Merge 8917a1e9b3 into 83b92ceb35

* Disable print_destroyed in tests on GraalPy

* Reenable test_iostream on GraalPy

.github/workflows/ci.yml

@@ -65,6 +65,24 @@ jobs:
           # Inject a couple Windows 2019 runs
           - runs-on: windows-2019
             python: '3.9'
+          # Inject a few runs with different runtime libraries
+          - runs-on: windows-2022
+            python: '3.9'
+            args: >
+              -DCMAKE_MSVC_RUNTIME_LIBRARY=MultiThreaded
+          - runs-on: windows-2022
+            python: '3.10'
+            args: >
+              -DCMAKE_MSVC_RUNTIME_LIBRARY=MultiThreadedDLL
+          # This needs a python built with MTd
+          # - runs-on: windows-2022
+          #   python: '3.11'
+          #   args: >
+          #     -DCMAKE_MSVC_RUNTIME_LIBRARY=MultiThreadedDebug
+          - runs-on: windows-2022
+            python: '3.12'
+            args: >
+              -DCMAKE_MSVC_RUNTIME_LIBRARY=MultiThreadedDebugDLL
           # Extra ubuntu latest job
           - runs-on: ubuntu-latest
             python: '3.11'
@@ -127,6 +145,7 @@ jobs:
         -DPYBIND11_DISABLE_HANDLE_TYPE_NAME_DEFAULT_IMPLEMENTATION=ON
         -DPYBIND11_SIMPLE_GIL_MANAGEMENT=ON
         -DPYBIND11_NUMPY_1_ONLY=ON
+        -DPYBIND11_PYTEST_ARGS=-v
         -DDOWNLOAD_CATCH=ON
         -DDOWNLOAD_EIGEN=ON
         -DCMAKE_CXX_STANDARD=11
@@ -156,6 +175,7 @@ jobs:
         -DPYBIND11_WERROR=ON
         -DPYBIND11_SIMPLE_GIL_MANAGEMENT=OFF
         -DPYBIND11_NUMPY_1_ONLY=ON
+        -DPYBIND11_PYTEST_ARGS=-v
         -DDOWNLOAD_CATCH=ON
         -DDOWNLOAD_EIGEN=ON
         -DCMAKE_CXX_STANDARD=17
@@ -175,6 +195,7 @@ jobs:
       run: >
         cmake -S . -B build3
         -DPYBIND11_WERROR=ON
+        -DPYBIND11_PYTEST_ARGS=-v
         -DDOWNLOAD_CATCH=ON
         -DDOWNLOAD_EIGEN=ON
         -DCMAKE_CXX_STANDARD=17
@@ -998,7 +1019,6 @@ jobs:
           git
           mingw-w64-${{matrix.env}}-gcc
           mingw-w64-${{matrix.env}}-python-pip
-          mingw-w64-${{matrix.env}}-python-numpy
           mingw-w64-${{matrix.env}}-cmake
           mingw-w64-${{matrix.env}}-make
           mingw-w64-${{matrix.env}}-python-pytest
@@ -1010,7 +1030,7 @@ jobs:
       with:
         msystem: ${{matrix.sys}}
         install: >-
-          git
+          mingw-w64-${{matrix.env}}-python-numpy
           mingw-w64-${{matrix.env}}-python-scipy
           mingw-w64-${{matrix.env}}-eigen3
 

.github/workflows/pip.yml

@@ -103,7 +103,7 @@ jobs:
     - uses: actions/download-artifact@v4
 
     - name: Generate artifact attestation for sdist and wheel
-      uses: actions/attest-build-provenance@1c608d11d69870c2092266b3f9a6f3abbf17002c # v1.4.3
+      uses: actions/attest-build-provenance@ef244123eb79f2f7a7e75d99086184180e6d0018 # v1.4.4
       with:
         subject-path: "*/pybind11*"
 

.pre-commit-config.yaml

@@ -25,14 +25,14 @@ repos:
 
 # Clang format the codebase automatically
 - repo: https://github.com/pre-commit/mirrors-clang-format
-  rev: "v18.1.8"
+  rev: "v19.1.3"
   hooks:
   - id: clang-format
     types_or: [c++, c, cuda]
 
 # Ruff, the Python auto-correcting linter/formatter written in Rust
 - repo: https://github.com/astral-sh/ruff-pre-commit
-  rev: v0.6.3
+  rev: v0.7.2
   hooks:
   - id: ruff
     args: ["--fix", "--show-fixes"]
@@ -40,7 +40,7 @@ repos:
 
 # Check static types with mypy
 - repo: https://github.com/pre-commit/mirrors-mypy
-  rev: "v1.11.2"
+  rev: "v1.13.0"
   hooks:
   - id: mypy
     args: []
@@ -62,7 +62,7 @@ repos:
 
 # Standard hooks
 - repo: https://github.com/pre-commit/pre-commit-hooks
-  rev: "v4.6.0"
+  rev: "v5.0.0"
   hooks:
   - id: check-added-large-files
   - id: check-case-conflict
@@ -79,7 +79,7 @@ repos:
 
 # Also code format the docs
 - repo: https://github.com/adamchainz/blacken-docs
-  rev: "1.18.0"
+  rev: "1.19.1"
   hooks:
   - id: blacken-docs
     additional_dependencies:
@@ -108,7 +108,7 @@ repos:
 
 # Checks the manifest for missing files (native support)
 - repo: https://github.com/mgedmin/check-manifest
-  rev: "0.49"
+  rev: "0.50"
   hooks:
   - id: check-manifest
     # This is a slow hook, so only run this if --hook-stage manual is passed
@@ -142,14 +142,14 @@ repos:
 
 # PyLint has native support - not always usable, but works for us
 - repo: https://github.com/PyCQA/pylint
-  rev: "v3.2.7"
+  rev: "v3.3.1"
   hooks:
   - id: pylint
     files: ^pybind11
 
 # Check schemas on some of our YAML files
 - repo: https://github.com/python-jsonschema/check-jsonschema
-  rev: 0.29.2
+  rev: 0.29.4
   hooks:
   - id: check-readthedocs
   - id: check-github-workflows

CMakeLists.txt

@@ -143,6 +143,9 @@ set(PYBIND11_HEADERS
     include/pybind11/chrono.h
     include/pybind11/common.h
     include/pybind11/complex.h
+    include/pybind11/conduit/pybind11_conduit_v1.h
+    include/pybind11/conduit/pybind11_platform_abi_id.h
+    include/pybind11/conduit/wrap_include_python_h.h
     include/pybind11/options.h
     include/pybind11/eigen.h
     include/pybind11/eigen/common.h

docs/upgrade.rst

@@ -24,7 +24,8 @@ changes are that:
   function is not available anymore.
 
 Due to NumPy changes, you may experience difficulties updating to NumPy 2.
-Please see the [NumPy 2 migration guide](https://numpy.org/devdocs/numpy_2_0_migration_guide.html) for details.
+Please see the `NumPy 2 migration guide <https://numpy.org/devdocs/numpy_2_0_migration_guide.html>`_
+for details.
 For example, a more direct change could be that the default integer ``"int_"``
 (and ``"uint"``) is now ``ssize_t`` and not ``long`` (affects 64bit windows).
 

include/pybind11/cast.h

@@ -1012,10 +1012,18 @@ template <>
 struct handle_type_name<args> {
     static constexpr auto name = const_name("*args");
 };
+template <typename T>
+struct handle_type_name<Args<T>> {
+    static constexpr auto name = const_name("*args: ") + make_caster<T>::name;
+};
 template <>
 struct handle_type_name<kwargs> {
     static constexpr auto name = const_name("**kwargs");
 };
+template <typename T>
+struct handle_type_name<KWArgs<T>> {
+    static constexpr auto name = const_name("**kwargs: ") + make_caster<T>::name;
+};
 template <>
 struct handle_type_name<obj_attr_accessor> {
     static constexpr auto name = const_name<obj_attr_accessor>();

include/pybind11/conduit/README.txt

@@ -0,0 +1,15 @@
+NOTE
+----
+
+The C++ code here
+
+** only depends on <Python.h> **
+
+and nothing else.
+
+DO NOT ADD CODE WITH OTHER EXTERNAL DEPENDENCIES TO THIS DIRECTORY.
+
+Read on:
+
+pybind11_conduit_v1.h — Type-safe interoperability between different
+                        independent Python/C++ bindings systems.

include/pybind11/conduit/pybind11_conduit_v1.h

@@ -0,0 +1,111 @@
+// Copyright (c) 2024 The pybind Community.
+
+/* The pybind11_conduit_v1 feature enables type-safe interoperability between
+
+* different independent Python/C++ bindings systems,
+
+* including pybind11 versions with different PYBIND11_INTERNALS_VERSION's.
+
+The naming of the feature is a bit misleading:
+
+* The feature is in no way tied to pybind11 internals.
+
+* It just happens to originate from pybind11 and currently still lives there.
+
+* The only external dependency is <Python.h>.
+
+The implementation is a VERY light-weight dependency. It is designed to be
+compatible with any ISO C++11 (or higher) compiler, and does NOT require
+C++ Exception Handling to be enabled.
+
+Please see https://github.com/pybind/pybind11/pull/5296 for more background.
+
+The implementation involves a
+
+def _pybind11_conduit_v1_(
+    self,
+    pybind11_platform_abi_id: bytes,
+    cpp_type_info_capsule: capsule,
+    pointer_kind: bytes) -> capsule
+
+method that is meant to be added to Python objects wrapping C++ objects
+(e.g. pybind11::class_-wrapped types).
+
+The design of the _pybind11_conduit_v1_ feature provides two layers of
+protection against C++ ABI mismatches:
+
+* The first and most important layer is that the pybind11_platform_abi_id's
+  must match between extensions. — This will never be perfect, but is the same
+  pragmatic approach used in pybind11 since 2017
+  (https://github.com/pybind/pybind11/commit/96997a4b9d4ec3d389a570604394af5d5eee2557,
+  PYBIND11_INTERNALS_ID).
+
+* The second layer is that the typeid(std::type_info).name()'s must match
+  between extensions.
+
+The implementation below (which is shorter than this comment!), serves as a
+battle-tested specification. The main API is this one function:
+
+auto *cpp_pointer = pybind11_conduit_v1::get_type_pointer_ephemeral<YourType>(py_obj);
+
+It is meant to be a minimalistic reference implementation, intentionally
+without comprehensive error reporting. It is expected that major bindings
+systems will roll their own, compatible implementations, potentially with
+system-specific error reporting. The essential specifications all bindings
+systems need to agree on are merely:
+
+* PYBIND11_PLATFORM_ABI_ID (const char* literal).
+
+* The cpp_type_info capsule (see below: a void *ptr and a const char *name).
+
+* The cpp_conduit capsule (see below: a void *ptr and a const char *name).
+
+* "raw_pointer_ephemeral" means: the lifetime of the pointer is the lifetime
+  of the py_obj.
+
+*/
+
+// THIS MUST STAY AT THE TOP!
+#include "pybind11_platform_abi_id.h"
+
+#include <Python.h>
+#include <typeinfo>
+
+namespace pybind11_conduit_v1 {
+
+inline void *get_raw_pointer_ephemeral(PyObject *py_obj, const std::type_info *cpp_type_info) {
+    PyObject *cpp_type_info_capsule
+        = PyCapsule_New(const_cast<void *>(static_cast<const void *>(cpp_type_info)),
+                        typeid(std::type_info).name(),
+                        nullptr);
+    if (cpp_type_info_capsule == nullptr) {
+        return nullptr;
+    }
+    PyObject *cpp_conduit = PyObject_CallMethod(py_obj,
+                                                "_pybind11_conduit_v1_",
+                                                "yOy",
+                                                PYBIND11_PLATFORM_ABI_ID,
+                                                cpp_type_info_capsule,
+                                                "raw_pointer_ephemeral");
+    Py_DECREF(cpp_type_info_capsule);
+    if (cpp_conduit == nullptr) {
+        return nullptr;
+    }
+    void *raw_ptr = PyCapsule_GetPointer(cpp_conduit, cpp_type_info->name());
+    Py_DECREF(cpp_conduit);
+    if (PyErr_Occurred()) {
+        return nullptr;
+    }
+    return raw_ptr;
+}
+
+template <typename T>
+T *get_type_pointer_ephemeral(PyObject *py_obj) {
+    void *raw_ptr = get_raw_pointer_ephemeral(py_obj, &typeid(T));
+    if (raw_ptr == nullptr) {
+        return nullptr;
+    }
+    return static_cast<T *>(raw_ptr);
+}
+
+} // namespace pybind11_conduit_v1

