From 7e3af05d99a67da1bb5a487952a76813f2d5ccfc Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Mon, 2 Aug 2021 12:41:47 -0700
Subject: [PATCH 01/18] Removing pragma for GCC -Wattributes, fixing forward
 declarations.

---
 include/pybind11/detail/common.h |  7 +------
 include/pybind11/pybind11.h      | 10 ----------
 2 files changed, 1 insertion(+), 16 deletions(-)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index cc24a52fd3..c68ade2a45 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -99,12 +99,7 @@
 #  endif
 #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)
+#if defined(_MSC_VER)
 #  define PYBIND11_NOINLINE __declspec(noinline) inline
 #else
 #  define PYBIND11_NOINLINE __attribute__ ((noinline)) inline
diff --git a/include/pybind11/pybind11.h b/include/pybind11/pybind11.h
index 20df6a01fd..a337017ed8 100644
--- a/include/pybind11/pybind11.h
+++ b/include/pybind11/pybind11.h
@@ -10,11 +10,6 @@
 
 #pragma once
 
-#if defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8))
-#  pragma GCC diagnostic push
-#  pragma GCC diagnostic ignored "-Wattributes"
-#endif
-
 #include "attr.h"
 #include "gil.h"
 #include "options.h"
@@ -1874,7 +1869,6 @@ template <typename Type> class enum_ : public class_<Type> {
 
 PYBIND11_NAMESPACE_BEGIN(detail)
 
-
 PYBIND11_NOINLINE void keep_alive_impl(handle nurse, handle patient) {
     if (!nurse || !patient)
         pybind11_fail("Could not activate keep_alive!");
@@ -2383,7 +2377,3 @@ PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)
 #if defined(__GNUC__) && __GNUC__ == 7
 #    pragma GCC diagnostic pop // -Wnoexcept-type
 #endif
-
-#if defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8))
-#  pragma GCC diagnostic pop
-#endif

From 2cb5d65d9c7b60fffeb2e867f8067880ac86965d Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Mon, 2 Aug 2021 13:52:36 -0700
Subject: [PATCH 02/18] Introducing PYBIND11_NOINLINE_FWD to deal with CUDA,
 GCC7, GCC8.

---
 include/pybind11/attr.h                    |  6 ++--
 include/pybind11/detail/common.h           | 15 ++++++---
 include/pybind11/detail/internals.h        |  6 ++--
 include/pybind11/detail/type_caster_base.h | 38 +++++++++++-----------
 include/pybind11/detail/typeid.h           |  2 +-
 include/pybind11/gil.h                     | 10 +++---
 include/pybind11/numpy.h                   |  4 +--
 include/pybind11/pybind11.h                | 16 ++++-----
 include/pybind11/pytypes.h                 |  4 +--
 9 files changed, 53 insertions(+), 48 deletions(-)

diff --git a/include/pybind11/attr.h b/include/pybind11/attr.h
index ab1fe80446..36a4343400 100644
--- a/include/pybind11/attr.h
+++ b/include/pybind11/attr.h
@@ -124,7 +124,7 @@ 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_;
-void keep_alive_impl(size_t Nurse, size_t Patient, function_call &call, handle ret);
+PYBIND11_NOINLINE_FWD 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 {
@@ -220,7 +220,7 @@ struct function_record {
 
 /// Special data structure which (temporarily) holds metadata about a bound class
 struct type_record {
-    PYBIND11_NOINLINE type_record()
+    PYBIND11_NOINLINE_DCL type_record()
         : multiple_inheritance(false), dynamic_attr(false), buffer_protocol(false),
           default_holder(true), module_local(false), is_final(false) { }
 
@@ -278,7 +278,7 @@ struct type_record {
     /// Is the class inheritable from python classes?
     bool is_final : 1;
 
-    PYBIND11_NOINLINE void add_base(const std::type_info &base, void *(*caster)(void *)) {
+    PYBIND11_NOINLINE_DCL void add_base(const std::type_info &base, void *(*caster)(void *)) {
         auto base_info = detail::get_type_info(base, false);
         if (!base_info) {
             std::string tname(base.name());
diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index c68ade2a45..5fdb3cdc8c 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -99,10 +99,15 @@
 #  endif
 #endif
 
-#if defined(_MSC_VER)
-#  define PYBIND11_NOINLINE __declspec(noinline) inline
+#if defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8))
+#  define PYBIND11_NOINLINE_DCL inline
+#  define PYBIND11_NOINLINE_FWD inline
+#elif defined(_MSC_VER)
+#  define PYBIND11_NOINLINE_DCL __declspec(noinline) inline
+#  define PYBIND11_NOINLINE_FWD
 #else
-#  define PYBIND11_NOINLINE __attribute__ ((noinline)) inline
+#  define PYBIND11_NOINLINE_DCL __attribute__ ((noinline)) inline
+#  define PYBIND11_NOINLINE_FWD
 #endif
 
 #if defined(__MINGW32__)
@@ -778,8 +783,8 @@ 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 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); }
+[[noreturn]] PYBIND11_NOINLINE_DCL void pybind11_fail(const char *reason) { throw std::runtime_error(reason); }
+[[noreturn]] PYBIND11_NOINLINE_DCL void pybind11_fail(const std::string &reason) { throw std::runtime_error(reason); }
 
 template <typename T, typename SFINAE = void> struct format_descriptor { };
 
diff --git a/include/pybind11/detail/internals.h b/include/pybind11/detail/internals.h
index b177801a11..eac3079c68 100644
--- a/include/pybind11/detail/internals.h
+++ b/include/pybind11/detail/internals.h
@@ -257,7 +257,7 @@ inline void translate_local_exception(std::exception_ptr p) {
 #endif
 
 /// Return a reference to the current `internals` data
-PYBIND11_NOINLINE internals &get_internals() {
+PYBIND11_NOINLINE_DCL internals &get_internals() {
     auto **&internals_pp = get_internals_pp();
     if (internals_pp && *internals_pp)
         return **internals_pp;
@@ -352,14 +352,14 @@ 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.
-PYBIND11_NOINLINE void *get_shared_data(const std::string &name) {
+PYBIND11_NOINLINE_DCL 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()`.
-PYBIND11_NOINLINE void *set_shared_data(const std::string &name, void *data) {
+PYBIND11_NOINLINE_DCL void *set_shared_data(const std::string &name, void *data) {
     detail::get_internals().shared_data[name] = data;
     return data;
 }
diff --git a/include/pybind11/detail/type_caster_base.h b/include/pybind11/detail/type_caster_base.h
index 14cb27cad2..eae5a12f4a 100644
--- a/include/pybind11/detail/type_caster_base.h
+++ b/include/pybind11/detail/type_caster_base.h
@@ -54,7 +54,7 @@ class loader_life_support {
 
     /// 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) {
+    PYBIND11_NOINLINE_DCL static void add_patient(handle h) {
         auto &stack = get_internals().loader_patient_stack;
         if (stack.empty())
             throw cast_error("When called outside a bound function, py::cast() cannot "
@@ -81,7 +81,7 @@ class loader_life_support {
 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 void all_type_info_populate(PyTypeObject *t, std::vector<type_info *> &bases) {
+PYBIND11_NOINLINE_DCL 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 +150,7 @@ 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 detail::type_info* get_type_info(PyTypeObject *type) {
+PYBIND11_NOINLINE_DCL detail::type_info* get_type_info(PyTypeObject *type) {
     auto &bases = all_type_info(type);
     if (bases.empty())
         return nullptr;
@@ -176,7 +176,7 @@ 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 detail::type_info *get_type_info(const std::type_index &tp,
+PYBIND11_NOINLINE_DCL 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 +191,13 @@ PYBIND11_NOINLINE detail::type_info *get_type_info(const std::type_index &tp,
     return nullptr;
 }
 
-PYBIND11_NOINLINE handle get_type_handle(const std::type_info &tp, bool throw_if_missing) {
+PYBIND11_NOINLINE_DCL 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 handle find_registered_python_instance(void *src,
+PYBIND11_NOINLINE_DCL 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) {
@@ -325,7 +325,7 @@ struct values_and_holders {
  * The returned object should be short-lived: in particular, it must not outlive the called-upon
  * instance.
  */
-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*/) {
+PYBIND11_NOINLINE_DCL 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 +349,7 @@ PYBIND11_NOINLINE value_and_holder instance::get_value_and_holder(const type_inf
 #endif
 }
 
-PYBIND11_NOINLINE void instance::allocate_layout() {
+PYBIND11_NOINLINE_DCL void instance::allocate_layout() {
     auto &tinfo = all_type_info(Py_TYPE(this));
 
     const size_t n_types = tinfo.size();
@@ -397,19 +397,19 @@ PYBIND11_NOINLINE void instance::allocate_layout() {
     owned = true;
 }
 
-PYBIND11_NOINLINE void instance::deallocate_layout() const {
+PYBIND11_NOINLINE_DCL void instance::deallocate_layout() const {
     if (!simple_layout)
         PyMem_Free(nonsimple.values_and_holders);
 }
 
