ -26,12 +26,14 @@ struct descr {
     char text[N + 1]{'\0'};
 
     constexpr descr() = default;
+    // NOLINTNEXTLINE(google-explicit-constructor)
     constexpr descr(char const (&s)[N+1]) : descr(s, make_index_sequence<N>()) { }
 
     template <size_t... Is>
     constexpr descr(char const (&s)[N+1], index_sequence<Is...>) : text{s[Is]..., '\0'} { }
 
     template <typename... Chars>
+    // NOLINTNEXTLINE(google-explicit-constructor)
     constexpr descr(char c, Chars... cs) : text{c, static_cast<char>(cs)..., '\0'} { }
 
     static constexpr std::array<const std::type_info *, sizeof...(Ts) + 1> types() {
@@ -42,6 +44,7 @@ struct descr {
 template <size_t N1, size_t N2, typename... Ts1, typename... Ts2, size_t... Is1, size_t... Is2>
 constexpr descr<N1 + N2, Ts1..., Ts2...> plus_impl(const descr<N1, Ts1...> &a, const descr<N2, Ts2...> &b,
                                                    index_sequence<Is1...>, index_sequence<Is2...>) {
+    PYBIND11_WORKAROUND_INCORRECT_MSVC_C4100(b);
     return {a.text[Is1]..., b.text[Is2]...};
 }
 
@@ -74,7 +77,8 @@ constexpr enable_if_t<B, T1> _(const T1 &d, const T2 &) { return d; }
 template <bool B, typename T1, typename T2>
 constexpr enable_if_t<!B, T2> _(const T1 &, const T2 &d) { return d; }
 
-template <size_t Size> auto constexpr _() -> decltype(int_to_str<Size / 10, Size % 10>::digits) {
+template <size_t Size>
+auto constexpr _() -> remove_cv_t<decltype(int_to_str<Size / 10, Size % 10>::digits)> {
     return int_to_str<Size / 10, Size % 10>::digits;
 }
 

include/pybind11/detail/init.h

@@ -23,7 +23,7 @@ public:
     }
 
     template <typename> using cast_op_type = value_and_holder &;
-    operator value_and_holder &() { return *value; }
+    explicit operator value_and_holder &() { return *value; }
     static constexpr auto name = _<value_and_holder>();
 
 private:
@@ -94,8 +94,9 @@ void construct(...) {
 // construct an Alias from the returned base instance.
 template <typename Class>
 void construct(value_and_holder &v_h, Cpp<Class> *ptr, bool need_alias) {
+    PYBIND11_WORKAROUND_INCORRECT_MSVC_C4100(need_alias);
     no_nullptr(ptr);
-    if (Class::has_alias && need_alias && !is_alias<Class>(ptr)) {
+    if (PYBIND11_SILENCE_MSVC_C4127(Class::has_alias) && need_alias && !is_alias<Class>(ptr)) {
         // We're going to try to construct an alias by moving the cpp type.  Whether or not
         // that succeeds, we still need to destroy the original cpp pointer (either the
         // moved away leftover, if the alias construction works, or the value itself if we
@@ -131,10 +132,11 @@ void construct(value_and_holder &v_h, Alias<Class> *alias_ptr, bool) {
 // derived type (through those holder's implicit conversion from derived class holder constructors).
 template <typename Class>
 void construct(value_and_holder &v_h, Holder<Class> holder, bool need_alias) {
+    PYBIND11_WORKAROUND_INCORRECT_MSVC_C4100(need_alias);
     auto *ptr = holder_helper<Holder<Class>>::get(holder);
     no_nullptr(ptr);
     // If we need an alias, check that the held pointer is actually an alias instance
-    if (Class::has_alias && need_alias && !is_alias<Class>(ptr))
+    if (PYBIND11_SILENCE_MSVC_C4127(Class::has_alias) && need_alias && !is_alias<Class>(ptr))
         throw type_error("pybind11::init(): construction failed: returned holder-wrapped instance "
                          "is not an alias instance");
 