include/pybind11/conduit/pybind11_platform_abi_id.h

@@ -0,0 +1,88 @@
+#pragma once
+
+// Copyright (c) 2024 The pybind Community.
+
+// To maximize reusability:
+// DO NOT ADD CODE THAT REQUIRES C++ EXCEPTION HANDLING.
+
+#include "wrap_include_python_h.h"
+
+// Implementation details. DO NOT USE ELSEWHERE. (Unfortunately we cannot #undef them.)
+// This is duplicated here to maximize portability.
+#define PYBIND11_PLATFORM_ABI_ID_STRINGIFY(x) #x
+#define PYBIND11_PLATFORM_ABI_ID_TOSTRING(x) PYBIND11_PLATFORM_ABI_ID_STRINGIFY(x)
+
+// On MSVC, debug and release builds are not ABI-compatible!
+#if defined(_MSC_VER) && defined(_DEBUG)
+#    define PYBIND11_BUILD_TYPE "_debug"
+#else
+#    define PYBIND11_BUILD_TYPE ""
+#endif
+
+// Let's assume that different compilers are ABI-incompatible.
+// A user can manually set this string if they know their
+// compiler is compatible.
+#ifndef PYBIND11_COMPILER_TYPE
+#    if defined(_MSC_VER)
+#        define PYBIND11_COMPILER_TYPE "_msvc"
+#    elif defined(__INTEL_COMPILER)
+#        define PYBIND11_COMPILER_TYPE "_icc"
+#    elif defined(__clang__)
+#        define PYBIND11_COMPILER_TYPE "_clang"
+#    elif defined(__PGI)
+#        define PYBIND11_COMPILER_TYPE "_pgi"
+#    elif defined(__MINGW32__)
+#        define PYBIND11_COMPILER_TYPE "_mingw"
+#    elif defined(__CYGWIN__)
+#        define PYBIND11_COMPILER_TYPE "_gcc_cygwin"
+#    elif defined(__GNUC__)
+#        define PYBIND11_COMPILER_TYPE "_gcc"
+#    else
+#        define PYBIND11_COMPILER_TYPE "_unknown"
+#    endif
+#endif
+
+// Also standard libs
+#ifndef PYBIND11_STDLIB
+#    if defined(_LIBCPP_VERSION)
+#        define PYBIND11_STDLIB "_libcpp"
+#    elif defined(__GLIBCXX__) || defined(__GLIBCPP__)
+#        define PYBIND11_STDLIB "_libstdcpp"
+#    else
+#        define PYBIND11_STDLIB ""
+#    endif
+#endif
+
+#ifndef PYBIND11_BUILD_ABI
+#    if defined(__GXX_ABI_VERSION) // Linux/OSX.
+#        define PYBIND11_BUILD_ABI "_cxxabi" PYBIND11_PLATFORM_ABI_ID_TOSTRING(__GXX_ABI_VERSION)
+#    elif defined(_MSC_VER)               // See PR #4953.
+#        if defined(_MT) && defined(_DLL) // Corresponding to CL command line options /MD or /MDd.
+#            if (_MSC_VER) / 100 == 19
+#                define PYBIND11_BUILD_ABI "_md_mscver19"
+#            else
+#                error "Unknown major version for MSC_VER: PLEASE REVISE THIS CODE."
+#            endif
+#        elif defined(_MT) // Corresponding to CL command line options /MT or /MTd.
+#            define PYBIND11_BUILD_ABI "_mt_mscver" PYBIND11_PLATFORM_ABI_ID_TOSTRING(_MSC_VER)
+#        else
+#            if (_MSC_VER) / 100 == 19
+#                define PYBIND11_BUILD_ABI "_none_mscver19"
+#            else
+#                error "Unknown major version for MSC_VER: PLEASE REVISE THIS CODE."
+#            endif
+#        endif
+#    elif defined(__NVCOMPILER)       // NVHPC (PGI-based).
+#        define PYBIND11_BUILD_ABI "" // TODO: What should be here, to prevent UB?
+#    else
+#        error "Unknown platform or compiler: PLEASE REVISE THIS CODE."
+#    endif
+#endif
+
+#ifndef PYBIND11_INTERNALS_KIND
+#    define PYBIND11_INTERNALS_KIND ""
+#endif
+
+#define PYBIND11_PLATFORM_ABI_ID                                                                  \
+    PYBIND11_INTERNALS_KIND PYBIND11_COMPILER_TYPE PYBIND11_STDLIB PYBIND11_BUILD_ABI             \
+        PYBIND11_BUILD_TYPE

include/pybind11/conduit/wrap_include_python_h.h

@@ -0,0 +1,72 @@
+#pragma once
+
+// Copyright (c) 2024 The pybind Community.
+
+// STRONG REQUIREMENT:
+//   This header is a wrapper around `#include <Python.h>`, therefore it
+//   MUST BE INCLUDED BEFORE ANY STANDARD HEADERS are included.
+// See also:
+//   https://docs.python.org/3/c-api/intro.html#include-files
+// Quoting from there:
+//   Note: Since Python may define some pre-processor definitions which affect
+//   the standard headers on some systems, you must include Python.h before
+//   any standard headers are included.
+
+// To maximize reusability:
+// DO NOT ADD CODE THAT REQUIRES C++ EXCEPTION HANDLING.
+
+// Disable linking to pythonX_d.lib on Windows in debug mode.
+#if defined(_MSC_VER) && defined(_DEBUG) && !defined(Py_DEBUG)
+// Workaround for a VS 2022 issue.
+// See https://github.com/pybind/pybind11/pull/3497 for full context.
+// NOTE: This workaround knowingly violates the Python.h include order
+//       requirement (see above).
+#    include <yvals.h>
+#    if _MSVC_STL_VERSION >= 143
+#        include <crtdefs.h>
+#    endif
+#    define PYBIND11_DEBUG_MARKER
+#    undef _DEBUG
+#endif
+
+// Don't let Python.h #define (v)snprintf as macro because they are implemented
+// properly in Visual Studio since 2015.
+#if defined(_MSC_VER)
+#    define HAVE_SNPRINTF 1
+#endif
+
+#if defined(_MSC_VER)
+#    pragma warning(push)
+#    pragma warning(disable : 4505)
+// C4505: 'PySlice_GetIndicesEx': unreferenced local function has been removed
+#endif
+
+#include <Python.h>
+#include <frameobject.h>
+#include <pythread.h>
+
+#if defined(_MSC_VER)
+#    pragma warning(pop)
+#endif
+
+#if defined(PYBIND11_DEBUG_MARKER)
+#    define _DEBUG
+#    undef PYBIND11_DEBUG_MARKER
+#endif
+
+// Python #defines overrides on all sorts of core functions, which
+// tends to wreak havok in C++ codebases that expect these to work
+// like regular functions (potentially with several overloads).
+#if defined(isalnum)
+#    undef isalnum
+#    undef isalpha
+#    undef islower
+#    undef isspace
+#    undef isupper
+#    undef tolower
+#    undef toupper
+#endif
+
+#if defined(copysign)
+#    undef copysign
+#endif

include/pybind11/detail/class.h

@@ -583,9 +583,9 @@ extern "C" inline int pybind11_getbuffer(PyObject *obj, Py_buffer *view, int fla
         return -1;
     }
     std::memset(view, 0, sizeof(Py_buffer));
-    buffer_info *info = nullptr;
+    std::unique_ptr<buffer_info> info = nullptr;
     try {
-        info = tinfo->get_buffer(obj, tinfo->get_buffer_data);
+        info.reset(tinfo->get_buffer(obj, tinfo->get_buffer_data));
     } catch (...) {
         try_translate_exceptions();
         raise_from(PyExc_BufferError, "Error getting buffer");
@@ -596,29 +596,72 @@ extern "C" inline int pybind11_getbuffer(PyObject *obj, Py_buffer *view, int fla
     }
 
     if ((flags & PyBUF_WRITABLE) == PyBUF_WRITABLE && info->readonly) {
-        delete info;
         // view->obj = nullptr;  // Was just memset to 0, so not necessary
         set_error(PyExc_BufferError, "Writable buffer requested for readonly storage");
         return -1;
     }
-    view->obj = obj;
-    view->ndim = 1;
-    view->internal = info;
-    view->buf = info->ptr;
+
+    // Fill in all the information, and then downgrade as requested by the caller, or raise an
+    // error if that's not possible.
     view->itemsize = info->itemsize;
     view->len = view->itemsize;
     for (auto s : info->shape) {
         view->len *= s;
     }
+    view->ndim = static_cast<int>(info->ndim);
+    view->shape = info->shape.data();
+    view->strides = info->strides.data();
     view->readonly = static_cast<int>(info->readonly);
     if ((flags & PyBUF_FORMAT) == PyBUF_FORMAT) {
         view->format = const_cast<char *>(info->format.c_str());
     }
-    if ((flags & PyBUF_STRIDES) == PyBUF_STRIDES) {
-        view->ndim = (int) info->ndim;
-        view->strides = info->strides.data();
-        view->shape = info->shape.data();
+
+    // Note, all contiguity flags imply PyBUF_STRIDES and lower.
+    if ((flags & PyBUF_C_CONTIGUOUS) == PyBUF_C_CONTIGUOUS) {
+        if (PyBuffer_IsContiguous(view, 'C') == 0) {
+            std::memset(view, 0, sizeof(Py_buffer));
+            set_error(PyExc_BufferError,
+                      "C-contiguous buffer requested for discontiguous storage");
+            return -1;
         }
+    } else if ((flags & PyBUF_F_CONTIGUOUS) == PyBUF_F_CONTIGUOUS) {
+        if (PyBuffer_IsContiguous(view, 'F') == 0) {
+            std::memset(view, 0, sizeof(Py_buffer));
+            set_error(PyExc_BufferError,
+                      "Fortran-contiguous buffer requested for discontiguous storage");
+            return -1;
+        }
+    } else if ((flags & PyBUF_ANY_CONTIGUOUS) == PyBUF_ANY_CONTIGUOUS) {
+        if (PyBuffer_IsContiguous(view, 'A') == 0) {
+            std::memset(view, 0, sizeof(Py_buffer));
+            set_error(PyExc_BufferError, "Contiguous buffer requested for discontiguous storage");
+            return -1;
+        }
+
+    } else if ((flags & PyBUF_STRIDES) != PyBUF_STRIDES) {
+        // If no strides are requested, the buffer must be C-contiguous.
+        // https://docs.python.org/3/c-api/buffer.html#contiguity-requests
+        if (PyBuffer_IsContiguous(view, 'C') == 0) {
+            std::memset(view, 0, sizeof(Py_buffer));
+            set_error(PyExc_BufferError,
+                      "C-contiguous buffer requested for discontiguous storage");
+            return -1;
+        }
+
+        view->strides = nullptr;
+
+        // Since this is a contiguous buffer, it can also pretend to be 1D.
+        if ((flags & PyBUF_ND) != PyBUF_ND) {
+            view->shape = nullptr;
+            view->ndim = 0;
+        }
+    }
+
+    // Set these after all checks so they don't leak out into the caller, and can be automatically
+    // cleaned up on error.
+    view->buf = info->ptr;
+    view->internal = info.release();
+    view->obj = obj;
     Py_INCREF(view->obj);
     return 0;
 }

include/pybind11/detail/common.h

@@ -9,6 +9,11 @@
 
 #pragma once
 
+#include <pybind11/conduit/wrap_include_python_h.h>
+#if PY_VERSION_HEX < 0x03080000
+#    error "PYTHON < 3.8 IS UNSUPPORTED. pybind11 v2.13 was the last to support Python 3.7."
+#endif
+
 #define PYBIND11_VERSION_MAJOR 2
 #define PYBIND11_VERSION_MINOR 14
 #define PYBIND11_VERSION_PATCH 0.dev1
@@ -204,31 +209,6 @@
 #    define PYBIND11_MAYBE_UNUSED __attribute__((__unused__))
 #endif
 
-/* Don't let Python.h #define (v)snprintf as macro because they are implemented
-   properly in Visual Studio since 2015. */
-#if defined(_MSC_VER)
-#    define HAVE_SNPRINTF 1
-#endif
-
-/// Include Python header, disable linking to pythonX_d.lib on Windows in debug mode
-#if defined(_MSC_VER)
-PYBIND11_WARNING_PUSH
-PYBIND11_WARNING_DISABLE_MSVC(4505)
-// C4505: 'PySlice_GetIndicesEx': unreferenced local function has been removed (PyPy only)
-#    if defined(_DEBUG) && !defined(Py_DEBUG)
-// Workaround for a VS 2022 issue.
-// NOTE: This workaround knowingly violates the Python.h include order requirement:
-// https://docs.python.org/3/c-api/intro.html#include-files
-// See https://github.com/pybind/pybind11/pull/3497 for full context.
-#        include <yvals.h>
-#        if _MSVC_STL_VERSION >= 143
-#            include <crtdefs.h>
-#        endif
-#        define PYBIND11_DEBUG_MARKER
-#        undef _DEBUG
-#    endif
-#endif
-
 // https://en.cppreference.com/w/c/chrono/localtime
 #if defined(__STDC_LIB_EXT1__) && !defined(__STDC_WANT_LIB_EXT1__)
 #    define __STDC_WANT_LIB_EXT1__
@@ -263,30 +243,6 @@ PYBIND11_WARNING_DISABLE_MSVC(4505)
 #    endif
 #endif
 
-#include <Python.h>
-#if PY_VERSION_HEX < 0x03080000
-#    error "PYTHON < 3.8 IS UNSUPPORTED. pybind11 v2.13 was the last to support Python 3.7."
-#endif
-#include <frameobject.h>
-#include <pythread.h>
-
-/* Python #defines overrides on all sorts of core functions, which
-   tends to weak havok in C++ codebases that expect these to work
-   like regular functions (potentially with several overloads) */
-#if defined(isalnum)
-#    undef isalnum
-#    undef isalpha
-#    undef islower
-#    undef isspace
-#    undef isupper
-#    undef tolower
-#    undef toupper
-#endif
-
-#if defined(copysign)
-#    undef copysign
-#endif
-
 #if defined(PYBIND11_NUMPY_1_ONLY)
 #    define PYBIND11_INTERNAL_NUMPY_1_ONLY_DETECTED
 #endif
@@ -295,14 +251,6 @@ PYBIND11_WARNING_DISABLE_MSVC(4505)
 #    define PYBIND11_SIMPLE_GIL_MANAGEMENT
 #endif
 
-#if defined(_MSC_VER)
-#    if defined(PYBIND11_DEBUG_MARKER)
-#        define _DEBUG
-#        undef PYBIND11_DEBUG_MARKER
-#    endif
-PYBIND11_WARNING_POP
-#endif
-
 #include <cstddef>
 #include <cstring>
 #include <exception>
@@ -1145,14 +1093,14 @@ struct overload_cast_impl {
     }
 
     template <typename Return, typename Class>
-    constexpr auto operator()(Return (Class::*pmf)(Args...),
-                              std::false_type = {}) const noexcept -> decltype(pmf) {
+    constexpr auto operator()(Return (Class::*pmf)(Args...), std::false_type = {}) const noexcept
+        -> decltype(pmf) {
         return pmf;
     }
 
     template <typename Return, typename Class>
-    constexpr auto operator()(Return (Class::*pmf)(Args...) const,
-                              std::true_type) const noexcept -> decltype(pmf) {
+    constexpr auto operator()(Return (Class::*pmf)(Args...) const, std::true_type) const noexcept
+        -> decltype(pmf) {
         return pmf;
     }
 };

include/pybind11/detail/descr.h

@@ -156,9 +156,8 @@ constexpr auto concat(const descr<N, Ts...> &d, const Args &...args) {
 }
 #else
 template <size_t N, typename... Ts, typename... Args>
-constexpr auto concat(const descr<N, Ts...> &d,
-                      const Args &...args) -> decltype(std::declval<descr<N + 2, Ts...>>()
-                                                       + concat(args...)) {
+constexpr auto concat(const descr<N, Ts...> &d, const Args &...args)
+    -> decltype(std::declval<descr<N + 2, Ts...>>() + concat(args...)) {
     return d + const_name(", ") + concat(args...);
 }
 #endif

include/pybind11/detail/init.h

@@ -410,7 +410,7 @@ struct pickle_factory<Get, Set, RetState(Self), NewInstance(ArgState)> {
 
     template <typename Class, typename... Extra>
     void execute(Class &cl, const Extra &...extra) && {
-        cl.def("__getstate__", std::move(get));
+        cl.def("__getstate__", std::move(get), pos_only());
 