-PYBIND11_NOINLINE bool isinstance_generic(handle obj, const std::type_info &tp) {
+PYBIND11_NOINLINE_DCL 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 std::string error_string() {
+PYBIND11_NOINLINE_DCL std::string error_string() {
     if (!PyErr_Occurred()) {
         PyErr_SetString(PyExc_RuntimeError, "Unknown internal error occurred");
         return "Unknown internal error occurred";
@@ -456,7 +456,7 @@ PYBIND11_NOINLINE std::string error_string() {
     return errorString;
 }
 
-PYBIND11_NOINLINE handle get_object_handle(const void *ptr, const detail::type_info *type ) {
+PYBIND11_NOINLINE_DCL 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) {
@@ -483,12 +483,12 @@ inline PyThreadState *get_thread_state_unchecked() {
 }
 
 // Forward declarations
-void keep_alive_impl(handle nurse, handle patient);
+PYBIND11_NOINLINE_FWD 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)
+    PYBIND11_NOINLINE_DCL 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)
@@ -498,7 +498,7 @@ class type_caster_generic {
         return load_impl<type_caster_generic>(src, convert);
     }
 
-    PYBIND11_NOINLINE static handle cast(const void *_src, return_value_policy policy, handle parent,
+    PYBIND11_NOINLINE_DCL static handle cast(const void *_src, return_value_policy policy, handle parent,
                                          const detail::type_info *tinfo,
                                          void *(*copy_constructor)(const void *),
                                          void *(*move_constructor)(const void *),
@@ -622,7 +622,7 @@ class type_caster_generic {
     }
     void check_holder_compat() {}
 
-    PYBIND11_NOINLINE static void *local_load(PyObject *src, const type_info *ti) {
+    PYBIND11_NOINLINE_DCL static void *local_load(PyObject *src, const type_info *ti) {
         auto caster = type_caster_generic(ti);
         if (caster.load(src, false))
             return caster.value;
@@ -631,7 +631,7 @@ class type_caster_generic {
 
     /// Try to load with foreign typeinfo, if available. Used when there is no
     /// native typeinfo, or when the native one wasn't able to produce a value.
-    PYBIND11_NOINLINE bool try_load_foreign_module_local(handle src) {
+    PYBIND11_NOINLINE_DCL bool try_load_foreign_module_local(handle src) {
         constexpr auto *local_key = PYBIND11_MODULE_LOCAL_ID;
         const auto pytype = type::handle_of(src);
         if (!hasattr(pytype, local_key))
@@ -654,7 +654,7 @@ class type_caster_generic {
     // bits of code between here and copyable_holder_caster where the two classes need different
     // logic (without having to resort to virtual inheritance).
     template <typename ThisT>
-    PYBIND11_NOINLINE bool load_impl(handle src, bool convert) {
+    PYBIND11_NOINLINE_DCL bool load_impl(handle src, bool convert) {
         if (!src) return false;
         if (!typeinfo) return try_load_foreign_module_local(src);
 
@@ -744,7 +744,7 @@ class type_caster_generic {
     // Called to do type lookup and wrap the pointer and type in a pair when a dynamic_cast
     // isn't needed or can't be used.  If the type is unknown, sets the error and returns a pair
     // with .second = nullptr.  (p.first = nullptr is not an error: it becomes None).
-    PYBIND11_NOINLINE static std::pair<const void *, const type_info *> src_and_type(
+    PYBIND11_NOINLINE_DCL static std::pair<const void *, const type_info *> src_and_type(
             const void *src, const std::type_info &cast_type, const std::type_info *rtti_type = nullptr) {
         if (auto *tpi = get_type_info(cast_type))
             return {src, const_cast<const type_info *>(tpi)};
diff --git a/include/pybind11/detail/typeid.h b/include/pybind11/detail/typeid.h
index 39ba8ce0f7..2d5b8cefdf 100644
--- a/include/pybind11/detail/typeid.h
+++ b/include/pybind11/detail/typeid.h
@@ -29,7 +29,7 @@ inline void erase_all(std::string &string, const std::string &search) {
     }
 }
 
-PYBIND11_NOINLINE void clean_type_id(std::string &name) {
+PYBIND11_NOINLINE_DCL void clean_type_id(std::string &name) {
 #if defined(__GNUG__)
     int status = 0;
     std::unique_ptr<char, void (*)(void *)> res {
diff --git a/include/pybind11/gil.h b/include/pybind11/gil.h
index 32f1a8963b..25fa1c93c2 100644
--- a/include/pybind11/gil.h
+++ b/include/pybind11/gil.h
@@ -49,7 +49,7 @@ PYBIND11_NAMESPACE_END(detail)
 
 class gil_scoped_acquire {
 public:
-    PYBIND11_NOINLINE gil_scoped_acquire() {
+    PYBIND11_NOINLINE_DCL gil_scoped_acquire() {
         auto const &internals = detail::get_internals();
         tstate = (PyThreadState *) PYBIND11_TLS_GET_VALUE(internals.tstate);
 
@@ -85,7 +85,7 @@ class gil_scoped_acquire {
         ++tstate->gilstate_counter;
     }
 
-    PYBIND11_NOINLINE void dec_ref() {
+    PYBIND11_NOINLINE_DCL void dec_ref() {
         --tstate->gilstate_counter;
         #if !defined(NDEBUG)
             if (detail::get_thread_state_unchecked() != tstate)
@@ -111,11 +111,11 @@ class gil_scoped_acquire {
     /// could be shutting down when this is called, as thread deletion is not
     /// allowed during shutdown. Check _Py_IsFinalizing() on Python 3.7+, and
     /// protect subsequent code.
-    PYBIND11_NOINLINE void disarm() {
+    PYBIND11_NOINLINE_DCL void disarm() {
         active = false;
     }
 
-    PYBIND11_NOINLINE ~gil_scoped_acquire() {
+    PYBIND11_NOINLINE_DCL ~gil_scoped_acquire() {
         dec_ref();
         if (release)
            PyEval_SaveThread();
@@ -145,7 +145,7 @@ class gil_scoped_release {
     /// could be shutting down when this is called, as thread deletion is not
     /// allowed during shutdown. Check _Py_IsFinalizing() on Python 3.7+, and
     /// protect subsequent code.
-    PYBIND11_NOINLINE void disarm() {
+    PYBIND11_NOINLINE_DCL void disarm() {
         active = false;
     }
 
diff --git a/include/pybind11/numpy.h b/include/pybind11/numpy.h
index 2dc0f29f0b..31597f7330 100644
--- a/include/pybind11/numpy.h
+++ b/include/pybind11/numpy.h
@@ -104,7 +104,7 @@ struct numpy_internals {
     }
 };
 
-PYBIND11_NOINLINE void load_numpy_internals(numpy_internals* &ptr) {
+PYBIND11_NOINLINE_DCL void load_numpy_internals(numpy_internals* &ptr) {
     ptr = &get_or_create_shared_data<numpy_internals>("_numpy_internals");
 }
 
@@ -1110,7 +1110,7 @@ struct field_descriptor {
     dtype descr;
 };
 
-PYBIND11_NOINLINE void register_structured_dtype(
+PYBIND11_NOINLINE_DCL void register_structured_dtype(
     any_container<field_descriptor> fields,
     const std::type_info& tinfo, ssize_t itemsize,
     bool (*direct_converter)(PyObject *, void *&)) {
diff --git a/include/pybind11/pybind11.h b/include/pybind11/pybind11.h
index a337017ed8..4ac827fdcf 100644
--- a/include/pybind11/pybind11.h
+++ b/include/pybind11/pybind11.h
@@ -141,7 +141,7 @@ class cpp_function : public function {
     using unique_function_record = std::unique_ptr<detail::function_record, InitializingFunctionRecordDeleter>;
 
     /// Space optimization: don't inline this frequently instantiated fragment
-    PYBIND11_NOINLINE unique_function_record make_function_record() {
+    PYBIND11_NOINLINE_DCL unique_function_record make_function_record() {
         return unique_function_record(new detail::function_record());
     }
 
@@ -1051,7 +1051,7 @@ class module_ : public object {
         ``overwrite`` should almost always be false: attempting to overwrite objects that pybind11 has
         established will, in most cases, break things.
     \endrst */
-    PYBIND11_NOINLINE void add_object(const char *name, handle obj, bool overwrite = false) {
+    PYBIND11_NOINLINE_DCL void add_object(const char *name, handle obj, bool overwrite = false) {
         if (!overwrite && hasattr(*this, name))
             pybind11_fail("Error during initialization: multiple incompatible definitions with name \"" +
                     std::string(name) + "\"");
@@ -1666,7 +1666,7 @@ inline str enum_name(handle arg) {
 struct enum_base {
     enum_base(handle base, handle parent) : m_base(base), m_parent(parent) { }
 
