significant redesign of GIL state handling

Author: wjakob

include/pybind11/cast.h

@@ -55,6 +55,13 @@ PYBIND11_NOINLINE inline internals &get_internals() {
         internals_ptr = caps;
     } else {
         internals_ptr = new internals();
+        #if defined(WITH_THREAD)
+            PyEval_InitThreads();
+            PyThreadState *tstate = PyThreadState_Get();
+            internals_ptr->tstate = PyThread_create_key();
+            PyThread_set_key_value(internals_ptr->tstate, tstate);
+            internals_ptr->istate = tstate->interp;
+        #endif
         builtins[id] = capsule(internals_ptr);
     }
     return *internals_ptr;

include/pybind11/common.h

@@ -46,6 +46,7 @@
 
 #include <Python.h>
 #include <frameobject.h>
+#include <pythread.h>
 
 #ifdef isalnum
 #  undef isalnum
@@ -127,6 +128,9 @@
     } \
     PyObject *pybind11_init()
 
+extern "C" {
+    extern PyThreadState *_PyThreadState_Current;
+};
 
 NAMESPACE_BEGIN(pybind11)
 
@@ -233,6 +237,10 @@ struct internals {
     std::unordered_map<const void *, void*> registered_types_py;     // PyTypeObject* -> type_info
     std::unordered_map<const void *, void*> registered_instances;    // void * -> PyObject*
     std::unordered_set<std::pair<const PyObject *, const char *>, overload_hash> inactive_overload_cache;
+#if defined(WITH_THREAD)
+    int tstate = 0;
+    PyInterpreterState *istate = nullptr;
+#endif
 };
 
 /// Return a reference to the current 'internals' information

include/pybind11/pybind11.h

@@ -1039,21 +1039,106 @@ template <typename InputType, typename OutputType> void implicitly_convertible()
 }
 
 #if defined(WITH_THREAD)
-inline void init_threading() { PyEval_InitThreads(); }
+
+/* The functions below essentially reproduce the PyGILState_* API using a RAII
+ * pattern, but there are a few important differences:
+ *
+ * 1. When acquiring the GIL from an non-main thread during the finalization
+ *    phase, the GILState API blindly terminates the calling thread, which
+ *    is often not what is wanted. This API does not do this.
+ *
+ * 2. The gil_scoped_release function can optionally cut the relationship
+ *    of a PyThreadState and its associated thread, which allows moving it to
+ *    another thread (this is a fairly rare/advanced use case).
+ *
+ * 3. The reference count of an acquired thread state can be controlled. This
+ *    can be handy to prevent cases where callbacks issued from an external
+ *    thread constantly construct and destroy thread state data structures. */
 
 class gil_scoped_acquire {
-    PyGILState_STATE state;
 public:
-    inline gil_scoped_acquire() { state = PyGILState_Ensure(); }
-    inline ~gil_scoped_acquire() { PyGILState_Release(state); }
+    gil_scoped_acquire() {
+        auto const &internals = detail::get_internals();
+        tstate = (PyThreadState *) PyThread_get_key_value(internals.tstate);
+
+        if (!tstate) {
+            tstate = PyThreadState_New(internals.istate);
+            #if !defined(NDEBUG)
+                if (!tstate)
+                    pybind11_fail("scoped_acquire: could not create thread state!");
+            #endif
+            tstate->gilstate_counter = 0;
+            PyThread_set_key_value(internals.tstate, tstate);
+        } else {
+            release = _PyThreadState_Current != tstate;
+        }
+
+        if (release) {
+            PyInterpreterState *interp = tstate->interp;
+            /* Work around an annoying assertion in PyThreadState_Swap */
+            tstate->interp = nullptr;
+            PyEval_AcquireThread(tstate);
+            tstate->interp = interp;
+        }
+
+        inc_ref();
+    }
+
+    void inc_ref() {
+        ++tstate->gilstate_counter;
+    }
+
+    void dec_ref() {
+        --tstate->gilstate_counter;
+        #if !defined(NDEBUG)
+            if (_PyThreadState_Current != tstate)
+                pybind11_fail("scoped_acquire::dec_ref(): thread state must be current!");
+            if (tstate->gilstate_counter < 0)
+                pybind11_fail("scoped_acquire::dec_ref(): reference count underflow!");
+        #endif
+        if (tstate->gilstate_counter == 0) {
+            #if !defined(NDEBUG)
+                if (!release)
+                    pybind11_fail("scoped_acquire::dec_ref(): internal error!");
+            #endif
+            PyThreadState_Clear(tstate);
+            PyThreadState_DeleteCurrent();
+            PyThread_set_key_value(detail::get_internals().tstate, nullptr);
+            release = false;
+        }
+    }
+
+    ~gil_scoped_acquire() {
+        dec_ref();
+        if (release)
+           PyEval_SaveThread();
+    }
+private:
+    PyThreadState *tstate = nullptr;
+    bool release = true;
 };
 
 class gil_scoped_release {
-    PyThreadState *state;
 public:
-    inline gil_scoped_release() { state = PyEval_SaveThread(); }
-    inline ~gil_scoped_release() { PyEval_RestoreThread(state); }
+    gil_scoped_release(bool disassoc = false) : disassoc(disassoc) {
+        tstate = PyEval_SaveThread();
+        if (disassoc)
+            PyThread_set_key_value(detail::get_internals().tstate, nullptr);
+    }
+    ~gil_scoped_release() {
+        if (!tstate)
+            return;
+        PyEval_RestoreThread(tstate);
+        if (disassoc)
+            PyThread_set_key_value(detail::get_internals().tstate, tstate);
+    }
+private:
+    PyThreadState *tstate;
+    bool disassoc;
 };
+#else
+class gil_scoped_acquire { };
+class gil_scoped_release { };
 #endif
 
 inline function get_overload(const void *this_ptr, const char *name)  {