 #if defined(PYBIND11_CPP14)
         cl.def(

include/pybind11/detail/internals.h

@@ -15,6 +15,7 @@
 #    include <pybind11/gil.h>
 #endif
 
+#include <pybind11/conduit/pybind11_platform_abi_id.h>
 #include <pybind11/pytypes.h>
 
 #include <exception>
@@ -269,67 +270,6 @@ struct type_info {
     bool module_local : 1;
 };
 
-/// On MSVC, debug and release builds are not ABI-compatible!
-#if defined(_MSC_VER) && defined(_DEBUG)
-#    define PYBIND11_BUILD_TYPE "_debug"
-#else
-#    define PYBIND11_BUILD_TYPE ""
-#endif
-
-/// Let's assume that different compilers are ABI-incompatible.
-/// A user can manually set this string if they know their
-/// compiler is compatible.
-#ifndef PYBIND11_COMPILER_TYPE
-#    if defined(_MSC_VER)
-#        define PYBIND11_COMPILER_TYPE "_msvc"
-#    elif defined(__INTEL_COMPILER)
-#        define PYBIND11_COMPILER_TYPE "_icc"
-#    elif defined(__clang__)
-#        define PYBIND11_COMPILER_TYPE "_clang"
-#    elif defined(__PGI)
-#        define PYBIND11_COMPILER_TYPE "_pgi"
-#    elif defined(__MINGW32__)
-#        define PYBIND11_COMPILER_TYPE "_mingw"
-#    elif defined(__CYGWIN__)
-#        define PYBIND11_COMPILER_TYPE "_gcc_cygwin"
-#    elif defined(__GNUC__)
-#        define PYBIND11_COMPILER_TYPE "_gcc"
-#    else
-#        define PYBIND11_COMPILER_TYPE "_unknown"
-#    endif
-#endif
-
-/// Also standard libs
-#ifndef PYBIND11_STDLIB
-#    if defined(_LIBCPP_VERSION)
-#        define PYBIND11_STDLIB "_libcpp"
-#    elif defined(__GLIBCXX__) || defined(__GLIBCPP__)
-#        define PYBIND11_STDLIB "_libstdcpp"
-#    else
-#        define PYBIND11_STDLIB ""
-#    endif
-#endif
-
-/// On Linux/OSX, changes in __GXX_ABI_VERSION__ indicate ABI incompatibility.
-/// On MSVC, changes in _MSC_VER may indicate ABI incompatibility (#2898).
-#ifndef PYBIND11_BUILD_ABI
-#    if defined(__GXX_ABI_VERSION)
-#        define PYBIND11_BUILD_ABI "_cxxabi" PYBIND11_TOSTRING(__GXX_ABI_VERSION)
-#    elif defined(_MSC_VER)
-#        define PYBIND11_BUILD_ABI "_mscver" PYBIND11_TOSTRING(_MSC_VER)
-#    else
-#        define PYBIND11_BUILD_ABI ""
-#    endif
-#endif
-
-#ifndef PYBIND11_INTERNALS_KIND
-#    define PYBIND11_INTERNALS_KIND ""
-#endif
-
-#define PYBIND11_PLATFORM_ABI_ID                                                                  \
-    PYBIND11_INTERNALS_KIND PYBIND11_COMPILER_TYPE PYBIND11_STDLIB PYBIND11_BUILD_ABI             \
-        PYBIND11_BUILD_TYPE
-
 #define PYBIND11_INTERNALS_ID                                                                     \
     "__pybind11_internals_v" PYBIND11_TOSTRING(PYBIND11_INTERNALS_VERSION)                        \
         PYBIND11_PLATFORM_ABI_ID "__"
@@ -671,8 +611,8 @@ inline std::uint64_t mix64(std::uint64_t z) {
 }
 
 template <typename F>
-inline auto with_instance_map(const void *ptr,
-                              const F &cb) -> decltype(cb(std::declval<instance_map &>())) {
+inline auto with_instance_map(const void *ptr, const F &cb)
+    -> decltype(cb(std::declval<instance_map &>())) {
     auto &internals = get_internals();
 
 #ifdef Py_GIL_DISABLED

include/pybind11/detail/type_caster_base.h

@@ -117,7 +117,6 @@ PYBIND11_NOINLINE void all_type_info_populate(PyTypeObject *t, std::vector<type_
     for (handle parent : reinterpret_borrow<tuple>(t->tp_bases)) {
         check.push_back((PyTypeObject *) parent.ptr());
     }
-
     auto const &type_dict = get_internals().registered_types_py;
     for (size_t i = 0; i < check.size(); i++) {
         auto *type = check[i];
@@ -176,13 +175,7 @@ PYBIND11_NOINLINE void all_type_info_populate(PyTypeObject *t, std::vector<type_
  * The value is cached for the lifetime of the Python type.
  */
 inline const std::vector<detail::type_info *> &all_type_info(PyTypeObject *type) {
-    auto ins = all_type_info_get_cache(type);
-    if (ins.second) {
-        // New cache entry: populate it
-        all_type_info_populate(type, ins.first->second);
-    }
-
-    return ins.first->second;
+    return all_type_info_get_cache(type).first->second;
 }
 
 /**
@@ -1161,14 +1154,14 @@ protected:
        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 *) -> decltype(new T(std::declval<const T>()),
-                                                             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<is_move_constructible<T>::value>>
-    static auto make_move_constructor(const T *) -> decltype(new T(std::declval<T &&>()),
-                                                             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/eigen/tensor.h

@@ -124,9 +124,9 @@ struct eigen_tensor_helper<
 template <typename Type, bool ShowDetails, bool NeedsWriteable = false>
 struct get_tensor_descriptor {
     static constexpr auto details
-        = const_name<NeedsWriteable>(", flags.writeable", "")
-          + const_name<static_cast<int>(Type::Layout) == static_cast<int>(Eigen::RowMajor)>(
-              ", flags.c_contiguous", ", flags.f_contiguous");
+        = const_name<NeedsWriteable>(", flags.writeable", "") + const_name
+              < static_cast<int>(Type::Layout)
+          == static_cast<int>(Eigen::RowMajor) > (", flags.c_contiguous", ", flags.f_contiguous");
     static constexpr auto value
         = const_name("numpy.ndarray[") + npy_format_descriptor<typename Type::Scalar>::name
           + const_name("[") + eigen_tensor_helper<remove_cv_t<Type>>::dimensions_descriptor

include/pybind11/numpy.h

@@ -212,6 +212,7 @@ constexpr int platform_lookup(int I, Ints... Is) {
 }
 
 struct npy_api {
+    // If you change this code, please review `normalized_dtype_num` below.
     enum constants {
         NPY_ARRAY_C_CONTIGUOUS_ = 0x0001,
         NPY_ARRAY_F_CONTIGUOUS_ = 0x0002,
@@ -384,6 +385,74 @@ private:
     }
 };
 
+// This table normalizes typenums by mapping NPY_INT_, NPY_LONG, ... to NPY_INT32_, NPY_INT64, ...
+// This is needed to correctly handle situations where multiple typenums map to the same type,
+// e.g. NPY_LONG_ may be equivalent to NPY_INT_ or NPY_LONGLONG_ despite having a different
+// typenum. The normalized typenum should always match the values used in npy_format_descriptor.
+// If you change this code, please review `enum constants` above.
+static constexpr int normalized_dtype_num[npy_api::NPY_VOID_ + 1] = {
+    // NPY_BOOL_ =>
+    npy_api::NPY_BOOL_,
+    // NPY_BYTE_ =>
+    npy_api::NPY_BYTE_,
+    // NPY_UBYTE_ =>
+    npy_api::NPY_UBYTE_,
+    // NPY_SHORT_ =>
+    npy_api::NPY_INT16_,
+    // NPY_USHORT_ =>
+    npy_api::NPY_UINT16_,
+    // NPY_INT_ =>
+    sizeof(int) == sizeof(std::int16_t)   ? npy_api::NPY_INT16_
+    : sizeof(int) == sizeof(std::int32_t) ? npy_api::NPY_INT32_
+    : sizeof(int) == sizeof(std::int64_t) ? npy_api::NPY_INT64_
+                                          : npy_api::NPY_INT_,
+    // NPY_UINT_ =>
+    sizeof(unsigned int) == sizeof(std::uint16_t)   ? npy_api::NPY_UINT16_
+    : sizeof(unsigned int) == sizeof(std::uint32_t) ? npy_api::NPY_UINT32_
+    : sizeof(unsigned int) == sizeof(std::uint64_t) ? npy_api::NPY_UINT64_
+                                                    : npy_api::NPY_UINT_,
+    // NPY_LONG_ =>
+    sizeof(long) == sizeof(std::int16_t)   ? npy_api::NPY_INT16_
+    : sizeof(long) == sizeof(std::int32_t) ? npy_api::NPY_INT32_
+    : sizeof(long) == sizeof(std::int64_t) ? npy_api::NPY_INT64_
+                                           : npy_api::NPY_LONG_,
+    // NPY_ULONG_ =>
+    sizeof(unsigned long) == sizeof(std::uint16_t)   ? npy_api::NPY_UINT16_
+    : sizeof(unsigned long) == sizeof(std::uint32_t) ? npy_api::NPY_UINT32_
+    : sizeof(unsigned long) == sizeof(std::uint64_t) ? npy_api::NPY_UINT64_
+                                                     : npy_api::NPY_ULONG_,
+    // NPY_LONGLONG_ =>
+    sizeof(long long) == sizeof(std::int16_t)   ? npy_api::NPY_INT16_
+    : sizeof(long long) == sizeof(std::int32_t) ? npy_api::NPY_INT32_
+    : sizeof(long long) == sizeof(std::int64_t) ? npy_api::NPY_INT64_
+                                                : npy_api::NPY_LONGLONG_,
+    // NPY_ULONGLONG_ =>
+    sizeof(unsigned long long) == sizeof(std::uint16_t)   ? npy_api::NPY_UINT16_
+    : sizeof(unsigned long long) == sizeof(std::uint32_t) ? npy_api::NPY_UINT32_
+    : sizeof(unsigned long long) == sizeof(std::uint64_t) ? npy_api::NPY_UINT64_
+                                                          : npy_api::NPY_ULONGLONG_,
+    // NPY_FLOAT_ =>
+    npy_api::NPY_FLOAT_,
+    // NPY_DOUBLE_ =>
+    npy_api::NPY_DOUBLE_,
+    // NPY_LONGDOUBLE_ =>
+    npy_api::NPY_LONGDOUBLE_,
+    // NPY_CFLOAT_ =>
+    npy_api::NPY_CFLOAT_,
+    // NPY_CDOUBLE_ =>
+    npy_api::NPY_CDOUBLE_,
+    // NPY_CLONGDOUBLE_ =>
+    npy_api::NPY_CLONGDOUBLE_,
+    // NPY_OBJECT_ =>
+    npy_api::NPY_OBJECT_,
+    // NPY_STRING_ =>
+    npy_api::NPY_STRING_,
+    // NPY_UNICODE_ =>
+    npy_api::NPY_UNICODE_,
+    // NPY_VOID_ =>
+    npy_api::NPY_VOID_,
+};
+
 inline PyArray_Proxy *array_proxy(void *ptr) { return reinterpret_cast<PyArray_Proxy *>(ptr); }
 
 inline const PyArray_Proxy *array_proxy(const void *ptr) {
@@ -684,6 +753,13 @@ public:
         return detail::npy_format_descriptor<typename std::remove_cv<T>::type>::dtype();
     }
 
+    /// Return the type number associated with a C++ type.
+    /// This is the constexpr equivalent of `dtype::of<T>().num()`.
+    template <typename T>
+    static constexpr int num_of() {
+        return detail::npy_format_descriptor<typename std::remove_cv<T>::type>::value;
+    }
+
     /// Size of the data type in bytes.
 #ifdef PYBIND11_NUMPY_1_ONLY
     ssize_t itemsize() const { return detail::array_descriptor_proxy(m_ptr)->elsize; }
@@ -725,7 +801,9 @@ public:
         return detail::array_descriptor_proxy(m_ptr)->type;
     }
 
-    /// type number of dtype.
+    /// Type number of dtype. Note that different values may be returned for equivalent types,
+    /// e.g. even though ``long`` may be equivalent to ``int`` or ``long long``, they still have
+    /// different type numbers. Consider using `normalized_num` to avoid this.
     int num() const {
         // Note: The signature, `dtype::num` follows the naming of NumPy's public
         // Python API (i.e., ``dtype.num``), rather than its internal
@@ -733,6 +811,17 @@ public:
         return detail::array_descriptor_proxy(m_ptr)->type_num;
     }
 
+    /// Type number of dtype, normalized to match the return value of `num_of` for equivalent
+    /// types. This function can be used to write switch statements that correctly handle
+    /// equivalent types with different type numbers.
+    int normalized_num() const {
+        int value = num();
+        if (value >= 0 && value <= detail::npy_api::NPY_VOID_) {
+            return detail::normalized_dtype_num[value];
+        }
+        return value;
+    }
+
     /// Single character for byteorder
     char byteorder() const { return detail::array_descriptor_proxy(m_ptr)->byteorder; }
 