@@ -148,9 +150,10 @@ void construct(value_and_holder &v_h, Holder<Class> holder, bool need_alias) {
 // need it, we simply move-construct the cpp value into a new instance.
 template <typename Class>
 void construct(value_and_holder &v_h, Cpp<Class> &&result, bool need_alias) {
+    PYBIND11_WORKAROUND_INCORRECT_MSVC_C4100(need_alias);
     static_assert(std::is_move_constructible<Cpp<Class>>::value,
         "pybind11::init() return-by-value factory function requires a movable class");
-    if (Class::has_alias && need_alias)
+    if (PYBIND11_SILENCE_MSVC_C4127(Class::has_alias) && need_alias)
         construct_alias_from_cpp<Class>(is_alias_constructible<Class>{}, v_h, std::move(result));
     else
         v_h.value_ptr() = new Cpp<Class>(std::move(result));
@@ -219,7 +222,8 @@ template <typename Func, typename Return, typename... Args>
 struct factory<Func, void_type (*)(), Return(Args...)> {
     remove_reference_t<Func> class_factory;
 
-    factory(Func &&f) : class_factory(std::forward<Func>(f)) { }
+    // NOLINTNEXTLINE(google-explicit-constructor)
+    factory(Func &&f) : class_factory(std::forward<Func>(f)) {}
 
     // The given class either has no alias or has no separate alias factory;
     // this always constructs the class itself.  If the class is registered with an alias

include/pybind11/detail/internals.h

@@ -11,8 +11,30 @@
 
 #include "../pytypes.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);
+
 PYBIND11_NAMESPACE_BEGIN(detail)
+
 // Forward declarations
 inline PyTypeObject *make_static_property_type();
 inline PyTypeObject *make_default_metaclass();
@@ -21,30 +43,59 @@ 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
-#    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)
+// Avoid unnecessary allocation of `Py_tss_t`, since we cannot use
+// `Py_LIMITED_API` anyway.
+#    if PYBIND11_INTERNALS_VERSION > 4
+#        define PYBIND11_TLS_KEY_REF Py_tss_t &
+#        ifdef __GNUC__
+// Clang on macOS warns due to `Py_tss_NEEDS_INIT` not specifying an initializer
+// for every field.
+#            define PYBIND11_TLS_KEY_INIT(var)                                                    \
+                _Pragma("GCC diagnostic push")                                         /**/       \
+                    _Pragma("GCC diagnostic ignored \"-Wmissing-field-initializers\"") /**/       \
+                    Py_tss_t var                                                                  \
+                    = Py_tss_NEEDS_INIT;                                                          \
+                _Pragma("GCC diagnostic pop")
+#        else
+#            define PYBIND11_TLS_KEY_INIT(var) Py_tss_t var = Py_tss_NEEDS_INIT;
+#        endif
+#        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_KEY_REF Py_tss_t *
+#        define PYBIND11_TLS_KEY_INIT(var) Py_tss_t *var = nullptr;
+#        define PYBIND11_TLS_KEY_CREATE(var)                                                      \
+            (((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
 #else
-    // Usually an int but a long on Cygwin64 with Python 3.x
-#    define PYBIND11_TLS_KEY_INIT(var) decltype(PyThread_create_key()) var = 0
+// 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
-#        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)
+#    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.
+inline void tls_replace_value(PYBIND11_TLS_KEY_REF key, void *value) {
+    PyThread_delete_key_value(key);
+    PyThread_set_key_value(key, value);
+}
+#        define PYBIND11_TLS_DELETE_VALUE(key) PyThread_delete_key_value(key)
+#        define PYBIND11_TLS_REPLACE_VALUE(key, value)                                            \
+            ::pybind11::detail::tls_replace_value((key), (value))
 #    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))
+#        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
+#    define PYBIND11_TLS_FREE(key) (void) key
 #endif
 
 // Python loads modules by default with dlopen with the RTLD_LOCAL flag; under libc++ and possibly
@@ -100,24 +151,33 @@ struct internals {
     std::unordered_set<std::pair<const PyObject *, const char *>, override_hash> inactive_override_cache;
     type_map<std::vector<bool (*)(PyObject *, void *&)>> direct_conversions;
     std::unordered_map<const PyObject *, std::vector<PyObject *>> patients;
-    std::forward_list<void (*) (std::exception_ptr)> registered_exception_translators;
+    std::forward_list<ExceptionTranslator> registered_exception_translators;
     std::unordered_map<std::string, void *> shared_data; // Custom data to be shared across extensions
-    std::vector<PyObject *> loader_patient_stack; // Used by `loader_life_support`
+#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.
-        // PYBIND11_TLS_FREE is PyThread_tss_free on python 3.7+. On older python, it does 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
-        // of those have anything to do with CPython internals.
-        // PyMem_RawFree *requires* that the `tstate` be allocated with the CPython allocator.
+        // That *SHOULD BE* fine. The following details what happens when PyThread_tss_free is
+        // called. PYBIND11_TLS_FREE is PyThread_tss_free on python 3.7+. On older python, it does
+        // 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 of those have anything to do with CPython internals. PyMem_RawFree *requires*
+        // that the `tstate` be allocated with the CPython allocator.
         PYBIND11_TLS_FREE(tstate);
     }
 #endif
@@ -149,9 +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"
@@ -253,7 +310,7 @@ inline void translate_local_exception(std::exception_ptr p) {
 #endif
 
