Support free-threaded CPython (PEP 703)

Some additional locking is added in the free-threaded build when
`Py_GIL_DISABLED` is defined:

- Most accesses to internals are protected by a single mutex
- The registered_instances uses a striped lock to improve concurrency

Pybind11 modules can indicate support for running with the GIL disabled
by calling `set_gil_not_used()`.

Author: colesbury

include/pybind11/detail/class.h

@@ -205,39 +205,40 @@ extern "C" inline PyObject *pybind11_meta_call(PyObject *type, PyObject *args, P
 
 /// Cleanup the type-info for a pybind11-registered type.
 extern "C" inline void pybind11_meta_dealloc(PyObject *obj) {
-    auto *type = (PyTypeObject *) obj;
-    auto &internals = get_internals();
-
-    // A pybind11-registered type will:
-    // 1) be found in internals.registered_types_py
-    // 2) have exactly one associated `detail::type_info`
-    auto found_type = internals.registered_types_py.find(type);
-    if (found_type != internals.registered_types_py.end() && found_type->second.size() == 1
-        && found_type->second[0]->type == type) {
-
-        auto *tinfo = found_type->second[0];
-        auto tindex = std::type_index(*tinfo->cpptype);
-        internals.direct_conversions.erase(tindex);
-
-        if (tinfo->module_local) {
-            get_local_internals().registered_types_cpp.erase(tindex);
-        } else {
-            internals.registered_types_cpp.erase(tindex);
-        }
-        internals.registered_types_py.erase(tinfo->type);
-
-        // Actually just `std::erase_if`, but that's only available in C++20
-        auto &cache = internals.inactive_override_cache;
-        for (auto it = cache.begin(), last = cache.end(); it != last;) {
-            if (it->first == (PyObject *) tinfo->type) {
-                it = cache.erase(it);
+    with_internals([obj](internals &internals) {
+        auto *type = (PyTypeObject *) obj;
+
+        // A pybind11-registered type will:
+        // 1) be found in internals.registered_types_py
+        // 2) have exactly one associated `detail::type_info`
+        auto found_type = internals.registered_types_py.find(type);
+        if (found_type != internals.registered_types_py.end() && found_type->second.size() == 1
+            && found_type->second[0]->type == type) {
+
+            auto *tinfo = found_type->second[0];
+            auto tindex = std::type_index(*tinfo->cpptype);
+            internals.direct_conversions.erase(tindex);
+
+            if (tinfo->module_local) {
+                get_local_internals().registered_types_cpp.erase(tindex);
             } else {
-                ++it;
+                internals.registered_types_cpp.erase(tindex);
+            }
+            internals.registered_types_py.erase(tinfo->type);
+
+            // Actually just `std::erase_if`, but that's only available in C++20
+            auto &cache = internals.inactive_override_cache;
+            for (auto it = cache.begin(), last = cache.end(); it != last;) {
+                if (it->first == (PyObject *) tinfo->type) {
+                    it = cache.erase(it);
+                } else {
+                    ++it;
+                }
             }
-        }
 
-        delete tinfo;
-    }
+            delete tinfo;
+        }
+    });
 
     PyType_Type.tp_dealloc(obj);
 }
@@ -310,19 +311,20 @@ inline void traverse_offset_bases(void *valueptr,
 }
 
 inline bool register_instance_impl(void *ptr, instance *self) {
-    get_internals().registered_instances.emplace(ptr, self);
+    with_instance_map(ptr, [&](instance_map &instances) { instances.emplace(ptr, self); });
     return true; // unused, but gives the same signature as the deregister func
 }
 inline bool deregister_instance_impl(void *ptr, instance *self) {
-    auto &registered_instances = get_internals().registered_instances;
-    auto range = registered_instances.equal_range(ptr);
-    for (auto it = range.first; it != range.second; ++it) {
-        if (self == it->second) {
-            registered_instances.erase(it);
-            return true;
+    return with_instance_map(ptr, [&](instance_map &instances) {
+        auto range = instances.equal_range(ptr);
+        for (auto it = range.first; it != range.second; ++it) {
+            if (self == it->second) {
+                instances.erase(it);
+                return true;
+            }
         }
-    }
-    return false;
+        return false;
+    });
 }
 