@@ -1232,9 +1321,7 @@ public:
     // Reference to element at a given index
     template <typename... Ix>
     const T &at(Ix... index) const {
-        if ((ssize_t) sizeof...(index) != ndim()) {
-            fail_dim_check(sizeof...(index), "index dimension mismatch");
-        }
+        check_rank_precondition(sizeof...(index));
         return *(static_cast<const T *>(array::data())
                  + byte_offset(ssize_t(index)...) / itemsize());
     }
@@ -1242,13 +1329,33 @@ public:
     // Mutable reference to element at a given index
     template <typename... Ix>
     T &mutable_at(Ix... index) {
-        if ((ssize_t) sizeof...(index) != ndim()) {
-            fail_dim_check(sizeof...(index), "index dimension mismatch");
-        }
+        check_rank_precondition(sizeof...(index));
         return *(static_cast<T *>(array::mutable_data())
                  + byte_offset(ssize_t(index)...) / itemsize());
     }
 
+    // const-reference to element at a given index without bounds checking
+    template <typename... Ix>
+    const T &operator()(Ix... index) const {
+#if !defined(NDEBUG)
+        check_rank_precondition(sizeof...(index));
+        check_dimensions(index...);
+#endif
+        return *(static_cast<const T *>(array::data())
+                 + detail::byte_offset_unsafe(strides(), ssize_t(index)...) / itemsize());
+    }
+
+    // mutable reference to element at a given index without bounds checking
+    template <typename... Ix>
+    T &operator()(Ix... index) {
+#if !defined(NDEBUG)
+        check_rank_precondition(sizeof...(index));
+        check_dimensions(index...);
+#endif
+        return *(static_cast<T *>(array::mutable_data())
+                 + detail::byte_offset_unsafe(strides(), ssize_t(index)...) / itemsize());
+    }
+
     /**
      * Returns a proxy object that provides access to the array's data without bounds or
      * dimensionality checking.  Will throw if the array is missing the `writeable` flag.  Use with
@@ -1305,6 +1412,13 @@ protected:
                                                            | ExtraFlags,
                                                        nullptr);
     }
+
+private:
+    void check_rank_precondition(ssize_t dim) const {
+        if (dim != ndim()) {
+            fail_dim_check(dim, "index dimension mismatch");
+        }
+    }
 };
 
 template <typename T>
@@ -1428,7 +1542,11 @@ public:
 };
 
 template <typename T>
-struct npy_format_descriptor<T, enable_if_t<is_same_ignoring_cvref<T, PyObject *>::value>> {
+struct npy_format_descriptor<
+    T,
+    enable_if_t<is_same_ignoring_cvref<T, PyObject *>::value
+                || ((std::is_same<T, handle>::value || std::is_same<T, object>::value)
+                    && sizeof(T) == sizeof(PyObject *))>> {
     static constexpr auto name = const_name("object");
 
     static constexpr int value = npy_api::NPY_OBJECT_;

include/pybind11/pybind11.h

@@ -301,9 +301,20 @@ protected:
             constexpr bool has_kw_only_args = any_of<std::is_same<kw_only, Extra>...>::value,
                            has_pos_only_args = any_of<std::is_same<pos_only, Extra>...>::value,
                            has_arg_annotations = any_of<is_keyword<Extra>...>::value;
+            constexpr bool has_is_method = any_of<std::is_same<is_method, Extra>...>::value;
+            // The implicit `self` argument is not present and not counted in method definitions.
+            constexpr bool has_args = cast_in::args_pos >= 0;
+            constexpr bool is_method_with_self_arg_only = has_is_method && !has_args;
             static_assert(has_arg_annotations || !has_kw_only_args,
                           "py::kw_only requires the use of argument annotations");
-            static_assert(has_arg_annotations || !has_pos_only_args,
+            static_assert(((/* Need `py::arg("arg_name")` annotation in function/method. */
+                            has_arg_annotations)
+                           || (/* Allow methods with no arguments `def method(self, /): ...`.
+                                * A method has at least one argument `self`. There can be no
+                                * `py::arg` annotation. E.g. `class.def("method", py::pos_only())`.
+                                */
+                               is_method_with_self_arg_only))
+                              || !has_pos_only_args,
                           "py::pos_only requires the use of argument annotations (for docstrings "
                           "and aligning the annotations to the argument)");
 
@@ -2022,9 +2033,11 @@ struct enum_base {
                     .format(std::move(type_name), enum_name(arg), int_(arg));
             },
             name("__repr__"),
-            is_method(m_base));
+            is_method(m_base),
+            pos_only());
 
-        m_base.attr("name") = property(cpp_function(&enum_name, name("name"), is_method(m_base)));
+        m_base.attr("name")
+            = property(cpp_function(&enum_name, name("name"), is_method(m_base), pos_only()));
 
         m_base.attr("__str__") = cpp_function(
             [](handle arg) -> str {
@@ -2032,7 +2045,8 @@ struct enum_base {
                 return pybind11::str("{}.{}").format(std::move(type_name), enum_name(arg));
             },
             name("__str__"),
-            is_method(m_base));
+            is_method(m_base),
+            pos_only());
 
         if (options::show_enum_members_docstring()) {
             m_base.attr("__doc__") = static_property(
@@ -2087,7 +2101,8 @@ struct enum_base {
         },                                                                                        \
         name(op),                                                                                 \
         is_method(m_base),                                                                        \
-        arg("other"))
+        arg("other"),                                                                             \
+        pos_only())
 
 #define PYBIND11_ENUM_OP_CONV(op, expr)                                                           \
     m_base.attr(op) = cpp_function(                                                               \
@@ -2097,7 +2112,8 @@ struct enum_base {
         },                                                                                        \
         name(op),                                                                                 \
         is_method(m_base),                                                                        \
-        arg("other"))
+        arg("other"),                                                                             \
+        pos_only())
 
 #define PYBIND11_ENUM_OP_CONV_LHS(op, expr)                                                       \
     m_base.attr(op) = cpp_function(                                                               \
@@ -2107,7 +2123,8 @@ struct enum_base {
         },                                                                                        \
         name(op),                                                                                 \
         is_method(m_base),                                                                        \
-        arg("other"))
+        arg("other"),                                                                             \
+        pos_only())
 
         if (is_convertible) {
             PYBIND11_ENUM_OP_CONV_LHS("__eq__", !b.is_none() && a.equal(b));
@@ -2127,7 +2144,8 @@ struct enum_base {
                 m_base.attr("__invert__")
                     = cpp_function([](const object &arg) { return ~(int_(arg)); },
                                    name("__invert__"),
-                                   is_method(m_base));
+                                   is_method(m_base),
+                                   pos_only());
             }
         } else {
             PYBIND11_ENUM_OP_STRICT("__eq__", int_(a).equal(int_(b)), return false);
@@ -2147,11 +2165,15 @@ struct enum_base {
 #undef PYBIND11_ENUM_OP_CONV
 #undef PYBIND11_ENUM_OP_STRICT
 
-        m_base.attr("__getstate__") = cpp_function(
-            [](const object &arg) { return int_(arg); }, name("__getstate__"), is_method(m_base));
+        m_base.attr("__getstate__") = cpp_function([](const object &arg) { return int_(arg); },
+                                                   name("__getstate__"),
+                                                   is_method(m_base),
+                                                   pos_only());
 
-        m_base.attr("__hash__") = cpp_function(
-            [](const object &arg) { return int_(arg); }, name("__hash__"), is_method(m_base));
+        m_base.attr("__hash__") = cpp_function([](const object &arg) { return int_(arg); },
+                                               name("__hash__"),
+                                               is_method(m_base),
+                                               pos_only());
     }
 
     PYBIND11_NOINLINE void value(char const *name_, object value, const char *doc = nullptr) {
@@ -2243,9 +2265,9 @@ public:
         m_base.init(is_arithmetic, is_convertible);
 
         def(init([](Scalar i) { return static_cast<Type>(i); }), arg("value"));
-        def_property_readonly("value", [](Type value) { return (Scalar) value; });
-        def("__int__", [](Type value) { return (Scalar) value; });
-        def("__index__", [](Type value) { return (Scalar) value; });
+        def_property_readonly("value", [](Type value) { return (Scalar) value; }, pos_only());
+        def("__int__", [](Type value) { return (Scalar) value; }, pos_only());
+        def("__index__", [](Type value) { return (Scalar) value; }, pos_only());
         attr("__setstate__") = cpp_function(
             [](detail::value_and_holder &v_h, Scalar arg) {
                 detail::initimpl::setstate<Base>(
@@ -2254,7 +2276,8 @@ public:
             detail::is_new_style_constructor(),
             pybind11::name("__setstate__"),
             is_method(*this),
-            arg("state"));
+            arg("state"),
+            pos_only());
     }
 
     /// Export enumeration entries into the parent scope
@@ -2326,13 +2349,20 @@ keep_alive_impl(size_t Nurse, size_t Patient, function_call &call, handle ret) {
 inline std::pair<decltype(internals::registered_types_py)::iterator, bool>
 all_type_info_get_cache(PyTypeObject *type) {
     auto res = with_internals([type](internals &internals) {
-        return internals
+        auto ins = internals
                        .registered_types_py
 #ifdef __cpp_lib_unordered_map_try_emplace
                        .try_emplace(type);
 #else
                        .emplace(type, std::vector<detail::type_info *>());
 #endif
+        if (ins.second) {
+            // For free-threading mode, this call must be under
+            // the with_internals() mutex lock, to avoid that other threads
+            // continue running with the empty ins.first->second.
+            all_type_info_populate(type, ins.first->second);
+        }
+        return ins;
     });
     if (res.second) {
         // New cache entry created; set up a weak reference to automatically remove it if the type
@@ -2433,7 +2463,8 @@ iterator make_iterator_impl(Iterator first, Sentinel last, Extra &&...extra) {
 
     if (!detail::get_type_info(typeid(state), false)) {
         class_<state>(handle(), "iterator", pybind11::module_local())
-            .def("__iter__", [](state &s) -> state & { return s; })
+            .def(
+                "__iter__", [](state &s) -> state & { return s; }, pos_only())
             .def(
                 "__next__",
                 [](state &s) -> ValueType {
@@ -2450,6 +2481,7 @@ iterator make_iterator_impl(Iterator first, Sentinel last, Extra &&...extra) {
                     // NOLINTNEXTLINE(readability-const-return-type) // PR #3263
                 },
                 std::forward<Extra>(extra)...,
+                pos_only(),
                 Policy);
     }
 

include/pybind11/pytypes.h

@@ -2226,6 +2226,18 @@ class kwargs : public dict {
     PYBIND11_OBJECT_DEFAULT(kwargs, dict, PyDict_Check)
 };
 
+// Subclasses of args and kwargs to support type hinting
+// as defined in PEP 484. See #5357 for more info.
+template <typename T>
+class Args : public args {
+    using args::args;
+};
+
+template <typename T>
+class KWArgs : public kwargs {
+    using kwargs::kwargs;
+};
+
 class anyset : public object {
 public:
     PYBIND11_OBJECT(anyset, object, PyAnySet_Check)

tests/CMakeLists.txt

@@ -491,6 +491,12 @@ foreach(target ${test_targets})
   if(SKBUILD)
     install(TARGETS ${target} LIBRARY DESTINATION .)
   endif()
+
+  if("${target}" STREQUAL "exo_planet_c_api")
+    if(CMAKE_CXX_COMPILER_ID MATCHES "(GNU|Intel|Clang|NVHPC)")
+      target_compile_options(${target} PRIVATE -fno-exceptions)
+    endif()
+  endif()
 endforeach()
 
 # Provide nice organisation in IDEs

tests/conftest.py

@@ -198,10 +198,11 @@ def pytest_assertrepr_compare(op, left, right):  # noqa: ARG001
 
 
 def gc_collect():
-    """Run the garbage collector twice (needed when running
+    """Run the garbage collector three times (needed when running
     reference counting tests with PyPy)"""
     gc.collect()
     gc.collect()
+    gc.collect()
 
 
 def pytest_configure():

tests/constructor_stats.h

@@ -312,8 +312,16 @@ void print_created(T *inst, Values &&...values) {
 }
 template <class T, typename... Values>
 void print_destroyed(T *inst, Values &&...values) { // Prints but doesn't store given values
+    /*
+     * On GraalPy, destructors can trigger anywhere and this can cause random
+     * failures in unrelated tests.
+     */
+#if !defined(GRAALVM_PYTHON)
     print_constr_details(inst, "destroyed", values...);
     track_destroyed(inst);
+#else
+    py::detail::silence_unused_warnings(inst, values...);
+#endif
 }
 template <class T, typename... Values>
 void print_values(T *inst, Values &&...values) {

tests/exo_planet_c_api.cpp

@@ -1,59 +1,33 @@
 // Copyright (c) 2024 The pybind Community.
 
+// In production situations it is totally fine to build with
+// C++ Exception Handling enabled. However, here we want to ensure that
+// C++ Exception Handling is not required.
+#if defined(_MSC_VER) || defined(__EMSCRIPTEN__)
+// Too much trouble making the required cmake changes (see PR #5375).
+#else
+#    ifdef __cpp_exceptions
+// https://isocpp.org/std/standing-documents/sd-6-sg10-feature-test-recommendations#__cpp_exceptions
+#    error This test is meant to be built with C++ Exception Handling disabled, but __cpp_exceptions is defined.
+#    endif
+#    ifdef __EXCEPTIONS
+// https://gcc.gnu.org/onlinedocs/cpp/Common-Predefined-Macros.html
+#    error This test is meant to be built with C++ Exception Handling disabled, but __EXCEPTIONS is defined.
+#    endif
+#endif
+
 // THIS MUST STAY AT THE TOP!
-#include <pybind11/pybind11.h> // EXCLUSIVELY for PYBIND11_PLATFORM_ABI_ID
-// Potential future direction to maximize reusability:
-// (e.g. for use from SWIG, Cython, PyCLIF, nanobind):
-//     #include <pybind11/compat/platform_abi_id.h>
-// This would only depend on:
-//     1. A C++ compiler, WITHOUT requiring -fexceptions.
-//     2. Python.h
+#include <pybind11/conduit/pybind11_conduit_v1.h> // VERY light-weight dependency.
 