 /// Return a reference to the current `internals` data
-PYBIND11_NOINLINE inline internals &get_internals() {
+PYBIND11_NOINLINE internals &get_internals() {
     auto **&internals_pp = get_internals_pp();
     if (internals_pp && *internals_pp)
         return **internals_pp;
@@ -287,21 +344,21 @@ PYBIND11_NOINLINE inline internals &get_internals() {
         internals_ptr = new internals();
 #if defined(WITH_THREAD)
 
-        #if PY_VERSION_HEX < 0x03090000
-                PyEval_InitThreads();
-        #endif
+#    if PY_VERSION_HEX < 0x03090000
+        PyEval_InitThreads();
+#    endif
         PyThreadState *tstate = PyThreadState_Get();
-        #if PY_VERSION_HEX >= 0x03070000
-            internals_ptr->tstate = PyThread_tss_alloc();
-            if (!internals_ptr->tstate || (PyThread_tss_create(internals_ptr->tstate) != 0))
-                pybind11_fail("get_internals: could not successfully initialize the TSS key!");
-            PyThread_tss_set(internals_ptr->tstate, tstate);
-        #else
-            internals_ptr->tstate = PyThread_create_key();
-            if (internals_ptr->tstate == -1)
-                pybind11_fail("get_internals: could not successfully initialize the TLS key!");
-            PyThread_set_key_value(internals_ptr->tstate, tstate);
-        #endif
+        if (!PYBIND11_TLS_KEY_CREATE(internals_ptr->tstate)) {
+            pybind11_fail("get_internals: could not successfully initialize the tstate TSS key!");
+        }
+        PYBIND11_TLS_REPLACE_VALUE(internals_ptr->tstate, tstate);
+
+#    if PYBIND11_INTERNALS_VERSION > 4
+        if (!PYBIND11_TLS_KEY_CREATE(internals_ptr->loader_life_support_tls_key)) {
+            pybind11_fail("get_internals: could not successfully initialize the "
+                          "loader_life_support TSS key!");
+        }
+#    endif
         internals_ptr->istate = tstate->interp;
 #endif
         builtins[id] = capsule(internals_pp);
@@ -313,12 +370,57 @@ PYBIND11_NOINLINE inline internals &get_internals() {
     return **internals_pp;
 }
 
-/// Works like `internals.registered_types_cpp`, but for module-local registered types:
-inline type_map<type_info *> &registered_local_types_cpp() {
-    static type_map<type_info *> locals{};
-    return locals;
+// 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,
+// restricted to a single module. Whether a module has local internals or not should not
+// 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;
+    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.
+inline local_internals &get_local_internals() {
+  static local_internals locals;
+  return locals;
 }
 
+
 /// Constructs a std::string with the given arguments, stores it in `internals`, and returns its
 /// `c_str()`.  Such strings objects have a long storage duration -- the internal strings are only
 /// cleared when the program exits or after interpreter shutdown (when embedding), and so are
@@ -335,14 +437,14 @@ PYBIND11_NAMESPACE_END(detail)
 /// Returns a named pointer that is shared among all extension modules (using the same
 /// pybind11 version) running in the current interpreter. Names starting with underscores
 /// are reserved for internal usage. Returns `nullptr` if no matching entry was found.
-inline PYBIND11_NOINLINE void *get_shared_data(const std::string &name) {
+PYBIND11_NOINLINE void *get_shared_data(const std::string &name) {
     auto &internals = detail::get_internals();
     auto it = internals.shared_data.find(name);
     return it != internals.shared_data.end() ? it->second : nullptr;
 }
 