-    PYBIND11_NOINLINE void init(bool is_arithmetic, bool is_convertible) {
+    PYBIND11_NOINLINE_DCL void init(bool is_arithmetic, bool is_convertible) {
         m_base.attr("__entries") = dict();
         auto property = handle((PyObject *) &PyProperty_Type);
         auto static_property = handle((PyObject *) get_internals().static_property_type);
@@ -1792,7 +1792,7 @@ struct enum_base {
             [](const object &arg) { return int_(arg); }, name("__hash__"), is_method(m_base));
     }
 
-    PYBIND11_NOINLINE void value(char const* name_, object value, const char *doc = nullptr) {
+    PYBIND11_NOINLINE_DCL void value(char const* name_, object value, const char *doc = nullptr) {
         dict entries = m_base.attr("__entries");
         str name(name_);
         if (entries.contains(name)) {
@@ -1804,7 +1804,7 @@ struct enum_base {
         m_base.attr(name) = value;
     }
 
-    PYBIND11_NOINLINE void export_values() {
+    PYBIND11_NOINLINE_DCL void export_values() {
         dict entries = m_base.attr("__entries");
         for (auto kv : entries)
             m_parent.attr(kv.first) = kv.second[int_(0)];
@@ -1869,7 +1869,7 @@ template <typename Type> class enum_ : public class_<Type> {
 
 PYBIND11_NAMESPACE_BEGIN(detail)
 
-PYBIND11_NOINLINE void keep_alive_impl(handle nurse, handle patient) {
+PYBIND11_NOINLINE_DCL void keep_alive_impl(handle nurse, handle patient) {
     if (!nurse || !patient)
         pybind11_fail("Could not activate keep_alive!");
 
@@ -1896,7 +1896,7 @@ PYBIND11_NOINLINE void keep_alive_impl(handle nurse, handle patient) {
     }
 }
 
-PYBIND11_NOINLINE void keep_alive_impl(size_t Nurse, size_t Patient, function_call &call, handle ret) {
+PYBIND11_NOINLINE_DCL void keep_alive_impl(size_t Nurse, size_t Patient, function_call &call, handle ret) {
     auto get_arg = [&](size_t n) {
         if (n == 0)
             return ret;
@@ -2146,7 +2146,7 @@ exception<CppException> &register_local_exception(handle scope,
 }
 
 PYBIND11_NAMESPACE_BEGIN(detail)
-PYBIND11_NOINLINE void print(const tuple &args, const dict &kwargs) {
+PYBIND11_NOINLINE_DCL void print(const tuple &args, const dict &kwargs) {
     auto strings = tuple(args.size());
     for (size_t i = 0; i < args.size(); ++i) {
         strings[i] = str(args[i]);
diff --git a/include/pybind11/pytypes.h b/include/pybind11/pytypes.h
index 21e14960cf..f07f52f815 100644
--- a/include/pybind11/pytypes.h
+++ b/include/pybind11/pytypes.h
@@ -24,7 +24,7 @@ struct arg; struct arg_v;
 
 PYBIND11_NAMESPACE_BEGIN(detail)
 class args_proxy;
-bool isinstance_generic(handle obj, const std::type_info &tp);
+PYBIND11_NOINLINE_FWD bool isinstance_generic(handle obj, const std::type_info &tp);
 
 // Accessor forward declarations
 template <typename Policy> class accessor;
@@ -316,7 +316,7 @@ template <typename T> T reinterpret_borrow(handle h) { return {h, object::borrow
 template <typename T> T reinterpret_steal(handle h) { return {h, object::stolen_t{}}; }
 
 PYBIND11_NAMESPACE_BEGIN(detail)
-std::string error_string();
+PYBIND11_NOINLINE_FWD std::string error_string();
 PYBIND11_NAMESPACE_END(detail)
 
 #if defined(_MSC_VER)

From 43c6f22ca6934e346358b2d4a353e833ba09247a Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Mon, 2 Aug 2021 14:27:49 -0700
Subject: [PATCH 03/18] Updating PYBIND11_NOINLINE_DCL in Doxyfile.

---
 docs/Doxyfile | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/docs/Doxyfile b/docs/Doxyfile
index c8562952ef..22af784890 100644
--- a/docs/Doxyfile
+++ b/docs/Doxyfile
@@ -20,4 +20,4 @@ WARNINGS               = YES
 WARN_IF_UNDOCUMENTED   = NO
 PREDEFINED             = DOXYGEN_SHOULD_SKIP_THIS \
                          PY_MAJOR_VERSION=3 \
-                         PYBIND11_NOINLINE
+                         PYBIND11_NOINLINE_DCL

From d564545eb351696d59a8af422e831ee3de6faecf Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Mon, 2 Aug 2021 15:09:21 -0700
Subject: [PATCH 04/18] Trying noinline, noinline for {CUDA, GCC7, GCC8}

---
 include/pybind11/detail/common.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index 5fdb3cdc8c..f4e3a7c43e 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -100,8 +100,8 @@
 #endif
 
 #if defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8))
-#  define PYBIND11_NOINLINE_DCL inline
-#  define PYBIND11_NOINLINE_FWD inline
+#  define PYBIND11_NOINLINE_DCL __attribute__ ((noinline)) inline
+#  define PYBIND11_NOINLINE_FWD __attribute__ ((noinline)) inline
 #elif defined(_MSC_VER)
 #  define PYBIND11_NOINLINE_DCL __declspec(noinline) inline
 #  define PYBIND11_NOINLINE_FWD

From 4984ab5bfe8c7986b3360302e24d6d4dd88733e3 Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Mon, 2 Aug 2021 15:15:09 -0700
Subject: [PATCH 05/18] Trying noinline, inline for {CUDA, GCC7, GCC8}

---
 include/pybind11/detail/common.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index f4e3a7c43e..fdb174f344 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -101,7 +101,7 @@
 
 #if defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8))
 #  define PYBIND11_NOINLINE_DCL __attribute__ ((noinline)) inline
-#  define PYBIND11_NOINLINE_FWD __attribute__ ((noinline)) inline
+#  define PYBIND11_NOINLINE_FWD inline
 #elif defined(_MSC_VER)
 #  define PYBIND11_NOINLINE_DCL __declspec(noinline) inline
 #  define PYBIND11_NOINLINE_FWD

From aef2edbbbc1dd13ba06114e03d37597cdf7e4289 Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Mon, 2 Aug 2021 15:46:01 -0700
Subject: [PATCH 06/18] Adding GCC -Wattributes `pragma` in 3 header files.

---
 include/pybind11/attr.h                    | 12 ++++++++++++
 include/pybind11/detail/common.h           |  4 ++--
 include/pybind11/detail/type_caster_base.h | 14 +++++++++++++-
 include/pybind11/pytypes.h                 | 18 ++++++++++++++----
 4 files changed, 41 insertions(+), 7 deletions(-)

diff --git a/include/pybind11/attr.h b/include/pybind11/attr.h
index 36a4343400..42f67c8d1f 100644
--- a/include/pybind11/attr.h
+++ b/include/pybind11/attr.h
@@ -124,8 +124,20 @@ 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_;
+
+#if defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8))
+#  define PYBIND11_PRAGMA_NOINLINE_FWD
+#  pragma GCC diagnostic push
+#  pragma GCC diagnostic ignored "-Wattributes"
+#endif
+
 PYBIND11_NOINLINE_FWD void keep_alive_impl(size_t Nurse, size_t Patient, function_call &call, handle ret);
 
+#if defined(PYBIND11_PRAGMA_NOINLINE_FWD)
+#  pragma GCC diagnostic pop
+#  undef PYBIND11_PRAGMA_NOINLINE_FWD
+#endif
+
 /// Internal data structure which holds metadata about a keyword argument
 struct argument_record {
     const char *name;  ///< Argument name
diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index fdb174f344..401d03cb4e 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -99,8 +99,8 @@
 #  endif
 #endif
 
-#if defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8))
-#  define PYBIND11_NOINLINE_DCL __attribute__ ((noinline)) inline
+#if defined(PYBIND11_NOINLINE_DISABLED)
+#  define PYBIND11_NOINLINE_DCL inline
 #  define PYBIND11_NOINLINE_FWD inline
 #elif defined(_MSC_VER)
 #  define PYBIND11_NOINLINE_DCL __declspec(noinline) inline
diff --git a/include/pybind11/detail/type_caster_base.h b/include/pybind11/detail/type_caster_base.h
index eae5a12f4a..cce7a90df3 100644
--- a/include/pybind11/detail/type_caster_base.h
+++ b/include/pybind11/detail/type_caster_base.h
@@ -483,9 +483,21 @@ inline PyThreadState *get_thread_state_unchecked() {
 }
 
 // Forward declarations
-PYBIND11_NOINLINE_FWD void keep_alive_impl(handle nurse, handle patient);
 inline PyObject *make_new_instance(PyTypeObject *type);
 