 #include "test_cpp_conduit_traveler_types.h"
 
 #include <Python.h>
-#include <typeinfo>
 
 namespace {
 
-void *get_cpp_conduit_void_ptr(PyObject *py_obj, const std::type_info *cpp_type_info) {
-    PyObject *cpp_type_info_capsule
-        = PyCapsule_New(const_cast<void *>(static_cast<const void *>(cpp_type_info)),
-                        typeid(std::type_info).name(),
-                        nullptr);
-    if (cpp_type_info_capsule == nullptr) {
-        return nullptr;
-    }
-    PyObject *cpp_conduit = PyObject_CallMethod(py_obj,
-                                                "_pybind11_conduit_v1_",
-                                                "yOy",
-                                                PYBIND11_PLATFORM_ABI_ID,
-                                                cpp_type_info_capsule,
-                                                "raw_pointer_ephemeral");
-    Py_DECREF(cpp_type_info_capsule);
-    if (cpp_conduit == nullptr) {
-        return nullptr;
-    }
-    void *void_ptr = PyCapsule_GetPointer(cpp_conduit, cpp_type_info->name());
-    Py_DECREF(cpp_conduit);
-    if (PyErr_Occurred()) {
-        return nullptr;
-    }
-    return void_ptr;
-}
-
-template <typename T>
-T *get_cpp_conduit_type_ptr(PyObject *py_obj) {
-    void *void_ptr = get_cpp_conduit_void_ptr(py_obj, &typeid(T));
-    if (void_ptr == nullptr) {
-        return nullptr;
-    }
-    return static_cast<T *>(void_ptr);
-}
-
 extern "C" PyObject *wrapGetLuggage(PyObject * /*self*/, PyObject *traveler) {
-    const auto *cpp_traveler
-        = get_cpp_conduit_type_ptr<pybind11_tests::test_cpp_conduit::Traveler>(traveler);
+    const auto *cpp_traveler = pybind11_conduit_v1::get_type_pointer_ephemeral<
+        pybind11_tests::test_cpp_conduit::Traveler>(traveler);
     if (cpp_traveler == nullptr) {
         return nullptr;
     }
@@ -61,9 +35,8 @@ extern "C" PyObject *wrapGetLuggage(PyObject * /*self*/, PyObject *traveler) {
 }
 
 extern "C" PyObject *wrapGetPoints(PyObject * /*self*/, PyObject *premium_traveler) {
-    const auto *cpp_premium_traveler
-        = get_cpp_conduit_type_ptr<pybind11_tests::test_cpp_conduit::PremiumTraveler>(
-            premium_traveler);
+    const auto *cpp_premium_traveler = pybind11_conduit_v1::get_type_pointer_ephemeral<
+        pybind11_tests::test_cpp_conduit::PremiumTraveler>(premium_traveler);
     if (cpp_premium_traveler == nullptr) {
         return nullptr;
     }

tests/extra_python_package/test_files.py

@@ -50,6 +50,13 @@ main_headers = {
     "include/pybind11/warnings.h",
 }
 
+conduit_headers = {
+    "include/pybind11/conduit/README.txt",
+    "include/pybind11/conduit/pybind11_conduit_v1.h",
+    "include/pybind11/conduit/pybind11_platform_abi_id.h",
+    "include/pybind11/conduit/wrap_include_python_h.h",
+}
+
 detail_headers = {
     "include/pybind11/detail/class.h",
     "include/pybind11/detail/common.h",
@@ -97,7 +104,7 @@ py_files = {
     "setup_helpers.py",
 }
 
-headers = main_headers | detail_headers | eigen_headers | stl_headers
+headers = main_headers | conduit_headers | detail_headers | eigen_headers | stl_headers
 src_files = headers | cmake_files | pkgconfig_files
 all_files = src_files | py_files
 
@@ -106,6 +113,7 @@ sdist_files = {
     "pybind11",
     "pybind11/include",
     "pybind11/include/pybind11",
+    "pybind11/include/pybind11/conduit",
     "pybind11/include/pybind11/detail",
     "pybind11/include/pybind11/eigen",
     "pybind11/include/pybind11/stl",

tests/pybind11_tests.cpp

@@ -128,4 +128,9 @@ PYBIND11_MODULE(pybind11_tests, m, py::mod_gil_not_used()) {
     for (const auto &initializer : initializers()) {
         initializer(m);
     }
+
+    py::class_<TestContext>(m, "TestContext")
+        .def(py::init<>(&TestContext::createNewContextForInit))
+        .def("__enter__", &TestContext::contextEnter)
+        .def("__exit__", &TestContext::contextExit);
 }

tests/pybind11_tests.h

@@ -96,3 +96,24 @@ void ignoreOldStyleInitWarnings(F &&body) {
     )",
              py::dict(py::arg("body") = py::cpp_function(body)));
 }
+
+// See PR #5419 for background.
+class TestContext {
+public:
+    TestContext() = delete;
+    TestContext(const TestContext &) = delete;
+    TestContext(TestContext &&) = delete;
+    static TestContext *createNewContextForInit() { return new TestContext("new-context"); }
+
+    pybind11::object contextEnter() {
+        py::object contextObj = py::cast(*this);
+        return contextObj;
+    }
+    void contextExit(const pybind11::object & /*excType*/,
+                     const pybind11::object & /*excVal*/,
+                     const pybind11::object & /*excTb*/) {}
+
+private:
+    explicit TestContext(const std::string &context) : context(context) {}
+    std::string context;
+};

tests/test_buffers.cpp

@@ -167,6 +167,125 @@ TEST_SUBMODULE(buffers, m) {
                  sizeof(float)});
         });
 
+    // A matrix that uses Fortran storage order.
+    class FortranMatrix : public Matrix {
+    public:
+        FortranMatrix(py::ssize_t rows, py::ssize_t cols) : Matrix(cols, rows) {
+            print_created(this,
+                          std::to_string(rows) + "x" + std::to_string(cols) + " Fortran matrix");
+        }
+
+        float operator()(py::ssize_t i, py::ssize_t j) const { return Matrix::operator()(j, i); }
+
+        float &operator()(py::ssize_t i, py::ssize_t j) { return Matrix::operator()(j, i); }
+
+        using Matrix::data;
+
+        py::ssize_t rows() const { return Matrix::cols(); }
+        py::ssize_t cols() const { return Matrix::rows(); }
+    };
+    py::class_<FortranMatrix, Matrix>(m, "FortranMatrix", py::buffer_protocol())
+        .def(py::init<py::ssize_t, py::ssize_t>())
+
+        .def("rows", &FortranMatrix::rows)
+        .def("cols", &FortranMatrix::cols)
+
+        /// Bare bones interface
+        .def("__getitem__",
+             [](const FortranMatrix &m, std::pair<py::ssize_t, py::ssize_t> i) {
+                 if (i.first >= m.rows() || i.second >= m.cols()) {
+                     throw py::index_error();
+                 }
+                 return m(i.first, i.second);
+             })
+        .def("__setitem__",
+             [](FortranMatrix &m, std::pair<py::ssize_t, py::ssize_t> i, float v) {
+                 if (i.first >= m.rows() || i.second >= m.cols()) {
+                     throw py::index_error();
+                 }
+                 m(i.first, i.second) = v;
+             })
+        /// Provide buffer access
+        .def_buffer([](FortranMatrix &m) -> py::buffer_info {
+            return py::buffer_info(m.data(),             /* Pointer to buffer */
+                                   {m.rows(), m.cols()}, /* Buffer dimensions */
+                                   /* Strides (in bytes) for each index */
+                                   {sizeof(float), sizeof(float) * size_t(m.rows())});
+        });
+
+    // A matrix that uses a discontiguous underlying memory block.
+    class DiscontiguousMatrix : public Matrix {
+    public:
+        DiscontiguousMatrix(py::ssize_t rows,
+                            py::ssize_t cols,
+                            py::ssize_t row_factor,
+                            py::ssize_t col_factor)
+            : Matrix(rows * row_factor, cols * col_factor), m_row_factor(row_factor),
+              m_col_factor(col_factor) {
+            print_created(this,
+                          std::to_string(rows) + "(*" + std::to_string(row_factor) + ")x"
+                              + std::to_string(cols) + "(*" + std::to_string(col_factor)
+                              + ") matrix");
+        }
+
+        ~DiscontiguousMatrix() {
+            print_destroyed(this,
+                            std::to_string(rows() / m_row_factor) + "(*"
+                                + std::to_string(m_row_factor) + ")x"
+                                + std::to_string(cols() / m_col_factor) + "(*"
+                                + std::to_string(m_col_factor) + ") matrix");
+        }
+
+        float operator()(py::ssize_t i, py::ssize_t j) const {
+            return Matrix::operator()(i * m_row_factor, j * m_col_factor);
+        }
+
+        float &operator()(py::ssize_t i, py::ssize_t j) {
+            return Matrix::operator()(i * m_row_factor, j * m_col_factor);
+        }
+
+        using Matrix::data;
+
+        py::ssize_t rows() const { return Matrix::rows() / m_row_factor; }
+        py::ssize_t cols() const { return Matrix::cols() / m_col_factor; }
+        py::ssize_t row_factor() const { return m_row_factor; }
+        py::ssize_t col_factor() const { return m_col_factor; }
+
+    private:
+        py::ssize_t m_row_factor;
+        py::ssize_t m_col_factor;
+    };
+    py::class_<DiscontiguousMatrix, Matrix>(m, "DiscontiguousMatrix", py::buffer_protocol())
+        .def(py::init<py::ssize_t, py::ssize_t, py::ssize_t, py::ssize_t>())
+
+        .def("rows", &DiscontiguousMatrix::rows)
+        .def("cols", &DiscontiguousMatrix::cols)
+
+        /// Bare bones interface
+        .def("__getitem__",
+             [](const DiscontiguousMatrix &m, std::pair<py::ssize_t, py::ssize_t> i) {
+                 if (i.first >= m.rows() || i.second >= m.cols()) {
+                     throw py::index_error();
+                 }
+                 return m(i.first, i.second);
+             })
+        .def("__setitem__",
+             [](DiscontiguousMatrix &m, std::pair<py::ssize_t, py::ssize_t> i, float v) {
+                 if (i.first >= m.rows() || i.second >= m.cols()) {
+                     throw py::index_error();
+                 }
+                 m(i.first, i.second) = v;
+             })
+        /// Provide buffer access
+        .def_buffer([](DiscontiguousMatrix &m) -> py::buffer_info {
+            return py::buffer_info(m.data(),             /* Pointer to buffer */
+                                   {m.rows(), m.cols()}, /* Buffer dimensions */
+                                   /* Strides (in bytes) for each index */
+                                   {size_t(m.col_factor()) * sizeof(float) * size_t(m.cols())
+                                        * size_t(m.row_factor()),
+                                    size_t(m.col_factor()) * sizeof(float)});
+        });
+
     class BrokenMatrix : public Matrix {
     public:
         BrokenMatrix(py::ssize_t rows, py::ssize_t cols) : Matrix(rows, cols) {}
@@ -268,4 +387,56 @@ TEST_SUBMODULE(buffers, m) {
         });
 
     m.def("get_buffer_info", [](const py::buffer &buffer) { return buffer.request(); });
+
+    // Expose Py_buffer for testing.
+    m.attr("PyBUF_FORMAT") = PyBUF_FORMAT;
+    m.attr("PyBUF_SIMPLE") = PyBUF_SIMPLE;
+    m.attr("PyBUF_ND") = PyBUF_ND;
+    m.attr("PyBUF_STRIDES") = PyBUF_STRIDES;
+    m.attr("PyBUF_INDIRECT") = PyBUF_INDIRECT;
+    m.attr("PyBUF_C_CONTIGUOUS") = PyBUF_C_CONTIGUOUS;
+    m.attr("PyBUF_F_CONTIGUOUS") = PyBUF_F_CONTIGUOUS;
+    m.attr("PyBUF_ANY_CONTIGUOUS") = PyBUF_ANY_CONTIGUOUS;
+
+    m.def("get_py_buffer", [](const py::object &object, int flags) {
+        Py_buffer buffer;
+        memset(&buffer, 0, sizeof(Py_buffer));
+        if (PyObject_GetBuffer(object.ptr(), &buffer, flags) == -1) {
+            throw py::error_already_set();
+        }
+
+        auto SimpleNamespace = py::module_::import("types").attr("SimpleNamespace");
+        py::object result = SimpleNamespace("len"_a = buffer.len,
+                                            "readonly"_a = buffer.readonly,
+                                            "itemsize"_a = buffer.itemsize,
+                                            "format"_a = buffer.format,
+                                            "ndim"_a = buffer.ndim,
+                                            "shape"_a = py::none(),
+                                            "strides"_a = py::none(),
+                                            "suboffsets"_a = py::none());
+        if (buffer.shape != nullptr) {
+            py::list l;
+            for (auto i = 0; i < buffer.ndim; i++) {
+                l.append(buffer.shape[i]);
+            }
+            py::setattr(result, "shape", l);
+        }
+        if (buffer.strides != nullptr) {
+            py::list l;
+            for (auto i = 0; i < buffer.ndim; i++) {
+                l.append(buffer.strides[i]);
+            }
+            py::setattr(result, "strides", l);
+        }
+        if (buffer.suboffsets != nullptr) {
+            py::list l;
+            for (auto i = 0; i < buffer.ndim; i++) {
+                l.append(buffer.suboffsets[i]);
+            }
+            py::setattr(result, "suboffsets", l);
+        }
+
+        PyBuffer_Release(&buffer);
+        return result;
+    });
 }