 inline void register_instance(instance *self, void *valptr, const type_info *tinfo) {
@@ -377,27 +379,32 @@ extern "C" inline int pybind11_object_init(PyObject *self, PyObject *, PyObject
 }
 
 inline void add_patient(PyObject *nurse, PyObject *patient) {
-    auto &internals = get_internals();
     auto *instance = reinterpret_cast<detail::instance *>(nurse);
     instance->has_patients = true;
     Py_INCREF(patient);
-    internals.patients[nurse].push_back(patient);
+
+    with_internals([&](internals &internals) { internals.patients[nurse].push_back(patient); });
 }
 
 inline void clear_patients(PyObject *self) {
     auto *instance = reinterpret_cast<detail::instance *>(self);
-    auto &internals = get_internals();
-    auto pos = internals.patients.find(self);
+    std::vector<PyObject *> patients;
 
-    if (pos == internals.patients.end()) {
-        pybind11_fail("FATAL: Internal consistency check failed: Invalid clear_patients() call.");
-    }
+    with_internals([&](internals &internals) {
+        auto pos = internals.patients.find(self);
+
+        if (pos == internals.patients.end()) {
+            pybind11_fail(
+                "FATAL: Internal consistency check failed: Invalid clear_patients() call.");
+        }
+
+        // Clearing the patients can cause more Python code to run, which
+        // can invalidate the iterator. Extract the vector of patients
+        // from the unordered_map first.
+        patients = std::move(pos->second);
+        internals.patients.erase(pos);
+    });
 
-    // Clearing the patients can cause more Python code to run, which
-    // can invalidate the iterator. Extract the vector of patients
-    // from the unordered_map first.
-    auto patients = std::move(pos->second);
-    internals.patients.erase(pos);
     instance->has_patients = false;
     for (PyObject *&patient : patients) {
         Py_CLEAR(patient);
@@ -662,10 +669,13 @@ inline PyObject *make_new_python_type(const type_record &rec) {
 
     char *tp_doc = nullptr;
     if (rec.doc && options::show_user_defined_docstrings()) {
-        /* Allocate memory for docstring (using PyObject_MALLOC, since
-           Python will free this later on) */
+        /* Allocate memory for docstring (Python will free this later on) */
         size_t size = std::strlen(rec.doc) + 1;
+#if PY_VERSION_HEX >= 0x030D0000
+        tp_doc = (char *) PyMem_MALLOC(size);
+#else
         tp_doc = (char *) PyObject_MALLOC(size);
+#endif
         std::memcpy((void *) tp_doc, rec.doc, size);
     }
 

include/pybind11/detail/internals.h

@@ -18,6 +18,8 @@
 #include "../pytypes.h"
 
 #include <exception>
+#include <mutex>
+#include <thread>
 
 /// Tracks the `internals` and `type_info` ABI version independent of the main library version.
 ///
@@ -34,7 +36,10 @@
 /// further ABI-incompatible changes may be made before the ABI is officially
 /// changed to the new version.
 #ifndef PYBIND11_INTERNALS_VERSION
-#    if PY_VERSION_HEX >= 0x030C0000 || defined(_MSC_VER)
+#    if PY_VERSION_HEX >= 0x030D0000
+// Version bump for Python 3.13+.
+#        define PYBIND11_INTERNALS_VERSION 6
+#    elif PY_VERSION_HEX >= 0x030C0000 || defined(_MSC_VER)
 // Version bump for Python 3.12+, before first 3.12 beta release.
 // Version bump for MSVC piggy-backed on PR #4779. See comments there.
 #        define PYBIND11_INTERNALS_VERSION 5
@@ -168,15 +173,31 @@ struct override_hash {
     }
 };
 