 /// Set the shared data that can be later recovered by `get_shared_data()`.
-inline PYBIND11_NOINLINE void *set_shared_data(const std::string &name, void *data) {
+PYBIND11_NOINLINE void *set_shared_data(const std::string &name, void *data) {
     detail::get_internals().shared_data[name] = data;
     return data;
 }

include/pybind11/detail/type_caster_base.h

@@ -31,47 +31,67 @@ PYBIND11_NAMESPACE_BEGIN(detail)
 /// A life support system for temporary objects created by `type_caster::load()`.
 /// Adding a patient will keep it alive up until the enclosing function returns.
 class loader_life_support {
+private:
+    loader_life_support* parent = nullptr;
+    std::unordered_set<PyObject *> keep_alive;
+
+#if defined(WITH_THREAD)
+    // Store stack pointer in thread-local storage.
+    static PYBIND11_TLS_KEY_REF get_stack_tls_key() {
+#    if PYBIND11_INTERNALS_VERSION == 4
+        return get_local_internals().loader_life_support_tls_key;
+#    else
+        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() {
-        get_internals().loader_patient_stack.push_back(nullptr);
+        parent = get_stack_top();
+        set_stack_top(this);
     }
 
     /// ... and destroyed after it returns
     ~loader_life_support() {
-        auto &stack = get_internals().loader_patient_stack;
-        if (stack.empty())
+        if (get_stack_top() != this)
             pybind11_fail("loader_life_support: internal error");
-
-        auto ptr = stack.back();
-        stack.pop_back();
-        Py_CLEAR(ptr);
-
-        // A heuristic to reduce the stack's capacity (e.g. after long recursive calls)
-        if (stack.capacity() > 16 && !stack.empty() && stack.capacity() / stack.size() > 2)
-            stack.shrink_to_fit();
+        set_stack_top(parent);
+        for (auto* item : keep_alive)
+            Py_DECREF(item);
     }
 
     /// 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) {
-        auto &stack = get_internals().loader_patient_stack;
-        if (stack.empty())
+        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
+            // a location is challenging. Developers could consider printing out
+            // stack frame addresses here using something like __builtin_frame_address(0)
             throw cast_error("When called outside a bound function, py::cast() cannot "
                              "do Python -> C++ conversions which require the creation "
                              "of temporary values");
-
-        auto &list_ptr = stack.back();
-        if (list_ptr == nullptr) {
-            list_ptr = PyList_New(1);
-            if (!list_ptr)
-                pybind11_fail("loader_life_support: error allocating list");
-            PyList_SET_ITEM(list_ptr, 0, h.inc_ref().ptr());
-        } else {
-            auto result = PyList_Append(list_ptr, h.ptr());
-            if (result == -1)
-                pybind11_fail("loader_life_support: error adding patient");
         }
+
+        if (frame->keep_alive.insert(h.ptr()).second)
+            Py_INCREF(h.ptr());
     }
 };
 
@@ -81,7 +101,7 @@ public:
 inline std::pair<decltype(internals::registered_types_py)::iterator, bool> all_type_info_get_cache(PyTypeObject *type);
 
 // Populates a just-created cache entry.
-PYBIND11_NOINLINE inline void all_type_info_populate(PyTypeObject *t, std::vector<type_info *> &bases) {
+PYBIND11_NOINLINE void all_type_info_populate(PyTypeObject *t, std::vector<type_info *> &bases) {
     std::vector<PyTypeObject *> check;
     for (handle parent : reinterpret_borrow<tuple>(t->tp_bases))
         check.push_back((PyTypeObject *) parent.ptr());
@@ -150,7 +170,7 @@ inline const std::vector<detail::type_info *> &all_type_info(PyTypeObject *type)
  * ancestors are pybind11-registered.  Throws an exception if there are multiple bases--use
  * `all_type_info` instead if you want to support multiple bases.
  */
-PYBIND11_NOINLINE inline detail::type_info* get_type_info(PyTypeObject *type) {
+PYBIND11_NOINLINE detail::type_info* get_type_info(PyTypeObject *type) {
     auto &bases = all_type_info(type);
     if (bases.empty())
         return nullptr;
@@ -160,7 +180,7 @@ PYBIND11_NOINLINE inline detail::type_info* get_type_info(PyTypeObject *type) {
 }
 
 inline detail::type_info *get_local_type_info(const std::type_index &tp) {
-    auto &locals = registered_local_types_cpp();
+    auto &locals = get_local_internals().registered_types_cpp;
     auto it = locals.find(tp);
     if (it != locals.end())
         return it->second;
@@ -176,7 +196,7 @@ inline detail::type_info *get_global_type_info(const std::type_index &tp) {
 }
 
 /// Return the type info for a given C++ type; on lookup failure can either throw or return nullptr.
-PYBIND11_NOINLINE inline detail::type_info *get_type_info(const std::type_index &tp,
+PYBIND11_NOINLINE detail::type_info *get_type_info(const std::type_index &tp,
                                                           bool throw_if_missing = false) {
     if (auto ltype = get_local_type_info(tp))
         return ltype;
@@ -191,13 +211,13 @@ PYBIND11_NOINLINE inline detail::type_info *get_type_info(const std::type_index
     return nullptr;
 }
 