tests/test_buffers.py

@@ -239,3 +239,163 @@ def test_buffer_exception():
         memoryview(m.BrokenMatrix(1, 1))
     assert isinstance(excinfo.value.__cause__, RuntimeError)
     assert "for context" in str(excinfo.value.__cause__)
+
+
+@pytest.mark.parametrize("type", ["pybind11", "numpy"])
+def test_c_contiguous_to_pybuffer(type):
+    if type == "pybind11":
+        mat = m.Matrix(5, 4)
+    elif type == "numpy":
+        mat = np.empty((5, 4), dtype=np.float32)
+    else:
+        raise ValueError(f"Unknown parametrization {type}")
+
+    info = m.get_py_buffer(mat, m.PyBUF_SIMPLE)
+    assert info.format is None
+    assert info.itemsize == ctypes.sizeof(ctypes.c_float)
+    assert info.len == 5 * 4 * info.itemsize
+    assert info.ndim == 0  # See discussion on PR #5407.
+    assert info.shape is None
+    assert info.strides is None
+    assert info.suboffsets is None
+    assert not info.readonly
+    info = m.get_py_buffer(mat, m.PyBUF_SIMPLE | m.PyBUF_FORMAT)
+    assert info.format == "f"
+    assert info.itemsize == ctypes.sizeof(ctypes.c_float)
+    assert info.len == 5 * 4 * info.itemsize
+    assert info.ndim == 0  # See discussion on PR #5407.
+    assert info.shape is None
+    assert info.strides is None
+    assert info.suboffsets is None
+    assert not info.readonly
+    info = m.get_py_buffer(mat, m.PyBUF_ND)
+    assert info.itemsize == ctypes.sizeof(ctypes.c_float)
+    assert info.len == 5 * 4 * info.itemsize
+    assert info.ndim == 2
+    assert info.shape == [5, 4]
+    assert info.strides is None
+    assert info.suboffsets is None
+    assert not info.readonly
+    info = m.get_py_buffer(mat, m.PyBUF_STRIDES)
+    assert info.itemsize == ctypes.sizeof(ctypes.c_float)
+    assert info.len == 5 * 4 * info.itemsize
+    assert info.ndim == 2
+    assert info.shape == [5, 4]
+    assert info.strides == [4 * info.itemsize, info.itemsize]
+    assert info.suboffsets is None
+    assert not info.readonly
+    info = m.get_py_buffer(mat, m.PyBUF_INDIRECT)
+    assert info.itemsize == ctypes.sizeof(ctypes.c_float)
+    assert info.len == 5 * 4 * info.itemsize
+    assert info.ndim == 2
+    assert info.shape == [5, 4]
+    assert info.strides == [4 * info.itemsize, info.itemsize]
+    assert info.suboffsets is None  # Should be filled in here, but we don't use it.
+    assert not info.readonly
+
+
+@pytest.mark.parametrize("type", ["pybind11", "numpy"])
+def test_fortran_contiguous_to_pybuffer(type):
+    if type == "pybind11":
+        mat = m.FortranMatrix(5, 4)
+    elif type == "numpy":
+        mat = np.empty((5, 4), dtype=np.float32, order="F")
+    else:
+        raise ValueError(f"Unknown parametrization {type}")
+
+    # A Fortran-shaped buffer can only be accessed at PyBUF_STRIDES level or higher.
+    info = m.get_py_buffer(mat, m.PyBUF_STRIDES)
+    assert info.itemsize == ctypes.sizeof(ctypes.c_float)
+    assert info.len == 5 * 4 * info.itemsize
+    assert info.ndim == 2
+    assert info.shape == [5, 4]
+    assert info.strides == [info.itemsize, 5 * info.itemsize]
+    assert info.suboffsets is None
+    assert not info.readonly
+    info = m.get_py_buffer(mat, m.PyBUF_INDIRECT)
+    assert info.itemsize == ctypes.sizeof(ctypes.c_float)
+    assert info.len == 5 * 4 * info.itemsize
+    assert info.ndim == 2
+    assert info.shape == [5, 4]
+    assert info.strides == [info.itemsize, 5 * info.itemsize]
+    assert info.suboffsets is None  # Should be filled in here, but we don't use it.
+    assert not info.readonly
+
+
+@pytest.mark.parametrize("type", ["pybind11", "numpy"])
+def test_discontiguous_to_pybuffer(type):
+    if type == "pybind11":
+        mat = m.DiscontiguousMatrix(5, 4, 2, 3)
+    elif type == "numpy":
+        mat = np.empty((5 * 2, 4 * 3), dtype=np.float32)[::2, ::3]
+    else:
+        raise ValueError(f"Unknown parametrization {type}")
+
+    info = m.get_py_buffer(mat, m.PyBUF_STRIDES)
+    assert info.itemsize == ctypes.sizeof(ctypes.c_float)
+    assert info.len == 5 * 4 * info.itemsize
+    assert info.ndim == 2
+    assert info.shape == [5, 4]
+    assert info.strides == [2 * 4 * 3 * info.itemsize, 3 * info.itemsize]
+    assert info.suboffsets is None
+    assert not info.readonly
+
+
+@pytest.mark.parametrize("type", ["pybind11", "numpy"])
+def test_to_pybuffer_contiguity(type):
+    def check_strides(mat):
+        # The full block is memset to 0, so fill it with non-zero in real spots.
+        expected = np.arange(1, 5 * 4 + 1).reshape((5, 4))
+        for i in range(5):
+            for j in range(4):
+                mat[i, j] = expected[i, j]
+        # If all strides are correct, the exposed buffer should match the input.
+        np.testing.assert_array_equal(np.array(mat), expected)
+
+    if type == "pybind11":
+        cmat = m.Matrix(5, 4)  # C contiguous.
+        fmat = m.FortranMatrix(5, 4)  # Fortran contiguous.
+        dmat = m.DiscontiguousMatrix(5, 4, 2, 3)  # Not contiguous.
+        expected_exception = BufferError
+    elif type == "numpy":
+        cmat = np.empty((5, 4), dtype=np.float32)  # C contiguous.
+        fmat = np.empty((5, 4), dtype=np.float32, order="F")  # Fortran contiguous.
+        dmat = np.empty((5 * 2, 4 * 3), dtype=np.float32)[::2, ::3]  # Not contiguous.
+        # NumPy incorrectly raises ValueError; when the minimum NumPy requirement is
+        # above the version that fixes https://github.com/numpy/numpy/issues/3634 then
+        # BufferError can be used everywhere.
+        expected_exception = (BufferError, ValueError)
+    else:
+        raise ValueError(f"Unknown parametrization {type}")
+
+    check_strides(cmat)
+    # Should work in C-contiguous mode, but not Fortran order.
+    m.get_py_buffer(cmat, m.PyBUF_C_CONTIGUOUS)
+    m.get_py_buffer(cmat, m.PyBUF_ANY_CONTIGUOUS)
+    with pytest.raises(expected_exception):
+        m.get_py_buffer(cmat, m.PyBUF_F_CONTIGUOUS)
+
+    check_strides(fmat)
+    # These flags imply C-contiguity, so won't work.
+    with pytest.raises(expected_exception):
+        m.get_py_buffer(fmat, m.PyBUF_SIMPLE)
+    with pytest.raises(expected_exception):
+        m.get_py_buffer(fmat, m.PyBUF_ND)
+    # Should work in Fortran-contiguous mode, but not C order.
+    with pytest.raises(expected_exception):
+        m.get_py_buffer(fmat, m.PyBUF_C_CONTIGUOUS)
+    m.get_py_buffer(fmat, m.PyBUF_ANY_CONTIGUOUS)
+    m.get_py_buffer(fmat, m.PyBUF_F_CONTIGUOUS)
+
+    check_strides(dmat)
+    # Should never work.
+    with pytest.raises(expected_exception):
+        m.get_py_buffer(dmat, m.PyBUF_SIMPLE)
+    with pytest.raises(expected_exception):
+        m.get_py_buffer(dmat, m.PyBUF_ND)
+    with pytest.raises(expected_exception):
+        m.get_py_buffer(dmat, m.PyBUF_C_CONTIGUOUS)
+    with pytest.raises(expected_exception):
+        m.get_py_buffer(dmat, m.PyBUF_ANY_CONTIGUOUS)
+    with pytest.raises(expected_exception):
+        m.get_py_buffer(dmat, m.PyBUF_F_CONTIGUOUS)

tests/test_call_policies.py

@@ -190,6 +190,7 @@ def test_alive_gc_multi_derived(capture):
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_return_none(capture):
     n_inst = ConstructorStats.detail_reg_inst()
     with capture:
@@ -217,6 +218,7 @@ def test_return_none(capture):
     assert capture == "Releasing parent."
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_keep_alive_constructor(capture):
     n_inst = ConstructorStats.detail_reg_inst()
 

tests/test_class.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import sys
 from unittest import mock
 
 import pytest
@@ -27,6 +28,9 @@ def test_instance(msg):
 
     instance = m.NoConstructor.new_instance()
 
+    if env.GRAALPY:
+        pytest.skip("ConstructorStats is incompatible with GraalPy.")
+
     cstats = ConstructorStats.get(m.NoConstructor)
     assert cstats.alive() == 1
     del instance
@@ -35,6 +39,10 @@ def test_instance(msg):
 
 def test_instance_new():
     instance = m.NoConstructorNew()  # .__new__(m.NoConstructor.__class__)
+
+    if env.GRAALPY:
+        pytest.skip("ConstructorStats is incompatible with GraalPy.")
+
     cstats = ConstructorStats.get(m.NoConstructorNew)
     assert cstats.alive() == 1
     del instance
@@ -501,3 +509,31 @@ def test_pr4220_tripped_over_this():
         m.Empty0().get_msg()
         == "This is really only meant to exercise successful compilation."
     )
+
+
+@pytest.mark.skipif(sys.platform.startswith("emscripten"), reason="Requires threads")
+def test_all_type_info_multithreaded():
+    # See PR #5419 for background.
+    import threading
+
+    from pybind11_tests import TestContext
+
+    class Context(TestContext):
+        pass
+
+    num_runs = 10
+    num_threads = 4
+    barrier = threading.Barrier(num_threads)
+
+    def func():
+        barrier.wait()
+        with Context():
+            pass
+
+    for _ in range(num_runs):
+        threads = [threading.Thread(target=func) for _ in range(num_threads)]
+        for thread in threads:
+            thread.start()
+
+        for thread in threads:
+            thread.join()

tests/test_copy_move.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import pytest
 
+import env  # noqa: F401
 from pybind11_tests import copy_move_policies as m
 
 
@@ -17,6 +18,7 @@ def test_lacking_move_ctor():
     assert "is neither movable nor copyable!" in str(excinfo.value)
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_move_and_copy_casts():
     """Cast some values in C++ via custom type casters and count the number of moves/copies."""
 
@@ -44,6 +46,7 @@ def test_move_and_copy_casts():
     assert c_m.alive() + c_mc.alive() + c_c.alive() == 0
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_move_and_copy_loads():
     """Call some functions that load arguments via custom type casters and count the number of
     moves/copies."""
@@ -77,6 +80,7 @@ def test_move_and_copy_loads():
 
 
 @pytest.mark.skipif(not m.has_optional, reason="no <optional>")
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_move_and_copy_load_optional():
     """Tests move/copy loads of std::optional arguments"""
 

tests/test_custom_type_casters.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import pytest
 
+import env  # noqa: F401
 from pybind11_tests import custom_type_casters as m
 
 
@@ -94,6 +95,7 @@ def test_noconvert_args(msg):
     )
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def test_custom_caster_destruction():
     """Tests that returning a pointer to a type that gets converted with a custom type caster gets
     destroyed when the function has py::return_value_policy::take_ownership policy applied.

tests/test_eigen_matrix.py

@@ -395,6 +395,7 @@ def test_eigen_return_references():
     np.testing.assert_array_equal(a_copy5, c5want)
 
 
+@pytest.mark.skipif("env.GRAALPY", reason="Cannot reliably trigger GC")
 def assert_keeps_alive(cl, method, *args):
     cstats = ConstructorStats.get(cl)
     start_with = cstats.alive()

tests/test_enum.py

@@ -1,6 +1,8 @@
 # ruff: noqa: SIM201 SIM300 SIM202
 from __future__ import annotations
 
+import re
+
 import pytest
 
 import env  # noqa: F401
@@ -271,3 +273,61 @@ def test_docstring_signatures():
 def test_str_signature():
     for enum_type in [m.ScopedEnum, m.UnscopedEnum]:
         assert enum_type.__str__.__doc__.startswith("__str__")
+
+
+def test_generated_dunder_methods_pos_only():
+    for enum_type in [m.ScopedEnum, m.UnscopedEnum]:
+        for binary_op in [
+            "__eq__",
+            "__ne__",
+            "__ge__",
+            "__gt__",
+            "__lt__",
+            "__le__",
+            "__and__",
+            "__rand__",
+            # "__or__",  # fail with some compilers (__doc__ = "Return self|value.")
+            # "__ror__",  # fail with some compilers (__doc__ = "Return value|self.")
+            "__xor__",
+            "__rxor__",
+            "__rxor__",
+        ]:
+            method = getattr(enum_type, binary_op, None)
+            if method is not None:
+                assert (
+                    re.match(
+                        rf"^{binary_op}\(self: [\w\.]+, other: [\w\.]+, /\)",
+                        method.__doc__,
+                    )
+                    is not None
+                )
+        for unary_op in [
+            "__int__",
+            "__index__",
+            "__hash__",
+            "__str__",
+            "__repr__",
+        ]:
+            method = getattr(enum_type, unary_op, None)
+            if method is not None:
+                assert (
+                    re.match(
+                        rf"^{unary_op}\(self: [\w\.]+, /\)",
+                        method.__doc__,
+                    )
+                    is not None
+                )
+        assert (
+            re.match(
+                r"^__getstate__\(self: [\w\.]+, /\)",
+                enum_type.__getstate__.__doc__,
+            )
+            is not None
+        )
+        assert (
+            re.match(
+                r"^__setstate__\(self: [\w\.]+, state: [\w\.]+, /\)",
+                enum_type.__setstate__.__doc__,
+            )
+            is not None
+        )

tests/test_iostream.py

@@ -6,14 +6,8 @@ from io import StringIO
 
 import pytest
 
-import env  # noqa: F401
 from pybind11_tests import iostream as m
 
-pytestmark = pytest.mark.skipif(
-    "env.GRAALPY",
-    reason="Delayed prints from finalizers from other tests can end up in the output",
-)
-
 
 def test_captured(capsys):
     msg = "I've been redirected to Python, I hope!"

tests/test_kwargs_and_defaults.cpp

@@ -14,26 +14,6 @@
 
 #include <utility>
 
-// Classes needed for subclass test.
-class ArgsSubclass : public py::args {
-    using py::args::args;
-};
-class KWArgsSubclass : public py::kwargs {
-    using py::kwargs::kwargs;
-};
-namespace pybind11 {
-namespace detail {
-template <>
-struct handle_type_name<ArgsSubclass> {
-    static constexpr auto name = const_name("*Args");
-};
-template <>
-struct handle_type_name<KWArgsSubclass> {
-    static constexpr auto name = const_name("**KWArgs");
-};
-} // namespace detail
-} // namespace pybind11
-
 TEST_SUBMODULE(kwargs_and_defaults, m) {
     auto kw_func
         = [](int x, int y) { return "x=" + std::to_string(x) + ", y=" + std::to_string(y); };
@@ -345,7 +325,7 @@ TEST_SUBMODULE(kwargs_and_defaults, m) {
 