+#if defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8))
+#  define PYBIND11_PRAGMA_NOINLINE_FWD
+#  pragma GCC diagnostic push
+#  pragma GCC diagnostic ignored "-Wattributes"
+#endif
+
+PYBIND11_NOINLINE_FWD void keep_alive_impl(handle nurse, handle patient);
+
+#if defined(PYBIND11_PRAGMA_NOINLINE_FWD)
+#  pragma GCC diagnostic pop
+#  undef PYBIND11_PRAGMA_NOINLINE_FWD
+#endif
+
 class type_caster_generic {
 public:
     PYBIND11_NOINLINE_DCL type_caster_generic(const std::type_info &type_info)
diff --git a/include/pybind11/pytypes.h b/include/pybind11/pytypes.h
index f07f52f815..54e9ed28a5 100644
--- a/include/pybind11/pytypes.h
+++ b/include/pybind11/pytypes.h
@@ -23,8 +23,22 @@ class type;
 struct arg; struct arg_v;
 
 PYBIND11_NAMESPACE_BEGIN(detail)
+
 class args_proxy;
+
+#if defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8))
+#  define PYBIND11_PRAGMA_NOINLINE_FWD
+#  pragma GCC diagnostic push
+#  pragma GCC diagnostic ignored "-Wattributes"
+#endif
+
 PYBIND11_NOINLINE_FWD bool isinstance_generic(handle obj, const std::type_info &tp);
+PYBIND11_NOINLINE_FWD std::string error_string();
+
+#if defined(PYBIND11_PRAGMA_NOINLINE_FWD)
+#  pragma GCC diagnostic pop
+#  undef PYBIND11_PRAGMA_NOINLINE_FWD
+#endif
 
 // Accessor forward declarations
 template <typename Policy> class accessor;
@@ -315,10 +329,6 @@ template <typename T> T reinterpret_borrow(handle h) { return {h, object::borrow
 \endrst */
 template <typename T> T reinterpret_steal(handle h) { return {h, object::stolen_t{}}; }
 
-PYBIND11_NAMESPACE_BEGIN(detail)
-PYBIND11_NOINLINE_FWD std::string error_string();
-PYBIND11_NAMESPACE_END(detail)
-
 #if defined(_MSC_VER)
 #  pragma warning(push)
 #  pragma warning(disable: 4275 4251) // warning C4275: An exported class was derived from a class that wasn't exported. Can be ignored when derived from a STL class.

From 7181a118cb5f23854c8f19e7d07791990840f57a Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Mon, 2 Aug 2021 16:37:09 -0700
Subject: [PATCH 07/18] Introducing
 PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED, used in 9 header files.

---
 include/pybind11/attr.h                    | 21 +++++++++-----------
 include/pybind11/detail/common.h           | 16 +++++++++++++++
 include/pybind11/detail/internals.h        |  9 +++++++++
 include/pybind11/detail/type_caster_base.h | 23 ++++++++++------------
 include/pybind11/detail/typeid.h           |  9 +++++++++
 include/pybind11/gil.h                     | 10 +++++++++-
 include/pybind11/numpy.h                   |  9 +++++++++
 include/pybind11/pybind11.h                |  9 +++++++++
 include/pybind11/pytypes.h                 | 21 +++++++++-----------
 9 files changed, 89 insertions(+), 38 deletions(-)

diff --git a/include/pybind11/attr.h b/include/pybind11/attr.h
index 42f67c8d1f..2a88ee1dc6 100644
--- a/include/pybind11/attr.h
+++ b/include/pybind11/attr.h
@@ -12,6 +12,11 @@
 
 #include "cast.h"
 
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic push
+#  pragma GCC diagnostic ignored "-Wattributes"
+#endif
+
 PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 
 /// \addtogroup annotations
@@ -124,20 +129,8 @@ 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_;
-
-#if defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8))
-#  define PYBIND11_PRAGMA_NOINLINE_FWD
-#  pragma GCC diagnostic push
-#  pragma GCC diagnostic ignored "-Wattributes"
-#endif
-
 PYBIND11_NOINLINE_FWD void keep_alive_impl(size_t Nurse, size_t Patient, function_call &call, handle ret);
 
-#if defined(PYBIND11_PRAGMA_NOINLINE_FWD)
-#  pragma GCC diagnostic pop
-#  undef PYBIND11_PRAGMA_NOINLINE_FWD
-#endif
-
 /// Internal data structure which holds metadata about a keyword argument
 struct argument_record {
     const char *name;  ///< Argument name
@@ -574,3 +567,7 @@ constexpr bool expected_num_args(size_t nargs, bool has_args, bool has_kwargs) {
 
 PYBIND11_NAMESPACE_END(detail)
 PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)
+
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic pop
+#endif
diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index 401d03cb4e..5c0c785e00 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -110,6 +110,13 @@
 #  define PYBIND11_NOINLINE_FWD
 #endif
 
+#if !defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED) && \
+    !defined(PYBIND11_NOINLINE_DISABLED) && \
+    (defined(__CUDACC__) || \
+     (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8)))
+#  define PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED
+#endif
+
 #if defined(__MINGW32__)
 // For unknown reasons all PYBIND11_DEPRECATED member trigger a warning when declared
 // whether it is used or not
@@ -379,6 +386,11 @@ extern "C" {
     }                                                                                             \
     void PYBIND11_CONCAT(pybind11_init_, name)(::pybind11::module_ & (variable))
 
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic push
+#  pragma GCC diagnostic ignored "-Wattributes"
+#endif
+
 PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 
 using ssize_t = Py_ssize_t;
@@ -965,3 +977,7 @@ constexpr inline bool silence_msvc_c4127(bool cond) { return cond; }
 
 PYBIND11_NAMESPACE_END(detail)
 PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)
+
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic pop
+#endif
diff --git a/include/pybind11/detail/internals.h b/include/pybind11/detail/internals.h
index eac3079c68..83ced63a3b 100644
--- a/include/pybind11/detail/internals.h
+++ b/include/pybind11/detail/internals.h
@@ -11,6 +11,11 @@
 
 #include "../pytypes.h"
 
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic push
+#  pragma GCC diagnostic ignored "-Wattributes"
+#endif
+
 PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 
 using ExceptionTranslator = void (*)(std::exception_ptr);
@@ -380,3 +385,7 @@ T &get_or_create_shared_data(const std::string &name) {
 }
 
 PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)
+
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic pop
+#endif
diff --git a/include/pybind11/detail/type_caster_base.h b/include/pybind11/detail/type_caster_base.h
index cce7a90df3..f2f72106bf 100644
--- a/include/pybind11/detail/type_caster_base.h
+++ b/include/pybind11/detail/type_caster_base.h
@@ -25,6 +25,11 @@
 #include <utility>
 #include <vector>
 
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic push
+#  pragma GCC diagnostic ignored "-Wattributes"
+#endif
+
 PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 PYBIND11_NAMESPACE_BEGIN(detail)
 
@@ -483,20 +488,8 @@ inline PyThreadState *get_thread_state_unchecked() {
 }
 
 // Forward declarations
-inline PyObject *make_new_instance(PyTypeObject *type);
-
-#if defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8))
-#  define PYBIND11_PRAGMA_NOINLINE_FWD
-#  pragma GCC diagnostic push
-#  pragma GCC diagnostic ignored "-Wattributes"
-#endif
-
 PYBIND11_NOINLINE_FWD void keep_alive_impl(handle nurse, handle patient);
-
-#if defined(PYBIND11_PRAGMA_NOINLINE_FWD)
-#  pragma GCC diagnostic pop
-#  undef PYBIND11_PRAGMA_NOINLINE_FWD
-#endif
+inline PyObject *make_new_instance(PyTypeObject *type);
 
 class type_caster_generic {
 public:
@@ -961,3 +954,7 @@ template <typename type> class type_caster_base : public type_caster_generic {
 
 PYBIND11_NAMESPACE_END(detail)
 PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)
+
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic pop
+#endif
diff --git a/include/pybind11/detail/typeid.h b/include/pybind11/detail/typeid.h
index 2d5b8cefdf..bb24d9b615 100644
--- a/include/pybind11/detail/typeid.h
+++ b/include/pybind11/detail/typeid.h
@@ -18,6 +18,11 @@
 
 #include "common.h"
 
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic push
+#  pragma GCC diagnostic ignored "-Wattributes"
+#endif
+
 PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 PYBIND11_NAMESPACE_BEGIN(detail)
 /// Erase all occurrences of a substring
@@ -53,3 +58,7 @@ template <typename T> static std::string type_id() {
 }
 
 PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)
+
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic pop
+#endif
diff --git a/include/pybind11/gil.h b/include/pybind11/gil.h
index 25fa1c93c2..ac0bcdf3bc 100644
--- a/include/pybind11/gil.h
+++ b/include/pybind11/gil.h
@@ -12,8 +12,12 @@
 #include "detail/common.h"
 #include "detail/internals.h"
 
-PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic push
+#  pragma GCC diagnostic ignored "-Wattributes"
+#endif
 
+PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 
 PYBIND11_NAMESPACE_BEGIN(detail)
 
@@ -191,3 +195,7 @@ class gil_scoped_release {
 #endif
 
 PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)
+
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic pop
+#endif
diff --git a/include/pybind11/numpy.h b/include/pybind11/numpy.h
index 31597f7330..f295dfa51a 100644
--- a/include/pybind11/numpy.h
+++ b/include/pybind11/numpy.h
@@ -30,6 +30,11 @@
 #  pragma warning(disable: 4127) // warning C4127: Conditional expression is constant
 #endif
 
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic push
+#  pragma GCC diagnostic ignored "-Wattributes"
+#endif
+
 /* This will be true on all flat address space platforms and allows us to reduce the
    whole npy_intp / ssize_t / Py_intptr_t business down to just ssize_t for all size
    and dimension types (e.g. shape, strides, indexing), instead of inflicting this
@@ -1709,6 +1714,10 @@ Helper vectorize(Return (Class::*f)(Args...) const) {
 
 PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)
 
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic pop
+#endif
+
 #if defined(_MSC_VER)
 #pragma warning(pop)
 #endif
diff --git a/include/pybind11/pybind11.h b/include/pybind11/pybind11.h
index 4ac827fdcf..9d9140aec0 100644
--- a/include/pybind11/pybind11.h
+++ b/include/pybind11/pybind11.h
@@ -47,6 +47,11 @@
 #    pragma GCC diagnostic ignored "-Wnoexcept-type"
 #endif
 
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic push
+#  pragma GCC diagnostic ignored "-Wattributes"
+#endif
+
 PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 
 PYBIND11_NAMESPACE_BEGIN(detail)
@@ -2374,6 +2379,10 @@ inline function get_overload(const T *this_ptr, const char *name) {
 
 PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)
 
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic pop
+#endif
+
 #if defined(__GNUC__) && __GNUC__ == 7
 #    pragma GCC diagnostic pop // -Wnoexcept-type
 #endif
diff --git a/include/pybind11/pytypes.h b/include/pybind11/pytypes.h
index 54e9ed28a5..477a5cd221 100644
--- a/include/pybind11/pytypes.h
+++ b/include/pybind11/pytypes.h
@@ -14,6 +14,11 @@
 #include <utility>
 #include <type_traits>
 
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic push
+#  pragma GCC diagnostic ignored "-Wattributes"
+#endif
+
 PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
 
 /* A few forward declarations */
@@ -25,21 +30,9 @@ struct arg; struct arg_v;
 PYBIND11_NAMESPACE_BEGIN(detail)
 
 class args_proxy;
-
-#if defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8))
-#  define PYBIND11_PRAGMA_NOINLINE_FWD
-#  pragma GCC diagnostic push
-#  pragma GCC diagnostic ignored "-Wattributes"
-#endif
-
 PYBIND11_NOINLINE_FWD bool isinstance_generic(handle obj, const std::type_info &tp);
 PYBIND11_NOINLINE_FWD std::string error_string();
 
-#if defined(PYBIND11_PRAGMA_NOINLINE_FWD)
-#  pragma GCC diagnostic pop
-#  undef PYBIND11_PRAGMA_NOINLINE_FWD
-#endif
-
 // Accessor forward declarations
 template <typename Policy> class accessor;
 namespace accessor_policies {
@@ -1773,3 +1766,7 @@ PYBIND11_MATH_OPERATOR_BINARY(operator>>=, PyNumber_InPlaceRshift)
 
 PYBIND11_NAMESPACE_END(detail)
 PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)
+
+#if defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#  pragma GCC diagnostic pop
+#endif

From a893084f242da26d6a5439d60186da26e1c39259 Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Mon, 2 Aug 2021 18:12:07 -0700
Subject: [PATCH 08/18] Removing ICC pragma 2196, to see if it is still needed.

---
 include/pybind11/detail/common.h | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index 5c0c785e00..a227b615fd 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -56,9 +56,6 @@
 #  elif __INTEL_COMPILER < 1900 && defined(PYBIND11_CPP14)
 #    error pybind11 supports only C++11 with Intel C++ compiler v18. Use v19 or newer for C++14.
 #  endif
-/* The following pragma cannot be pop'ed:
-   https://community.intel.com/t5/Intel-C-Compiler/Inline-and-no-inline-warning/td-p/1216764 */
-#  pragma warning disable 2196 // warning #2196: routine is both "inline" and "noinline"
 #elif defined(__clang__) && !defined(__apple_build_version__)
 #  if __clang_major__ < 3 || (__clang_major__ == 3 && __clang_minor__ < 3)
 #    error pybind11 requires clang 3.3 or newer

From 71a2799ccf7d2a5cfeebddee39376b76b0f57ed8 Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Mon, 2 Aug 2021 18:20:06 -0700
Subject: [PATCH 09/18] Trying noinline, noinline for ICC

---
 include/pybind11/detail/common.h | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index a227b615fd..7a73c23b0c 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -99,6 +99,9 @@
 #if defined(PYBIND11_NOINLINE_DISABLED)
 #  define PYBIND11_NOINLINE_DCL inline
 #  define PYBIND11_NOINLINE_FWD inline
+#elif defined(__INTEL_COMPILER)
+#  define PYBIND11_NOINLINE_DCL __attribute__ ((noinline)) inline
+#  define PYBIND11_NOINLINE_FWD __attribute__ ((noinline)) inline
 #elif defined(_MSC_VER)
 #  define PYBIND11_NOINLINE_DCL __declspec(noinline) inline
 #  define PYBIND11_NOINLINE_FWD

From 2307390bf0650283081f9a2a1472b7b138a21bef Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Mon, 2 Aug 2021 18:24:21 -0700
Subject: [PATCH 10/18] Trying noinline, inline for ICC

---
 include/pybind11/detail/common.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index 7a73c23b0c..253f3be0c8 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -101,7 +101,7 @@
 #  define PYBIND11_NOINLINE_FWD inline
 #elif defined(__INTEL_COMPILER)
 #  define PYBIND11_NOINLINE_DCL __attribute__ ((noinline)) inline
-#  define PYBIND11_NOINLINE_FWD __attribute__ ((noinline)) inline
+#  define PYBIND11_NOINLINE_FWD inline
 #elif defined(_MSC_VER)
 #  define PYBIND11_NOINLINE_DCL __declspec(noinline) inline
 #  define PYBIND11_NOINLINE_FWD

From 236edb2beeff16012cc88da4e7ccf81944035f78 Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Tue, 3 Aug 2021 12:51:51 -0700
Subject: [PATCH 11/18] Restoring ICC pragma 2196, introducing
 PYBIND11_NOINLINE_FORCED, defined for testing.

---
 include/pybind11/detail/common.h | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index 253f3be0c8..56027b95fe 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -56,6 +56,9 @@
 #  elif __INTEL_COMPILER < 1900 && defined(PYBIND11_CPP14)
 #    error pybind11 supports only C++11 with Intel C++ compiler v18. Use v19 or newer for C++14.
 #  endif
+/* The following pragma cannot be pop'ed:
+   https://community.intel.com/t5/Intel-C-Compiler/Inline-and-no-inline-warning/td-p/1216764 */
+#  pragma warning disable 2196 // warning #2196: routine is both "inline" and "noinline"
 #elif defined(__clang__) && !defined(__apple_build_version__)
 #  if __clang_major__ < 3 || (__clang_major__ == 3 && __clang_minor__ < 3)
 #    error pybind11 requires clang 3.3 or newer
@@ -96,6 +99,17 @@
 #  endif
 #endif
 
+// For CUDA, GCC7, GCC8 (and maybe some other platforms):
+// PYBIND11_NOINLINE_FORCED is incompatible with `-Wattributes -Werror`.
+// When defining PYBIND11_NOINLINE_FORCED, it is best to also use
+// `-Wno-attributes` (or not to use `-Werror` and ignore the warnings).
+#define PYBIND11_NOINLINE_FORCED
+#if !defined(PYBIND11_NOINLINE_FORCED) && \
+    (defined(__CUDACC__) || \
+     (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8)))
+#  define PYBIND11_NOINLINE_DISABLED
+#endif
+
 #if defined(PYBIND11_NOINLINE_DISABLED)
 #  define PYBIND11_NOINLINE_DCL inline
 #  define PYBIND11_NOINLINE_FWD inline
@@ -110,6 +124,11 @@
 #  define PYBIND11_NOINLINE_FWD
 #endif
 
+// DECISION TO BE MADE before this PR is merged:
+// ----------------------------------------------------------------------------
+// This code and the 9 x 2 pragma blocks referring to
+// PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED could be purged.
+// ----------------------------------------------------------------------------
 #if !defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED) && \
     !defined(PYBIND11_NOINLINE_DISABLED) && \
     (defined(__CUDACC__) || \

From c22bd6567576d750d2ac6abcfd4f9310002c833d Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Tue, 3 Aug 2021 13:26:08 -0700
Subject: [PATCH 12/18] Removing code accidentally left in (was for
 experimentation only).