-PYBIND11_NOINLINE inline handle get_type_handle(const std::type_info &tp, bool throw_if_missing) {
+PYBIND11_NOINLINE handle get_type_handle(const std::type_info &tp, bool throw_if_missing) {
     detail::type_info *type_info = get_type_info(tp, throw_if_missing);
     return handle(type_info ? ((PyObject *) type_info->type) : nullptr);
 }
 
 // Searches the inheritance graph for a registered Python instance, using all_type_info().
-PYBIND11_NOINLINE inline handle find_registered_python_instance(void *src,
+PYBIND11_NOINLINE handle find_registered_python_instance(void *src,
                                                                 const detail::type_info *tinfo) {
     auto it_instances = get_internals().registered_instances.equal_range(src);
     for (auto it_i = it_instances.first; it_i != it_instances.second; ++it_i) {
@@ -225,7 +245,7 @@ struct value_and_holder {
     value_and_holder() = default;
 
     // Used for past-the-end iterator
-    value_and_holder(size_t index) : index{index} {}
+    explicit value_and_holder(size_t index) : index{index} {}
 
     template <typename V = void> V *&value_ptr() const {
         return reinterpret_cast<V *&>(vh[0]);
@@ -241,7 +261,8 @@ struct value_and_holder {
                    ? inst->simple_holder_constructed
                    : (inst->nonsimple.status[index] & instance::status_holder_constructed) != 0u;
     }
-    void set_holder_constructed(bool v = true) const {
+    // NOLINTNEXTLINE(readability-make-member-function-const)
+    void set_holder_constructed(bool v = true) {
         if (inst->simple_layout)
             inst->simple_holder_constructed = v;
         else if (v)
@@ -254,7 +275,8 @@ struct value_and_holder {
             ? inst->simple_instance_registered
             : ((inst->nonsimple.status[index] & instance::status_instance_registered) != 0);
     }
-    void set_instance_registered(bool v = true) const {
+    // NOLINTNEXTLINE(readability-make-member-function-const)
+    void set_instance_registered(bool v = true) {
         if (inst->simple_layout)
             inst->simple_instance_registered = v;
         else if (v)
@@ -272,7 +294,8 @@ private:
     const type_vec &tinfo;
 
 public:
-    values_and_holders(instance *inst) : inst{inst}, tinfo(all_type_info(Py_TYPE(inst))) {}
+    explicit values_and_holders(instance *inst)
+        : inst{inst}, tinfo(all_type_info(Py_TYPE(inst))) {}
 
     struct iterator {
     private:
@@ -288,7 +311,8 @@ public:
                  0 /* index */)
         {}
         // Past-the-end iterator:
-        iterator(size_t end) : curr(end) {}
+        explicit iterator(size_t end) : curr(end) {}
+
     public:
         bool operator==(const iterator &other) const { return curr.index == other.curr.index; }
         bool operator!=(const iterator &other) const { return curr.index != other.curr.index; }
@@ -325,7 +349,7 @@ public:
  * The returned object should be short-lived: in particular, it must not outlive the called-upon
  * instance.
  */
-PYBIND11_NOINLINE inline value_and_holder instance::get_value_and_holder(const type_info *find_type /*= nullptr default in common.h*/, bool throw_if_missing /*= true in common.h*/) {
+PYBIND11_NOINLINE value_and_holder instance::get_value_and_holder(const type_info *find_type /*= nullptr default in common.h*/, bool throw_if_missing /*= true in common.h*/) {
     // Optimize common case:
     if (!find_type || Py_TYPE(this) == find_type->type)
         return value_and_holder(this, find_type, 0, 0);
@@ -349,7 +373,7 @@ PYBIND11_NOINLINE inline value_and_holder instance::get_value_and_holder(const t
 #endif
 }
 
-PYBIND11_NOINLINE inline void instance::allocate_layout() {
+PYBIND11_NOINLINE void instance::allocate_layout() {
     auto &tinfo = all_type_info(Py_TYPE(this));
 
     const size_t n_types = tinfo.size();
@@ -397,19 +421,20 @@ PYBIND11_NOINLINE inline void instance::allocate_layout() {
     owned = true;
 }
 