     // Test support for args and kwargs subclasses
     m.def("args_kwargs_subclass_function",
-          [](const ArgsSubclass &args, const KWArgsSubclass &kwargs) {
+          [](const py::Args<std::string> &args, const py::KWArgs<std::string> &kwargs) {
               return py::make_tuple(args, kwargs);
           });
 }

tests/test_kwargs_and_defaults.py

@@ -20,7 +20,7 @@ def test_function_signatures(doc):
     )
     assert (
         doc(m.args_kwargs_subclass_function)
-        == "args_kwargs_subclass_function(*Args, **KWArgs) -> tuple"
+        == "args_kwargs_subclass_function(*args: str, **kwargs: str) -> tuple"
     )
     assert (
         doc(m.KWClass.foo0)

tests/test_methods_and_attributes.cpp

@@ -294,7 +294,7 @@ TEST_SUBMODULE(methods_and_attributes, m) {
                                  static_cast<py::str (ExampleMandA::*)(int, int)>(
                                      &ExampleMandA::overloaded));
                     })
-        .def("__str__", &ExampleMandA::toString)
+        .def("__str__", &ExampleMandA::toString, py::pos_only())
         .def_readwrite("value", &ExampleMandA::value);
 
     // test_copy_method

tests/test_methods_and_attributes.py

@@ -19,6 +19,13 @@ NO_DELETER_MSG = (
 )
 
 
+def test_self_only_pos_only():
+    assert (
+        m.ExampleMandA.__str__.__doc__
+        == "__str__(self: pybind11_tests.methods_and_attributes.ExampleMandA, /) -> str\n"
+    )
+
+
 def test_methods_and_attributes():
     instance1 = m.ExampleMandA()
     instance2 = m.ExampleMandA(32)

tests/test_numpy_array.cpp

@@ -131,6 +131,15 @@ arr_t &mutate_at_t(arr_t &a, Ix... idx) {
     a.mutable_at(idx...)++;
     return a;
 }
+template <typename... Ix>
+arr_t &subscript_via_call_operator_t(arr_t &a, Ix... idx) {
+    a(idx...)++;
+    return a;
+}
+template <typename... Ix>
+py::ssize_t const_subscript_via_call_operator_t(const arr_t &a, Ix... idx) {
+    return a(idx...);
+}
 
 #define def_index_fn(name, type)                                                                  \
     sm.def(#name, [](type a) { return name(a); });                                                \
@@ -156,6 +165,55 @@ py::handle auxiliaries(T &&r, T2 &&r2) {
     return l.release();
 }
 
+template <typename PyObjectType>
+PyObjectType convert_to_pyobjecttype(py::object obj);
+
+template <>
+PyObject *convert_to_pyobjecttype<PyObject *>(py::object obj) {
+    return obj.release().ptr();
+}
+
+template <>
+py::handle convert_to_pyobjecttype<py::handle>(py::object obj) {
+    return obj.release();
+}
+
+template <>
+py::object convert_to_pyobjecttype<py::object>(py::object obj) {
+    return obj;
+}
+
+template <typename PyObjectType>
+std::string pass_array_return_sum_str_values(const py::array_t<PyObjectType> &objs) {
+    std::string sum_str_values;
+    for (const auto &obj : objs) {
+        sum_str_values += py::str(obj.attr("value"));
+    }
+    return sum_str_values;
+}
+
+template <typename PyObjectType>
+py::list pass_array_return_as_list(const py::array_t<PyObjectType> &objs) {
+    return objs;
+}
+
+template <typename PyObjectType>
+py::array_t<PyObjectType> return_array_cpp_loop(const py::list &objs) {
+    py::size_t arr_size = py::len(objs);
+    py::array_t<PyObjectType> arr_from_list(static_cast<py::ssize_t>(arr_size));
+    PyObjectType *data = arr_from_list.mutable_data();
+    for (py::size_t i = 0; i < arr_size; i++) {
+        assert(!data[i]);
+        data[i] = convert_to_pyobjecttype<PyObjectType>(objs[i].attr("value"));
+    }
+    return arr_from_list;
+}
+
+template <typename PyObjectType>
+py::array_t<PyObjectType> return_array_from_list(const py::list &objs) {
+    return objs;
+}
+
 // note: declaration at local scope would create a dangling reference!
 static int data_i = 42;
 
@@ -197,6 +255,13 @@ TEST_SUBMODULE(numpy_array, sm) {
     sm.def("nbytes", [](const arr &a) { return a.nbytes(); });
     sm.def("owndata", [](const arr &a) { return a.owndata(); });
 
+    sm.attr("defined_NDEBUG") =
+#ifdef NDEBUG
+        true;
+#else
+        false;
+#endif
+
     // test_index_offset
     def_index_fn(index_at, const arr &);
     def_index_fn(index_at_t, const arr_t &);
@@ -210,6 +275,8 @@ TEST_SUBMODULE(numpy_array, sm) {
     def_index_fn(mutate_data_t, arr_t &);
     def_index_fn(at_t, const arr_t &);
     def_index_fn(mutate_at_t, arr_t &);
+    def_index_fn(subscript_via_call_operator_t, arr_t &);
+    def_index_fn(const_subscript_via_call_operator_t, const arr_t &);
 
     // test_make_c_f_array
     sm.def("make_f_array", [] { return py::array_t<float>({2, 2}, {4, 8}); });
@@ -520,28 +587,21 @@ TEST_SUBMODULE(numpy_array, sm) {
     sm.def("round_trip_float", [](double d) { return d; });
 
     sm.def("pass_array_pyobject_ptr_return_sum_str_values",
-           [](const py::array_t<PyObject *> &objs) {
-               std::string sum_str_values;
-               for (const auto &obj : objs) {
-                   sum_str_values += py::str(obj.attr("value"));
-               }
-               return sum_str_values;
-           });
+           pass_array_return_sum_str_values<PyObject *>);
+    sm.def("pass_array_handle_return_sum_str_values",
+           pass_array_return_sum_str_values<py::handle>);
+    sm.def("pass_array_object_return_sum_str_values",
+           pass_array_return_sum_str_values<py::object>);
 
-    sm.def("pass_array_pyobject_ptr_return_as_list",
-           [](const py::array_t<PyObject *> &objs) -> py::list { return objs; });
+    sm.def("pass_array_pyobject_ptr_return_as_list", pass_array_return_as_list<PyObject *>);
+    sm.def("pass_array_handle_return_as_list", pass_array_return_as_list<py::handle>);
+    sm.def("pass_array_object_return_as_list", pass_array_return_as_list<py::object>);
 
-    sm.def("return_array_pyobject_ptr_cpp_loop", [](const py::list &objs) {
-        py::size_t arr_size = py::len(objs);
-        py::array_t<PyObject *> arr_from_list(static_cast<py::ssize_t>(arr_size));
-        PyObject **data = arr_from_list.mutable_data();
-        for (py::size_t i = 0; i < arr_size; i++) {
-            assert(data[i] == nullptr);
-            data[i] = py::cast<PyObject *>(objs[i].attr("value"));
-        }
-        return arr_from_list;
-    });
+    sm.def("return_array_pyobject_ptr_cpp_loop", return_array_cpp_loop<PyObject *>);
+    sm.def("return_array_handle_cpp_loop", return_array_cpp_loop<py::handle>);
+    sm.def("return_array_object_cpp_loop", return_array_cpp_loop<py::object>);
 
-    sm.def("return_array_pyobject_ptr_from_list",
-           [](const py::list &objs) -> py::array_t<PyObject *> { return objs; });
+    sm.def("return_array_pyobject_ptr_from_list", return_array_from_list<PyObject *>);
+    sm.def("return_array_handle_from_list", return_array_from_list<py::handle>);
+    sm.def("return_array_object_from_list", return_array_from_list<py::object>);
 }

tests/test_numpy_array.py

@@ -111,20 +111,32 @@ def test_data(arr, args, ret):
     assert all(m.data(arr, *args)[(1 if byteorder == "little" else 0) :: 2] == 0)
 
 
+@pytest.mark.parametrize(
+    "func",
+    [
+        m.at_t,
+        m.mutate_at_t,
+        m.const_subscript_via_call_operator_t,
+        m.subscript_via_call_operator_t,
+    ][: 2 if m.defined_NDEBUG else 99],
+)
 @pytest.mark.parametrize("dim", [0, 1, 3])
-def test_at_fail(arr, dim):
-    for func in m.at_t, m.mutate_at_t:
+def test_elem_reference(arr, func, dim):
     with pytest.raises(IndexError) as excinfo:
         func(arr, *([0] * dim))
     assert str(excinfo.value) == f"index dimension mismatch: {dim} (ndim = 2)"
 
 
-def test_at(arr):
-    assert m.at_t(arr, 0, 2) == 3
-    assert m.at_t(arr, 1, 0) == 4
+@pytest.mark.parametrize("func", [m.at_t, m.const_subscript_via_call_operator_t])
+def test_const_elem_reference(arr, func):
+    assert func(arr, 0, 2) == 3
+    assert func(arr, 1, 0) == 4
 
-    assert all(m.mutate_at_t(arr, 0, 2).ravel() == [1, 2, 4, 4, 5, 6])
-    assert all(m.mutate_at_t(arr, 1, 0).ravel() == [1, 2, 4, 5, 5, 6])
+
+@pytest.mark.parametrize("func", [m.mutate_at_t, m.subscript_via_call_operator_t])
+def test_mutable_elem_reference(arr, func):
+    assert all(func(arr, 0, 2).ravel() == [1, 2, 4, 4, 5, 6])
+    assert all(func(arr, 1, 0).ravel() == [1, 2, 4, 5, 5, 6])
 
 
 def test_mutate_readonly(arr):
@@ -153,8 +165,9 @@ def test_mutate_data(arr):
     assert all(m.mutate_data_t(arr, 1, 2).ravel() == [6, 19, 27, 68, 84, 197])
 
 
-def test_bounds_check(arr):
-    for func in (
+@pytest.mark.parametrize(
+    "func",
+    [
         m.index_at,
         m.index_at_t,
         m.data,
@@ -163,7 +176,11 @@ def test_bounds_check(arr):
         m.mutate_data_t,
         m.at_t,
         m.mutate_at_t,
-    ):
+        m.const_subscript_via_call_operator_t,
+        m.subscript_via_call_operator_t,
+    ][: 8 if m.defined_NDEBUG else 99],
+)
+def test_bounds_check(arr, func):
     with pytest.raises(IndexError) as excinfo:
         func(arr, 2, 0)
     assert str(excinfo.value) == "index 2 is out of bounds for axis 0 with size 2"
@@ -629,45 +646,61 @@ def UnwrapPyValueHolder(vhs):
     return [vh.value for vh in vhs]
 
 
-def test_pass_array_pyobject_ptr_return_sum_str_values_ndarray():
+PASS_ARRAY_PYOBJECT_RETURN_SUM_STR_VALUES_FUNCTIONS = [
+    m.pass_array_pyobject_ptr_return_sum_str_values,
+    m.pass_array_handle_return_sum_str_values,
+    m.pass_array_object_return_sum_str_values,
+]
+
+
+@pytest.mark.parametrize(
+    "pass_array", PASS_ARRAY_PYOBJECT_RETURN_SUM_STR_VALUES_FUNCTIONS
+)
+def test_pass_array_object_return_sum_str_values_ndarray(pass_array):
     # Intentionally all temporaries, do not change.
     assert (
-        m.pass_array_pyobject_ptr_return_sum_str_values(
-            np.array(WrapWithPyValueHolder(-3, "four", 5.0), dtype=object)
-        )
+        pass_array(np.array(WrapWithPyValueHolder(-3, "four", 5.0), dtype=object))
         == "-3four5.0"
     )
 
 
-def test_pass_array_pyobject_ptr_return_sum_str_values_list():
+@pytest.mark.parametrize(
+    "pass_array", PASS_ARRAY_PYOBJECT_RETURN_SUM_STR_VALUES_FUNCTIONS
+)
+def test_pass_array_object_return_sum_str_values_list(pass_array):
     # Intentionally all temporaries, do not change.
-    assert (
-        m.pass_array_pyobject_ptr_return_sum_str_values(
-            WrapWithPyValueHolder(2, "three", -4.0)
-        )
-        == "2three-4.0"
-    )
+    assert pass_array(WrapWithPyValueHolder(2, "three", -4.0)) == "2three-4.0"
 
 
-def test_pass_array_pyobject_ptr_return_as_list():
+@pytest.mark.parametrize(
+    "pass_array",
+    [
+        m.pass_array_pyobject_ptr_return_as_list,
+        m.pass_array_handle_return_as_list,
+        m.pass_array_object_return_as_list,
+    ],
+)
+def test_pass_array_object_return_as_list(pass_array):
     # Intentionally all temporaries, do not change.
     assert UnwrapPyValueHolder(
-        m.pass_array_pyobject_ptr_return_as_list(
-            np.array(WrapWithPyValueHolder(-1, "two", 3.0), dtype=object)
-        )
+        pass_array(np.array(WrapWithPyValueHolder(-1, "two", 3.0), dtype=object))
     ) == [-1, "two", 3.0]
 
 
 @pytest.mark.parametrize(
-    ("return_array_pyobject_ptr", "unwrap"),
+    ("return_array", "unwrap"),
     [
         (m.return_array_pyobject_ptr_cpp_loop, list),
+        (m.return_array_handle_cpp_loop, list),
+        (m.return_array_object_cpp_loop, list),
         (m.return_array_pyobject_ptr_from_list, UnwrapPyValueHolder),
+        (m.return_array_handle_from_list, UnwrapPyValueHolder),
+        (m.return_array_object_from_list, UnwrapPyValueHolder),
     ],
 )
-def test_return_array_pyobject_ptr_cpp_loop(return_array_pyobject_ptr, unwrap):
+def test_return_array_object_cpp_loop(return_array, unwrap):
     # Intentionally all temporaries, do not change.
-    arr_from_list = return_array_pyobject_ptr(WrapWithPyValueHolder(6, "seven", -8.0))
+    arr_from_list = return_array(WrapWithPyValueHolder(6, "seven", -8.0))
     assert isinstance(arr_from_list, np.ndarray)
     assert arr_from_list.dtype == np.dtype("O")
     assert unwrap(arr_from_list) == [6, "seven", -8.0]

tests/test_numpy_dtypes.cpp

@@ -11,6 +11,9 @@
 
 #include "pybind11_tests.h"
 