---
 include/pybind11/detail/common.h | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index 56027b95fe..cb3733688c 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -113,9 +113,6 @@
 #if defined(PYBIND11_NOINLINE_DISABLED)
 #  define PYBIND11_NOINLINE_DCL inline
 #  define PYBIND11_NOINLINE_FWD inline
-#elif defined(__INTEL_COMPILER)
-#  define PYBIND11_NOINLINE_DCL __attribute__ ((noinline)) inline
-#  define PYBIND11_NOINLINE_FWD inline
 #elif defined(_MSC_VER)
 #  define PYBIND11_NOINLINE_DCL __declspec(noinline) inline
 #  define PYBIND11_NOINLINE_FWD

From 3c6098503000fc996fe602e7838cd43cba71a715 Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Tue, 3 Aug 2021 13:30:08 -0700
Subject: [PATCH 13/18] Removing one-time-test define.

---
 include/pybind11/detail/common.h | 1 -
 1 file changed, 1 deletion(-)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index cb3733688c..13e56cc51a 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -103,7 +103,6 @@
 // PYBIND11_NOINLINE_FORCED is incompatible with `-Wattributes -Werror`.
 // When defining PYBIND11_NOINLINE_FORCED, it is best to also use
 // `-Wno-attributes` (or not to use `-Werror` and ignore the warnings).
-#define PYBIND11_NOINLINE_FORCED
 #if !defined(PYBIND11_NOINLINE_FORCED) && \
     (defined(__CUDACC__) || \
      (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8)))

From 3b2e642d6bed5def182a4a1c646fbfd6c2ff0bc8 Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Tue, 3 Aug 2021 14:33:48 -0700
Subject: [PATCH 14/18] Removing PYBIND11_NOINLINE_FWD macro (after learning
 that it makes no sense).

---
 docs/Doxyfile                              |  2 +-
 include/pybind11/attr.h                    |  6 ++--
 include/pybind11/detail/common.h           | 15 ++++-----
 include/pybind11/detail/internals.h        |  6 ++--
 include/pybind11/detail/type_caster_base.h | 38 +++++++++++-----------
 include/pybind11/detail/typeid.h           |  2 +-
 include/pybind11/gil.h                     | 10 +++---
 include/pybind11/numpy.h                   |  4 +--
 include/pybind11/pybind11.h                | 16 ++++-----
 include/pybind11/pytypes.h                 |  4 +--
 10 files changed, 51 insertions(+), 52 deletions(-)

diff --git a/docs/Doxyfile b/docs/Doxyfile
index 22af784890..c8562952ef 100644
--- a/docs/Doxyfile
+++ b/docs/Doxyfile
@@ -20,4 +20,4 @@ WARNINGS               = YES
 WARN_IF_UNDOCUMENTED   = NO
 PREDEFINED             = DOXYGEN_SHOULD_SKIP_THIS \
                          PY_MAJOR_VERSION=3 \
-                         PYBIND11_NOINLINE_DCL
+                         PYBIND11_NOINLINE
diff --git a/include/pybind11/attr.h b/include/pybind11/attr.h
index 2a88ee1dc6..5efd7b4f87 100644
--- a/include/pybind11/attr.h
+++ b/include/pybind11/attr.h
@@ -129,7 +129,7 @@ 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_;
-PYBIND11_NOINLINE_FWD 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 {
@@ -225,7 +225,7 @@ struct function_record {
 
 /// Special data structure which (temporarily) holds metadata about a bound class
 struct type_record {
-    PYBIND11_NOINLINE_DCL type_record()
+    PYBIND11_NOINLINE type_record()
         : multiple_inheritance(false), dynamic_attr(false), buffer_protocol(false),
           default_holder(true), module_local(false), is_final(false) { }
 
@@ -283,7 +283,7 @@ struct type_record {
     /// Is the class inheritable from python classes?
     bool is_final : 1;
 
-    PYBIND11_NOINLINE_DCL void add_base(const std::type_info &base, void *(*caster)(void *)) {
+    PYBIND11_NOINLINE void add_base(const std::type_info &base, void *(*caster)(void *)) {
         auto base_info = detail::get_type_info(base, false);
         if (!base_info) {
             std::string tname(base.name());
diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index 13e56cc51a..68af33fb53 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -109,15 +109,14 @@
 #  define PYBIND11_NOINLINE_DISABLED
 #endif
 
+// Note: Forward declaration should never be `inline`:
+// https://stackoverflow.com/questions/9317473/forward-declaration-of-inline-functions
 #if defined(PYBIND11_NOINLINE_DISABLED)
-#  define PYBIND11_NOINLINE_DCL inline
-#  define PYBIND11_NOINLINE_FWD inline
+#  define PYBIND11_NOINLINE inline
 #elif defined(_MSC_VER)
-#  define PYBIND11_NOINLINE_DCL __declspec(noinline) inline
-#  define PYBIND11_NOINLINE_FWD
+#  define PYBIND11_NOINLINE __declspec(noinline) inline
 #else
-#  define PYBIND11_NOINLINE_DCL __attribute__ ((noinline)) inline
-#  define PYBIND11_NOINLINE_FWD
+#  define PYBIND11_NOINLINE __attribute__ ((noinline)) inline
 #endif
 
 // DECISION TO BE MADE before this PR is merged:
@@ -810,8 +809,8 @@ 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_DCL void pybind11_fail(const char *reason) { throw std::runtime_error(reason); }
-[[noreturn]] PYBIND11_NOINLINE_DCL 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 { };
 
diff --git a/include/pybind11/detail/internals.h b/include/pybind11/detail/internals.h
index 83ced63a3b..c5f341cb03 100644
--- a/include/pybind11/detail/internals.h
+++ b/include/pybind11/detail/internals.h
@@ -262,7 +262,7 @@ inline void translate_local_exception(std::exception_ptr p) {
 #endif
 
 /// Return a reference to the current `internals` data
-PYBIND11_NOINLINE_DCL internals &get_internals() {
+PYBIND11_NOINLINE internals &get_internals() {
     auto **&internals_pp = get_internals_pp();
     if (internals_pp && *internals_pp)
         return **internals_pp;
@@ -357,14 +357,14 @@ 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.
-PYBIND11_NOINLINE_DCL 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()`.
-PYBIND11_NOINLINE_DCL 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;
 }
diff --git a/include/pybind11/detail/type_caster_base.h b/include/pybind11/detail/type_caster_base.h
index f2f72106bf..eb28233612 100644
--- a/include/pybind11/detail/type_caster_base.h
+++ b/include/pybind11/detail/type_caster_base.h
@@ -59,7 +59,7 @@ class loader_life_support {
 
     /// 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_DCL static void add_patient(handle h) {
+    PYBIND11_NOINLINE static void add_patient(handle h) {
         auto &stack = get_internals().loader_patient_stack;
         if (stack.empty())
             throw cast_error("When called outside a bound function, py::cast() cannot "
@@ -86,7 +86,7 @@ class loader_life_support {
 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_DCL 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());
@@ -155,7 +155,7 @@ 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_DCL 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;
@@ -181,7 +181,7 @@ 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_DCL 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;
@@ -196,13 +196,13 @@ PYBIND11_NOINLINE_DCL detail::type_info *get_type_info(const std::type_index &tp
     return nullptr;
 }
 
-PYBIND11_NOINLINE_DCL 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_DCL 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) {
@@ -330,7 +330,7 @@ struct values_and_holders {
  * The returned object should be short-lived: in particular, it must not outlive the called-upon
  * instance.
  */
-PYBIND11_NOINLINE_DCL 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);
@@ -354,7 +354,7 @@ PYBIND11_NOINLINE_DCL value_and_holder instance::get_value_and_holder(const type
 #endif
 }
 
-PYBIND11_NOINLINE_DCL 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();
@@ -402,19 +402,19 @@ PYBIND11_NOINLINE_DCL void instance::allocate_layout() {
     owned = true;
 }
 
-PYBIND11_NOINLINE_DCL void instance::deallocate_layout() const {
+PYBIND11_NOINLINE void instance::deallocate_layout() const {
     if (!simple_layout)
         PyMem_Free(nonsimple.values_and_holders);
 }
 
-PYBIND11_NOINLINE_DCL 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_DCL 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";
@@ -461,7 +461,7 @@ PYBIND11_NOINLINE_DCL std::string error_string() {
     return errorString;
 }
 
-PYBIND11_NOINLINE_DCL 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) {
@@ -488,12 +488,12 @@ inline PyThreadState *get_thread_state_unchecked() {
 }
 
 // Forward declarations
-PYBIND11_NOINLINE_FWD 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_DCL type_caster_generic(const std::type_info &type_info)
+    PYBIND11_NOINLINE 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)
@@ -503,7 +503,7 @@ class type_caster_generic {
         return load_impl<type_caster_generic>(src, convert);
     }
 