-PYBIND11_NOINLINE inline void instance::deallocate_layout() const {
+// NOLINTNEXTLINE(readability-make-member-function-const)
+PYBIND11_NOINLINE void instance::deallocate_layout() {
     if (!simple_layout)
         PyMem_Free(nonsimple.values_and_holders);
 }
 
-PYBIND11_NOINLINE inline bool isinstance_generic(handle obj, const std::type_info &tp) {
+PYBIND11_NOINLINE bool isinstance_generic(handle obj, const std::type_info &tp) {
     handle type = detail::get_type_handle(tp, false);
     if (!type)
         return false;
     return isinstance(obj, type);
 }
 
-PYBIND11_NOINLINE inline std::string error_string() {
+PYBIND11_NOINLINE std::string error_string() {
     if (!PyErr_Occurred()) {
         PyErr_SetString(PyExc_RuntimeError, "Unknown internal error occurred");
         return "Unknown internal error occurred";
@@ -456,7 +481,7 @@ PYBIND11_NOINLINE inline std::string error_string() {
     return errorString;
 }
 
-PYBIND11_NOINLINE inline handle get_object_handle(const void *ptr, const detail::type_info *type ) {
+PYBIND11_NOINLINE handle get_object_handle(const void *ptr, const detail::type_info *type ) {
     auto &instances = get_internals().registered_instances;
     auto range = instances.equal_range(ptr);
     for (auto it = range.first; it != range.second; ++it) {
@@ -483,16 +508,16 @@ inline PyThreadState *get_thread_state_unchecked() {
 }
 
 // Forward declarations
-inline void keep_alive_impl(handle nurse, handle patient);
+void keep_alive_impl(handle nurse, handle patient);
 inline PyObject *make_new_instance(PyTypeObject *type);
 
 class type_caster_generic {
 public:
-    PYBIND11_NOINLINE type_caster_generic(const std::type_info &type_info)
-        : typeinfo(get_type_info(type_info)), cpptype(&type_info) { }
+    PYBIND11_NOINLINE explicit type_caster_generic(const std::type_info &type_info)
+        : typeinfo(get_type_info(type_info)), cpptype(&type_info) {}
 
-    type_caster_generic(const type_info *typeinfo)
-        : typeinfo(typeinfo), cpptype(typeinfo ? typeinfo->cpptype : nullptr) { }
+    explicit type_caster_generic(const type_info *typeinfo)
+        : typeinfo(typeinfo), cpptype(typeinfo ? typeinfo->cpptype : nullptr) {}
 
     bool load(handle src, bool convert) {
         return load_impl<type_caster_generic>(src, convert);
@@ -920,23 +945,26 @@ public:
 
     template <typename T> using cast_op_type = detail::cast_op_type<T>;
 
+    // NOLINTNEXTLINE(google-explicit-constructor)
     operator itype*() { return (type *) value; }
+    // NOLINTNEXTLINE(google-explicit-constructor)
     operator itype&() { if (!value) throw reference_cast_error(); return *((itype *) value); }
 
 protected:
     using Constructor = void *(*)(const void *);
 
     /* Only enabled when the types are {copy,move}-constructible *and* when the type
-       does not have a private operator new implementation. */
+       does not have a private operator new implementation. A comma operator is used in the decltype
+       argument to apply SFINAE to the public copy/move constructors.*/
     template <typename T, typename = enable_if_t<is_copy_constructible<T>::value>>
-    static auto make_copy_constructor(const T *x) -> decltype(new T(*x), Constructor{}) {
+    static auto make_copy_constructor(const T *) -> decltype(new T(std::declval<const T>()), Constructor{}) {
         return [](const void *arg) -> void * {
             return new T(*reinterpret_cast<const T *>(arg));
         };
     }
 
     template <typename T, typename = enable_if_t<std::is_move_constructible<T>::value>>
-    static auto make_move_constructor(const T *x) -> decltype(new T(std::move(*const_cast<T *>(x))), Constructor{}) {
+    static auto make_move_constructor(const T *) -> decltype(new T(std::declval<T&&>()), Constructor{}) {
         return [](const void *arg) -> void * {
             return new T(std::move(*const_cast<T *>(reinterpret_cast<const T *>(arg))));
         };

include/pybind11/detail/typeid.h

@@ -29,7 +29,7 @@ inline void erase_all(std::string &string, const std::string &search) {
     }
 }
 
-PYBIND11_NOINLINE inline void clean_type_id(std::string &name) {
+PYBIND11_NOINLINE void clean_type_id(std::string &name) {
 #if defined(__GNUG__)
     int status = 0;
     std::unique_ptr<char, void (*)(void *)> res {