+using instance_map = std::unordered_multimap<const void *, instance *>;
+
+struct instance_map_shard {
+    std::mutex mutex;
+    instance_map registered_instances;
+    char padding[64 - (sizeof(std::mutex) + sizeof(instance_map)) % 64];
+};
+
 /// Internal data structure used to track registered instances and types.
 /// Whenever binary incompatible changes are made to this structure,
 /// `PYBIND11_INTERNALS_VERSION` must be incremented.
 struct internals {
+#if PYBIND11_INTERNALS_VERSION >= 6
+    std::mutex mutex;
+#endif
     // std::type_index -> pybind11's type information
     type_map<type_info *> registered_types_cpp;
     // PyTypeObject* -> base type_info(s)
     std::unordered_map<PyTypeObject *, std::vector<type_info *>> registered_types_py;
-    std::unordered_multimap<const void *, instance *> registered_instances; // void * -> instance*
+#if PYBIND11_INTERNALS_VERSION >= 6
+    std::unique_ptr<instance_map_shard[]> instance_shards; // void * -> instance*
+    size_t instance_shards_mask;
+#else
+    instance_map registered_instances; // void * -> instance*
+#endif
     std::unordered_set<std::pair<const PyObject *, const char *>, override_hash>
         inactive_override_cache;
     type_map<std::vector<bool (*)(PyObject *, void *&)>> direct_conversions;
@@ -462,7 +483,8 @@ inline object get_python_state_dict() {
 }
 
 inline object get_internals_obj_from_state_dict(handle state_dict) {
-    return reinterpret_borrow<object>(dict_getitemstring(state_dict.ptr(), PYBIND11_INTERNALS_ID));
+    return reinterpret_steal<object>(
+        dict_getitemstringref(state_dict.ptr(), PYBIND11_INTERNALS_ID));
 }
 
 inline internals **get_internals_pp_from_capsule(handle obj) {
@@ -474,6 +496,20 @@ inline internals **get_internals_pp_from_capsule(handle obj) {
     return static_cast<internals **>(raw_ptr);
 }
 
+inline uint64_t next_pow2(uint64_t x) {
+    // Round-up to the next power of two.
+    // See https://graphics.stanford.edu/~seander/bithacks.html#RoundUpPowerOf2
+    x--;
+    x |= (x >> 1);
+    x |= (x >> 2);
+    x |= (x >> 4);
+    x |= (x >> 8);
+    x |= (x >> 16);
+    x |= (x >> 32);
+    x++;
+    return x;
+}
+
 /// Return a reference to the current `internals` data
 PYBIND11_NOINLINE internals &get_internals() {
     auto **&internals_pp = get_internals_pp();
@@ -542,6 +578,18 @@ PYBIND11_NOINLINE internals &get_internals() {
         internals_ptr->static_property_type = make_static_property_type();
         internals_ptr->default_metaclass = make_default_metaclass();
         internals_ptr->instance_base = make_object_base_type(internals_ptr->default_metaclass);
+#if PYBIND11_INTERNALS_VERSION >= 6
+#    if defined(Py_GIL_DISABLED)
+        size_t num_shards = (size_t) next_pow2(2 * std::thread::hardware_concurrency());
+        if (num_shards == 0) {
+            num_shards = 1;
+        }
+#    else
+        size_t num_shards = 1;
+#    endif
+        internals_ptr->instance_shards.reset(new instance_map_shard[num_shards]);
+        internals_ptr->instance_shards_mask = num_shards - 1;
+#endif // PYBIND11_INTERNALS_VERSION >= 6
     }
     return **internals_pp;
 }
@@ -602,13 +650,77 @@ inline local_internals &get_local_internals() {
     return *locals;
 }
 