-    PYBIND11_NOINLINE_DCL static handle cast(const void *_src, return_value_policy policy, handle parent,
+    PYBIND11_NOINLINE static handle cast(const void *_src, return_value_policy policy, handle parent,
                                          const detail::type_info *tinfo,
                                          void *(*copy_constructor)(const void *),
                                          void *(*move_constructor)(const void *),
@@ -627,7 +627,7 @@ class type_caster_generic {
     }
     void check_holder_compat() {}
 
-    PYBIND11_NOINLINE_DCL static void *local_load(PyObject *src, const type_info *ti) {
+    PYBIND11_NOINLINE static void *local_load(PyObject *src, const type_info *ti) {
         auto caster = type_caster_generic(ti);
         if (caster.load(src, false))
             return caster.value;
@@ -636,7 +636,7 @@ class type_caster_generic {
 
     /// Try to load with foreign typeinfo, if available. Used when there is no
     /// native typeinfo, or when the native one wasn't able to produce a value.
-    PYBIND11_NOINLINE_DCL bool try_load_foreign_module_local(handle src) {
+    PYBIND11_NOINLINE bool try_load_foreign_module_local(handle src) {
         constexpr auto *local_key = PYBIND11_MODULE_LOCAL_ID;
         const auto pytype = type::handle_of(src);
         if (!hasattr(pytype, local_key))
@@ -659,7 +659,7 @@ class type_caster_generic {
     // bits of code between here and copyable_holder_caster where the two classes need different
     // logic (without having to resort to virtual inheritance).
     template <typename ThisT>
-    PYBIND11_NOINLINE_DCL bool load_impl(handle src, bool convert) {
+    PYBIND11_NOINLINE bool load_impl(handle src, bool convert) {
         if (!src) return false;
         if (!typeinfo) return try_load_foreign_module_local(src);
 
@@ -749,7 +749,7 @@ class type_caster_generic {
     // Called to do type lookup and wrap the pointer and type in a pair when a dynamic_cast
     // isn't needed or can't be used.  If the type is unknown, sets the error and returns a pair
     // with .second = nullptr.  (p.first = nullptr is not an error: it becomes None).
-    PYBIND11_NOINLINE_DCL static std::pair<const void *, const type_info *> src_and_type(
+    PYBIND11_NOINLINE static std::pair<const void *, const type_info *> src_and_type(
             const void *src, const std::type_info &cast_type, const std::type_info *rtti_type = nullptr) {
         if (auto *tpi = get_type_info(cast_type))
             return {src, const_cast<const type_info *>(tpi)};
diff --git a/include/pybind11/detail/typeid.h b/include/pybind11/detail/typeid.h
index bb24d9b615..cb07ba7fbc 100644
--- a/include/pybind11/detail/typeid.h
+++ b/include/pybind11/detail/typeid.h
@@ -34,7 +34,7 @@ inline void erase_all(std::string &string, const std::string &search) {
     }
 }
 
-PYBIND11_NOINLINE_DCL 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 {
diff --git a/include/pybind11/gil.h b/include/pybind11/gil.h
index ac0bcdf3bc..5bcfafcf37 100644
--- a/include/pybind11/gil.h
+++ b/include/pybind11/gil.h
@@ -53,7 +53,7 @@ PYBIND11_NAMESPACE_END(detail)
 
 class gil_scoped_acquire {
 public:
-    PYBIND11_NOINLINE_DCL gil_scoped_acquire() {
+    PYBIND11_NOINLINE gil_scoped_acquire() {
         auto const &internals = detail::get_internals();
         tstate = (PyThreadState *) PYBIND11_TLS_GET_VALUE(internals.tstate);
 
@@ -89,7 +89,7 @@ class gil_scoped_acquire {
         ++tstate->gilstate_counter;
     }
 
-    PYBIND11_NOINLINE_DCL void dec_ref() {
+    PYBIND11_NOINLINE void dec_ref() {
         --tstate->gilstate_counter;
         #if !defined(NDEBUG)
             if (detail::get_thread_state_unchecked() != tstate)
@@ -115,11 +115,11 @@ class gil_scoped_acquire {
     /// could be shutting down when this is called, as thread deletion is not
     /// allowed during shutdown. Check _Py_IsFinalizing() on Python 3.7+, and
     /// protect subsequent code.
-    PYBIND11_NOINLINE_DCL void disarm() {
+    PYBIND11_NOINLINE void disarm() {
         active = false;
     }
 
-    PYBIND11_NOINLINE_DCL ~gil_scoped_acquire() {
+    PYBIND11_NOINLINE ~gil_scoped_acquire() {
         dec_ref();
         if (release)
            PyEval_SaveThread();
@@ -149,7 +149,7 @@ class gil_scoped_release {
     /// could be shutting down when this is called, as thread deletion is not
     /// allowed during shutdown. Check _Py_IsFinalizing() on Python 3.7+, and
     /// protect subsequent code.
-    PYBIND11_NOINLINE_DCL void disarm() {
+    PYBIND11_NOINLINE void disarm() {
         active = false;
     }
 
diff --git a/include/pybind11/numpy.h b/include/pybind11/numpy.h
index f295dfa51a..941c268b5c 100644
--- a/include/pybind11/numpy.h
+++ b/include/pybind11/numpy.h
@@ -109,7 +109,7 @@ struct numpy_internals {
     }
 };
 
-PYBIND11_NOINLINE_DCL void load_numpy_internals(numpy_internals* &ptr) {
+PYBIND11_NOINLINE void load_numpy_internals(numpy_internals* &ptr) {
     ptr = &get_or_create_shared_data<numpy_internals>("_numpy_internals");
 }
 
@@ -1115,7 +1115,7 @@ struct field_descriptor {
     dtype descr;
 };
 
-PYBIND11_NOINLINE_DCL void register_structured_dtype(
+PYBIND11_NOINLINE void register_structured_dtype(
     any_container<field_descriptor> fields,
     const std::type_info& tinfo, ssize_t itemsize,
     bool (*direct_converter)(PyObject *, void *&)) {
diff --git a/include/pybind11/pybind11.h b/include/pybind11/pybind11.h
index 9d9140aec0..ab8ac22532 100644
--- a/include/pybind11/pybind11.h
+++ b/include/pybind11/pybind11.h
@@ -146,7 +146,7 @@ class cpp_function : public function {
     using unique_function_record = std::unique_ptr<detail::function_record, InitializingFunctionRecordDeleter>;
 
     /// Space optimization: don't inline this frequently instantiated fragment
-    PYBIND11_NOINLINE_DCL unique_function_record make_function_record() {
+    PYBIND11_NOINLINE unique_function_record make_function_record() {
         return unique_function_record(new detail::function_record());
     }
 
@@ -1056,7 +1056,7 @@ class module_ : public object {
         ``overwrite`` should almost always be false: attempting to overwrite objects that pybind11 has
         established will, in most cases, break things.
     \endrst */
-    PYBIND11_NOINLINE_DCL void add_object(const char *name, handle obj, bool overwrite = false) {
+    PYBIND11_NOINLINE void add_object(const char *name, handle obj, bool overwrite = false) {
         if (!overwrite && hasattr(*this, name))
             pybind11_fail("Error during initialization: multiple incompatible definitions with name \"" +
                     std::string(name) + "\"");
@@ -1671,7 +1671,7 @@ inline str enum_name(handle arg) {
 struct enum_base {
     enum_base(handle base, handle parent) : m_base(base), m_parent(parent) { }
 
-    PYBIND11_NOINLINE_DCL void init(bool is_arithmetic, bool is_convertible) {
+    PYBIND11_NOINLINE void init(bool is_arithmetic, bool is_convertible) {
         m_base.attr("__entries") = dict();
         auto property = handle((PyObject *) &PyProperty_Type);
         auto static_property = handle((PyObject *) get_internals().static_property_type);
@@ -1797,7 +1797,7 @@ struct enum_base {
             [](const object &arg) { return int_(arg); }, name("__hash__"), is_method(m_base));
     }
 
-    PYBIND11_NOINLINE_DCL void value(char const* name_, object value, const char *doc = nullptr) {
+    PYBIND11_NOINLINE void value(char const* name_, object value, const char *doc = nullptr) {
         dict entries = m_base.attr("__entries");
         str name(name_);
         if (entries.contains(name)) {
@@ -1809,7 +1809,7 @@ struct enum_base {
         m_base.attr(name) = value;
     }
 
-    PYBIND11_NOINLINE_DCL void export_values() {
+    PYBIND11_NOINLINE void export_values() {
         dict entries = m_base.attr("__entries");
         for (auto kv : entries)
             m_parent.attr(kv.first) = kv.second[int_(0)];
@@ -1874,7 +1874,7 @@ template <typename Type> class enum_ : public class_<Type> {
 
 PYBIND11_NAMESPACE_BEGIN(detail)
 
-PYBIND11_NOINLINE_DCL void keep_alive_impl(handle nurse, handle patient) {
+PYBIND11_NOINLINE void keep_alive_impl(handle nurse, handle patient) {
     if (!nurse || !patient)
         pybind11_fail("Could not activate keep_alive!");
 