+#include <cstdint>
+#include <stdexcept>
+
 #ifdef __GNUC__
 #    define PYBIND11_PACKED(cls) cls __attribute__((__packed__))
 #else
@@ -297,6 +300,15 @@ py::list test_dtype_ctors() {
     return list;
 }
 
+template <typename T>
+py::array_t<T> dispatch_array_increment(py::array_t<T> arr) {
+    py::array_t<T> res(arr.shape(0));
+    for (py::ssize_t i = 0; i < arr.shape(0); ++i) {
+        res.mutable_at(i) = T(arr.at(i) + 1);
+    }
+    return res;
+}
+
 struct A {};
 struct B {};
 
@@ -496,6 +508,98 @@ TEST_SUBMODULE(numpy_dtypes, m) {
         }
         return list;
     });
+    m.def("test_dtype_num_of", []() -> py::list {
+        py::list res;
+#define TEST_DTYPE(T) res.append(py::make_tuple(py::dtype::of<T>().num(), py::dtype::num_of<T>()));
+        TEST_DTYPE(bool)
+        TEST_DTYPE(char)
+        TEST_DTYPE(unsigned char)
+        TEST_DTYPE(short)
+        TEST_DTYPE(unsigned short)
+        TEST_DTYPE(int)
+        TEST_DTYPE(unsigned int)
+        TEST_DTYPE(long)
+        TEST_DTYPE(unsigned long)
+        TEST_DTYPE(long long)
+        TEST_DTYPE(unsigned long long)
+        TEST_DTYPE(float)
+        TEST_DTYPE(double)
+        TEST_DTYPE(long double)
+        TEST_DTYPE(std::complex<float>)
+        TEST_DTYPE(std::complex<double>)
+        TEST_DTYPE(std::complex<long double>)
+        TEST_DTYPE(int8_t)
+        TEST_DTYPE(uint8_t)
+        TEST_DTYPE(int16_t)
+        TEST_DTYPE(uint16_t)
+        TEST_DTYPE(int32_t)
+        TEST_DTYPE(uint32_t)
+        TEST_DTYPE(int64_t)
+        TEST_DTYPE(uint64_t)
+#undef TEST_DTYPE
+        return res;
+    });
+    m.def("test_dtype_normalized_num", []() -> py::list {
+        py::list res;
+#define TEST_DTYPE(NT, T)                                                                         \
+    res.append(py::make_tuple(py::dtype(py::detail::npy_api::NT).normalized_num(),                \
+                              py::dtype::num_of<T>()));
+        TEST_DTYPE(NPY_BOOL_, bool)
+        TEST_DTYPE(NPY_BYTE_, char);
+        TEST_DTYPE(NPY_UBYTE_, unsigned char);
+        TEST_DTYPE(NPY_SHORT_, short);
+        TEST_DTYPE(NPY_USHORT_, unsigned short);
+        TEST_DTYPE(NPY_INT_, int);
+        TEST_DTYPE(NPY_UINT_, unsigned int);
+        TEST_DTYPE(NPY_LONG_, long);
+        TEST_DTYPE(NPY_ULONG_, unsigned long);
+        TEST_DTYPE(NPY_LONGLONG_, long long);
+        TEST_DTYPE(NPY_ULONGLONG_, unsigned long long);
+        TEST_DTYPE(NPY_FLOAT_, float);
+        TEST_DTYPE(NPY_DOUBLE_, double);
+        TEST_DTYPE(NPY_LONGDOUBLE_, long double);
+        TEST_DTYPE(NPY_CFLOAT_, std::complex<float>);
+        TEST_DTYPE(NPY_CDOUBLE_, std::complex<double>);
+        TEST_DTYPE(NPY_CLONGDOUBLE_, std::complex<long double>);
+        TEST_DTYPE(NPY_INT8_, int8_t);
+        TEST_DTYPE(NPY_UINT8_, uint8_t);
+        TEST_DTYPE(NPY_INT16_, int16_t);
+        TEST_DTYPE(NPY_UINT16_, uint16_t);
+        TEST_DTYPE(NPY_INT32_, int32_t);
+        TEST_DTYPE(NPY_UINT32_, uint32_t);
+        TEST_DTYPE(NPY_INT64_, int64_t);
+        TEST_DTYPE(NPY_UINT64_, uint64_t);
+#undef TEST_DTYPE
+        return res;
+    });
+    m.def("test_dtype_switch", [](const py::array &arr) -> py::array {
+        switch (arr.dtype().normalized_num()) {
+            case py::dtype::num_of<int8_t>():
+                return dispatch_array_increment<int8_t>(arr);
+            case py::dtype::num_of<uint8_t>():
+                return dispatch_array_increment<uint8_t>(arr);
+            case py::dtype::num_of<int16_t>():
+                return dispatch_array_increment<int16_t>(arr);
+            case py::dtype::num_of<uint16_t>():
+                return dispatch_array_increment<uint16_t>(arr);
+            case py::dtype::num_of<int32_t>():
+                return dispatch_array_increment<int32_t>(arr);
+            case py::dtype::num_of<uint32_t>():
+                return dispatch_array_increment<uint32_t>(arr);
+            case py::dtype::num_of<int64_t>():
+                return dispatch_array_increment<int64_t>(arr);
+            case py::dtype::num_of<uint64_t>():
+                return dispatch_array_increment<uint64_t>(arr);
+            case py::dtype::num_of<float>():
+                return dispatch_array_increment<float>(arr);
+            case py::dtype::num_of<double>():
+                return dispatch_array_increment<double>(arr);
+            case py::dtype::num_of<long double>():
+                return dispatch_array_increment<long double>(arr);
+            default:
+                throw std::runtime_error("Unsupported dtype");
+        }
+    });
     m.def("test_dtype_methods", []() {
         py::list list;
         auto dt1 = py::dtype::of<int32_t>();

tests/test_numpy_dtypes.py

@@ -188,6 +188,28 @@ def test_dtype(simple_dtype):
             chr(np.dtype(ch).flags) for ch in expected_chars
         ]
 
+    for a, b in m.test_dtype_num_of():
+        assert a == b
+
+    for a, b in m.test_dtype_normalized_num():
+        assert a == b
+
+    arr = np.array([4, 84, 21, 36])
+    # Note: "ulong" does not work in NumPy 1.x, so we use "L"
+    assert (m.test_dtype_switch(arr.astype("byte")) == arr + 1).all()
+    assert (m.test_dtype_switch(arr.astype("ubyte")) == arr + 1).all()
+    assert (m.test_dtype_switch(arr.astype("short")) == arr + 1).all()
+    assert (m.test_dtype_switch(arr.astype("ushort")) == arr + 1).all()
+    assert (m.test_dtype_switch(arr.astype("intc")) == arr + 1).all()
+    assert (m.test_dtype_switch(arr.astype("uintc")) == arr + 1).all()
+    assert (m.test_dtype_switch(arr.astype("long")) == arr + 1).all()
+    assert (m.test_dtype_switch(arr.astype("L")) == arr + 1).all()
+    assert (m.test_dtype_switch(arr.astype("longlong")) == arr + 1).all()
+    assert (m.test_dtype_switch(arr.astype("ulonglong")) == arr + 1).all()
+    assert (m.test_dtype_switch(arr.astype("single")) == arr + 1).all()
+    assert (m.test_dtype_switch(arr.astype("double")) == arr + 1).all()
+    assert (m.test_dtype_switch(arr.astype("longdouble")) == arr + 1).all()
+
 
 def test_recarray(simple_dtype, packed_dtype):
     elements = [(False, 0, 0.0, -0.0), (True, 1, 1.5, -2.5), (False, 2, 3.0, -5.0)]

tests/test_opaque_types.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import pytest
 
+import env
 from pybind11_tests import ConstructorStats, UserType
 from pybind11_tests import opaque_types as m
 
@@ -30,6 +31,8 @@ def test_pointers(msg):
     living_before = ConstructorStats.get(UserType).alive()
     assert m.get_void_ptr_value(m.return_void_ptr()) == 0x1234
     assert m.get_void_ptr_value(UserType())  # Should also work for other C++ types
+
+    if not env.GRAALPY:
         assert ConstructorStats.get(UserType).alive() == living_before
 
     with pytest.raises(TypeError) as excinfo:

tests/test_operator_overloading.py

@@ -2,7 +2,7 @@ from __future__ import annotations
 
 import pytest
 
-import env  # noqa: F401
+import env
 from pybind11_tests import ConstructorStats
 from pybind11_tests import operators as m
 
@@ -51,6 +51,9 @@ def test_operator_overloading():
     v2 /= v1
     assert str(v2) == "[2.000000, 8.000000]"
 
+    if env.GRAALPY:
+        pytest.skip("ConstructorStats is incompatible with GraalPy.")
+
     cstats = ConstructorStats.get(m.Vector2)
     assert cstats.alive() == 3
     del v1

tests/test_pickling.py

@@ -93,3 +93,20 @@ def test_roundtrip_simple_cpp_derived():
     # Issue #3062: pickleable base C++ classes can incur object slicing
     #              if derived typeid is not registered with pybind11
     assert not m.check_dynamic_cast_SimpleCppDerived(p2)
+
+
+def test_new_style_pickle_getstate_pos_only():
+    assert (
+        re.match(
+            r"^__getstate__\(self: [\w\.]+, /\)", m.PickleableNew.__getstate__.__doc__
+        )
+        is not None
+    )
+    if hasattr(m, "PickleableWithDictNew"):
+        assert (
+            re.match(
+                r"^__getstate__\(self: [\w\.]+, /\)",
+                m.PickleableWithDictNew.__getstate__.__doc__,
+            )
+            is not None
+        )

tests/test_sequences_and_iterators.py

@@ -1,5 +1,7 @@
 from __future__ import annotations
 
+import re
+
 import pytest
 from pytest import approx  # noqa: PT013
 
@@ -253,16 +255,12 @@ def test_python_iterator_in_cpp():
 
 def test_iterator_passthrough():
     """#181: iterator passthrough did not compile"""
-    from pybind11_tests.sequences_and_iterators import iterator_passthrough
-
     values = [3, 5, 7, 9, 11, 13, 15]
-    assert list(iterator_passthrough(iter(values))) == values
+    assert list(m.iterator_passthrough(iter(values))) == values
 
 
 def test_iterator_rvp():
     """#388: Can't make iterators via make_iterator() with different r/v policies"""
-    import pybind11_tests.sequences_and_iterators as m
-
     assert list(m.make_iterator_1()) == [1, 2, 3]
     assert list(m.make_iterator_2()) == [1, 2, 3]
     assert not isinstance(m.make_iterator_1(), type(m.make_iterator_2()))
@@ -274,3 +272,25 @@ def test_carray_iterator():
     arr_h = m.CArrayHolder(*args_gt)
     args = list(arr_h)
     assert args_gt == args
+
+
+def test_generated_dunder_methods_pos_only():
+    string_map = m.StringMap({"hi": "bye", "black": "white"})
+    for it in (
+        m.make_iterator_1(),
+        m.make_iterator_2(),
+        m.iterator_passthrough(iter([3, 5, 7])),
+        iter(m.Sequence(5)),
+        iter(string_map),
+        string_map.items(),
+        string_map.values(),
+        iter(m.CArrayHolder(*[float(i) for i in range(3)])),
+    ):
+        assert (
+            re.match(r"^__iter__\(self: [\w\.]+, /\)", type(it).__iter__.__doc__)
+            is not None
+        )
+        assert (
+            re.match(r"^__next__\(self: [\w\.]+, /\)", type(it).__next__.__doc__)
+            is not None
+        )

tools/setup_global.py.in

@@ -26,12 +26,14 @@ class InstallHeadersNested(install_headers):
 
 
 main_headers = glob.glob("pybind11/include/pybind11/*.h")
+conduit_headers = sum([glob.glob(f"pybind11/include/pybind11/conduit/*.{ext}")
+                       for ext in ("h", "txt")], [])
 detail_headers = glob.glob("pybind11/include/pybind11/detail/*.h")
 eigen_headers = glob.glob("pybind11/include/pybind11/eigen/*.h")
 stl_headers = glob.glob("pybind11/include/pybind11/stl/*.h")
 cmake_files = glob.glob("pybind11/share/cmake/pybind11/*.cmake")
 pkgconfig_files = glob.glob("pybind11/share/pkgconfig/*.pc")
-headers = main_headers + detail_headers + stl_headers + eigen_headers
+headers = main_headers + conduit_headers + detail_headers + eigen_headers + stl_headers
 
 cmdclass = {"install_headers": InstallHeadersNested}
 $extra_cmd
@@ -55,6 +57,7 @@ setup(
         (base + "share/cmake/pybind11", cmake_files),
         (base + "share/pkgconfig", pkgconfig_files),
         (base + "include/pybind11", main_headers),
+        (base + "include/pybind11/conduit", conduit_headers),
         (base + "include/pybind11/detail", detail_headers),
         (base + "include/pybind11/eigen", eigen_headers),
         (base + "include/pybind11/stl", stl_headers),

tools/setup_main.py.in

@@ -14,6 +14,7 @@ setup(
     packages=[
         "pybind11",
         "pybind11.include.pybind11",
+        "pybind11.include.pybind11.conduit",
         "pybind11.include.pybind11.detail",
         "pybind11.include.pybind11.eigen",
         "pybind11.include.pybind11.stl",
@@ -23,6 +24,7 @@ setup(
     package_data={
         "pybind11": ["py.typed"],
         "pybind11.include.pybind11": ["*.h"],
+        "pybind11.include.pybind11.conduit": ["*.h", "*.txt"],
         "pybind11.include.pybind11.detail": ["*.h"],
         "pybind11.include.pybind11.eigen": ["*.h"],
         "pybind11.include.pybind11.stl": ["*.h"],