+#if PYBIND11_INTERNALS_VERSION >= 6 && defined(Py_GIL_DISABLED)
+#    define PYBIND11_LOCK_INTERNALS(internals) std::unique_lock<std::mutex> lock((internals).mutex)
+#else
+#    define PYBIND11_LOCK_INTERNALS(internals)
+#endif
+
+template <typename F>
+inline auto with_internals(const F &cb) -> decltype(cb(get_internals())) {
+    auto &internals = get_internals();
+    PYBIND11_LOCK_INTERNALS(internals);
+    return cb(internals);
+}
+
+inline uint64_t splitmix64(uint64_t z) {
+    z = (z ^ (z >> 30)) * 0xbf58476d1ce4e5b9;
+    z = (z ^ (z >> 27)) * 0x94d049bb133111eb;
+    return z ^ (z >> 31);
+}
+
+template <typename F>
+inline auto with_instance_map(const void *ptr,
+                              const F &cb) -> decltype(cb(std::declval<instance_map &>())) {
+    auto &internals = get_internals();
+
+#if PYBIND11_INTERNALS_VERSION >= 6
+    // Hash address to compute shard, but ignore low bits. We'd like allocations
+    // from the same thread/core to map to the same shard and allocations from
+    // other threads/cores to map to other shards. Using the high bits is a good
+    // heuristic because memory allocators often have a per-thread
+    // arena/superblock/segment from which smaller allocations are served.
+    auto addr = reinterpret_cast<uintptr_t>(ptr);
+    uint64_t hash = splitmix64((uint64_t) (addr >> 20));
+    size_t idx = (size_t) hash & internals.instance_shards_mask;
+
+    auto &shard = internals.instance_shards[idx];
+#    if defined(Py_GIL_DISABLED)
+    std::unique_lock<std::mutex> lock(shard.mutex);
+#    endif
+    return cb(shard.registered_instances);
+#else
+    (void) ptr;
+    return cb(internals.registered_instances);
+#endif
+}
+
+inline size_t num_registered_instances() {
+    auto &internals = get_internals();
+#if PYBIND11_INTERNALS_VERSION >= 6
+    size_t count = 0;
+    for (size_t i = 0; i <= internals.instance_shards_mask; ++i) {
+        auto &shard = internals.instance_shards[i];
+        std::unique_lock<std::mutex> lock(shard.mutex);
+        count += shard.registered_instances.size();
+    }
+    return count;
+#else
+    return internals.registered_instances.size();
+#endif
+}
+
 /// Constructs a std::string with the given arguments, stores it in `internals`, and returns its
 /// `c_str()`.  Such strings objects have a long storage duration -- the internal strings are only
 /// cleared when the program exits or after interpreter shutdown (when embedding), and so are
 /// suitable for c-style strings needed by Python internals (such as PyTypeObject's tp_name).
 template <typename... Args>
 const char *c_str(Args &&...args) {
-    auto &strings = get_internals().static_strings;
+    // GCC 4.8 doesn't like parameter unpack within lambda capture, so use
+    // PYBIND11_LOCK_INTERNALS.
+    auto &internals = get_internals();
+    PYBIND11_LOCK_INTERNALS(internals);
+    auto &strings = internals.static_strings;
     strings.emplace_front(std::forward<Args>(args)...);
     return strings.front().c_str();
 }
@@ -638,30 +750,34 @@ PYBIND11_NAMESPACE_END(detail)
 /// pybind11 version) running in the current interpreter. Names starting with underscores
 /// are reserved for internal usage. Returns `nullptr` if no matching entry was found.
 PYBIND11_NOINLINE void *get_shared_data(const std::string &name) {
-    auto &internals = detail::get_internals();
-    auto it = internals.shared_data.find(name);
-    return it != internals.shared_data.end() ? it->second : nullptr;
+    return detail::with_internals([&](detail::internals &internals) {
+        auto it = internals.shared_data.find(name);
+        return it != internals.shared_data.end() ? it->second : nullptr;
+    });
 }
 
 /// Set the shared data that can be later recovered by `get_shared_data()`.
 PYBIND11_NOINLINE void *set_shared_data(const std::string &name, void *data) {
-    detail::get_internals().shared_data[name] = data;
-    return data;
+    return detail::with_internals([&](detail::internals &internals) {
+        internals.shared_data[name] = data;
+        return data;
+    });
 }
 
 /// Returns a typed reference to a shared data entry (by using `get_shared_data()`) if
 /// such entry exists. Otherwise, a new object of default-constructible type `T` is
 /// added to the shared data under the given name and a reference to it is returned.
 template <typename T>
 T &get_or_create_shared_data(const std::string &name) {
-    auto &internals = detail::get_internals();
-    auto it = internals.shared_data.find(name);
-    T *ptr = (T *) (it != internals.shared_data.end() ? it->second : nullptr);
-    if (!ptr) {
-        ptr = new T();
-        internals.shared_data[name] = ptr;
-    }
-    return *ptr;
+    return *detail::with_internals([&](detail::internals &internals) {
+        auto it = internals.shared_data.find(name);
+        T *ptr = (T *) (it != internals.shared_data.end() ? it->second : nullptr);
+        if (!ptr) {
+            ptr = new T();
+            internals.shared_data[name] = ptr;
+        }
+        return ptr;
+    });
 }
 
 PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)