@@ -1901,7 +1901,7 @@ PYBIND11_NOINLINE_DCL void keep_alive_impl(handle nurse, handle patient) {
     }
 }
 
-PYBIND11_NOINLINE_DCL void keep_alive_impl(size_t Nurse, size_t Patient, function_call &call, handle ret) {
+PYBIND11_NOINLINE void keep_alive_impl(size_t Nurse, size_t Patient, function_call &call, handle ret) {
     auto get_arg = [&](size_t n) {
         if (n == 0)
             return ret;
@@ -2151,7 +2151,7 @@ exception<CppException> &register_local_exception(handle scope,
 }
 
 PYBIND11_NAMESPACE_BEGIN(detail)
-PYBIND11_NOINLINE_DCL void print(const tuple &args, const dict &kwargs) {
+PYBIND11_NOINLINE void print(const tuple &args, const dict &kwargs) {
     auto strings = tuple(args.size());
     for (size_t i = 0; i < args.size(); ++i) {
         strings[i] = str(args[i]);
diff --git a/include/pybind11/pytypes.h b/include/pybind11/pytypes.h
index 477a5cd221..ba19c18556 100644
--- a/include/pybind11/pytypes.h
+++ b/include/pybind11/pytypes.h
@@ -30,8 +30,8 @@ struct arg; struct arg_v;
 PYBIND11_NAMESPACE_BEGIN(detail)
 
 class args_proxy;
-PYBIND11_NOINLINE_FWD bool isinstance_generic(handle obj, const std::type_info &tp);
-PYBIND11_NOINLINE_FWD std::string error_string();
+bool isinstance_generic(handle obj, const std::type_info &tp);
+std::string error_string();
 
 // Accessor forward declarations
 template <typename Policy> class accessor;

From a304a25ef400e16cada9ee233d586d92fcec1ea9 Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Tue, 3 Aug 2021 15:10:18 -0700
Subject: [PATCH 15/18] Testing with PYBIND11_NOINLINE_DISABLED. Minor
 non-functional enhancements.

---
 include/pybind11/detail/common.h | 21 ++++++++++++---------
 1 file changed, 12 insertions(+), 9 deletions(-)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index 68af33fb53..15aa1d893f 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -56,9 +56,6 @@
 #  elif __INTEL_COMPILER < 1900 && defined(PYBIND11_CPP14)
 #    error pybind11 supports only C++11 with Intel C++ compiler v18. Use v19 or newer for C++14.
 #  endif
-/* The following pragma cannot be pop'ed:
-   https://community.intel.com/t5/Intel-C-Compiler/Inline-and-no-inline-warning/td-p/1216764 */
-#  pragma warning disable 2196 // warning #2196: routine is both "inline" and "noinline"
 #elif defined(__clang__) && !defined(__apple_build_version__)
 #  if __clang_major__ < 3 || (__clang_major__ == 3 && __clang_minor__ < 3)
 #    error pybind11 requires clang 3.3 or newer
@@ -99,6 +96,8 @@
 #  endif
 #endif
 
+#define PYBIND11_NOINLINE_DISABLED
+
 // For CUDA, GCC7, GCC8 (and maybe some other platforms):
 // PYBIND11_NOINLINE_FORCED is incompatible with `-Wattributes -Werror`.
 // When defining PYBIND11_NOINLINE_FORCED, it is best to also use
@@ -109,7 +108,7 @@
 #  define PYBIND11_NOINLINE_DISABLED
 #endif
 
-// Note: Forward declaration should never be `inline`:
+// Note: Forward declarations should never be `inline`:
 // https://stackoverflow.com/questions/9317473/forward-declaration-of-inline-functions
 #if defined(PYBIND11_NOINLINE_DISABLED)
 #  define PYBIND11_NOINLINE inline
@@ -124,11 +123,15 @@
 // This code and the 9 x 2 pragma blocks referring to
 // PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED could be purged.
 // ----------------------------------------------------------------------------
-#if !defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED) && \
-    !defined(PYBIND11_NOINLINE_DISABLED) && \
-    (defined(__CUDACC__) || \
-     (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8)))
-#  define PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED
+#if !defined(PYBIND11_NOINLINE_DISABLED)
+#    if (defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8)))          \
+        && !defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)
+#        define PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED
+#    elif defined(__INTEL_COMPILER)
+// The following pragma cannot be pop'ed:
+// https://community.intel.com/t5/Intel-C-Compiler/Inline-and-no-inline-warning/td-p/1216764
+#        pragma warning disable 2196 // warning #2196: routine is both "inline" and "noinline"
+#    endif
 #endif
 
 #if defined(__MINGW32__)

From 2fade636731afb1661236e12c69ba3620acc3b8d Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Tue, 3 Aug 2021 17:20:15 -0700
Subject: [PATCH 16/18] Removing #define PYBIND11_NOINLINE_DISABLED (test was
 successful).

---
 include/pybind11/detail/common.h | 2 --
 1 file changed, 2 deletions(-)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index 15aa1d893f..890ebc7251 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -96,8 +96,6 @@
 #  endif
 #endif
 
-#define PYBIND11_NOINLINE_DISABLED
-
 // For CUDA, GCC7, GCC8 (and maybe some other platforms):
 // PYBIND11_NOINLINE_FORCED is incompatible with `-Wattributes -Werror`.
 // When defining PYBIND11_NOINLINE_FORCED, it is best to also use

From af0d4ca348dfae5104644fc223a5e5f6f88f79bf Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Mon, 9 Aug 2021 11:48:02 -0700
Subject: [PATCH 17/18] Fixing up after git rebase -X theirs master

---
 include/pybind11/detail/common.h | 5 +++--
 include/pybind11/pytypes.h       | 6 ++++--
 2 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index 890ebc7251..159cf96116 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -106,9 +106,10 @@
 #  define PYBIND11_NOINLINE_DISABLED
 #endif
 
-// Note: Forward declarations should never be `inline`:
+// 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)
+#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
diff --git a/include/pybind11/pytypes.h b/include/pybind11/pytypes.h
index ba19c18556..e6d985630b 100644
--- a/include/pybind11/pytypes.h
+++ b/include/pybind11/pytypes.h
@@ -28,10 +28,8 @@ class type;
 struct arg; struct arg_v;
 
 PYBIND11_NAMESPACE_BEGIN(detail)
-
 class args_proxy;
 bool isinstance_generic(handle obj, const std::type_info &tp);
-std::string error_string();
 
 // Accessor forward declarations
 template <typename Policy> class accessor;
@@ -322,6 +320,10 @@ template <typename T> T reinterpret_borrow(handle h) { return {h, object::borrow
 \endrst */
 template <typename T> T reinterpret_steal(handle h) { return {h, object::stolen_t{}}; }
 
+PYBIND11_NAMESPACE_BEGIN(detail)
+std::string error_string();
+PYBIND11_NAMESPACE_END(detail)
+
 #if defined(_MSC_VER)
 #  pragma warning(push)
 #  pragma warning(disable: 4275 4251) // warning C4275: An exported class was derived from a class that wasn't exported. Can be ignored when derived from a STL class.

From bec4e6d2db331fe51a886ddd04fabf69a40ba945 Mon Sep 17 00:00:00 2001
From: "Ralf W. Grosse-Kunstleve" <rwgk@google.com>
Date: Mon, 9 Aug 2021 11:53:06 -0700
Subject: [PATCH 18/18] Removing PYBIND11_NOINLINE_FORCED (to be moved to an
 alternative PR).

---
 include/pybind11/detail/common.h | 15 ---------------
 1 file changed, 15 deletions(-)

diff --git a/include/pybind11/detail/common.h b/include/pybind11/detail/common.h
index 159cf96116..8b06d771b2 100644
--- a/include/pybind11/detail/common.h
+++ b/include/pybind11/detail/common.h
@@ -96,16 +96,6 @@
 #  endif
 #endif
 
-// For CUDA, GCC7, GCC8 (and maybe some other platforms):
-// PYBIND11_NOINLINE_FORCED is incompatible with `-Wattributes -Werror`.
-// When defining PYBIND11_NOINLINE_FORCED, it is best to also use
-// `-Wno-attributes` (or not to use `-Werror` and ignore the warnings).
-#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
@@ -117,11 +107,6 @@
 #  define PYBIND11_NOINLINE __attribute__ ((noinline)) inline
 #endif
 
-// DECISION TO BE MADE before this PR is merged:
-// ----------------------------------------------------------------------------
-// This code and the 9 x 2 pragma blocks referring to
-// PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED could be purged.
-// ----------------------------------------------------------------------------
 #if !defined(PYBIND11_NOINLINE_DISABLED)
 #    if (defined(__CUDACC__) || (defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8)))          \
         && !defined(PYBIND11_NOINLINE_GCC_PRAGMA_ATTRIBUTES_NEEDED)