bpo-38644: Add _PyObject_Call() (GH-17089)

* Add pycore_call.h internal header file.
* Add _PyObject_Call(): PyObject_Call() with tstate
* Add _PyObject_CallNoArgTstate(): _PyObject_CallNoArg() with tstate
* Add _PyObject_FastCallDictTstate(): _PyObject_FastCallDict()
  with tstate
* _PyObject_Call_Prepend() now takes tstate
* Replace _PyObject_FastCall() calls
  with _PyObject_VectorcallTstate() calls

10 files changed, 216 insertions, 137 deletions

diff --git a/Include/cpython/abstract.h b/Include/cpython/abstract.h
index fef538e..2c4eae7 100644
--- a/Include/cpython/abstract.h
+++ b/Include/cpython/abstract.h
@@ -137,7 +137,8 @@ PyAPI_FUNC(PyObject *) PyVectorcall_Call(PyObject *callable, PyObject *tuple, Py
 static inline PyObject *
 _PyObject_FastCall(PyObject *func, PyObject *const *args, Py_ssize_t nargs)
 {
-    return _PyObject_Vectorcall(func, args, (size_t)nargs, NULL);
+    PyThreadState *tstate = PyThreadState_GET();
+    return _PyObject_VectorcallTstate(tstate, func, args, (size_t)nargs, NULL);
 }
 
 /* Call a callable without any arguments
@@ -145,7 +146,8 @@ _PyObject_FastCall(PyObject *func, PyObject *const *args, Py_ssize_t nargs)
    PyObject_CallNoArgs(). */
 static inline PyObject *
 _PyObject_CallNoArg(PyObject *func) {
-    return _PyObject_Vectorcall(func, NULL, 0, NULL);
+    PyThreadState *tstate = PyThreadState_GET();
+    return _PyObject_VectorcallTstate(tstate, func, NULL, 0, NULL);
 }
 
 static inline PyObject *
@@ -155,16 +157,11 @@ _PyObject_CallOneArg(PyObject *func, PyObject *arg)
     PyObject *_args[2];
     PyObject **args = _args + 1;  // For PY_VECTORCALL_ARGUMENTS_OFFSET
     args[0] = arg;
-    return _PyObject_Vectorcall(func, args,
-                                1 | PY_VECTORCALL_ARGUMENTS_OFFSET, NULL);
+    PyThreadState *tstate = PyThreadState_GET();
+    size_t nargsf = 1 | PY_VECTORCALL_ARGUMENTS_OFFSET;
+    return _PyObject_VectorcallTstate(tstate, func, args, nargsf, NULL);
 }
 
-PyAPI_FUNC(PyObject *) _PyObject_Call_Prepend(
-    PyObject *callable,
-    PyObject *obj,
-    PyObject *args,
-    PyObject *kwargs);
-
 PyAPI_FUNC(PyObject *) _PyObject_VectorcallMethod(
     PyObject *name, PyObject *const *args,
     size_t nargsf, PyObject *kwnames);
diff --git a/Include/internal/pycore_call.h b/Include/internal/pycore_call.h
new file mode 100644
index 0000000..f7d856a
--- /dev/null
+++ b/Include/internal/pycore_call.h
@@ -0,0 +1,39 @@
+#ifndef Py_INTERNAL_CALL_H
+#define Py_INTERNAL_CALL_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef Py_BUILD_CORE
+#  error "this header requires Py_BUILD_CORE define"
+#endif
+
+PyAPI_FUNC(PyObject *) _PyObject_Call_Prepend(
+    PyThreadState *tstate,
+    PyObject *callable,
+    PyObject *obj,
+    PyObject *args,
+    PyObject *kwargs);
+
+PyAPI_FUNC(PyObject *) _PyObject_FastCallDictTstate(
+    PyThreadState *tstate,
+    PyObject *callable,
+    PyObject *const *args,
+    size_t nargsf,
+    PyObject *kwargs);
+
+PyAPI_FUNC(PyObject *) _PyObject_Call(
+    PyThreadState *tstate,
+    PyObject *callable,
+    PyObject *args,
+    PyObject *kwargs);
+
+static inline PyObject *
+_PyObject_CallNoArgTstate(PyThreadState *tstate, PyObject *func) {
+    return _PyObject_VectorcallTstate(tstate, func, NULL, 0, NULL);
+}
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* !Py_INTERNAL_CALL_H */
diff --git a/Makefile.pre.in b/Makefile.pre.in
index 3c607d0..d08c78d 100644
--- a/Makefile.pre.in
+++ b/Makefile.pre.in
@@ -1076,6 +1076,7 @@ PYTHON_HEADERS= \
 		\
 		$(srcdir)/Include/internal/pycore_accu.h \
 		$(srcdir)/Include/internal/pycore_atomic.h \
+		$(srcdir)/Include/internal/pycore_call.h \
 		$(srcdir)/Include/internal/pycore_ceval.h \
 		$(srcdir)/Include/internal/pycore_code.h \
 		$(srcdir)/Include/internal/pycore_condvar.h \
diff --git a/Objects/call.c b/Objects/call.c
index 96174c4..ae08d18 100644
--- a/Objects/call.c
+++ b/Objects/call.c
@@ -1,4 +1,5 @@
 #include "Python.h"
+#include "pycore_call.h"
 #include "pycore_ceval.h"   /* _PyEval_EvalFrame() */
 #include "pycore_object.h"
 #include "pycore_pyerrors.h"
@@ -18,11 +19,12 @@ _PyStack_UnpackDict_Free(PyObject *const *stack, Py_ssize_t nargs,
 
 
 static PyObject *
-null_error(void)
+null_error(PyThreadState *tstate)
 {
-    if (!PyErr_Occurred())
-        PyErr_SetString(PyExc_SystemError,
-                        "null argument to internal routine");
+    if (!_PyErr_Occurred(tstate)) {
+        _PyErr_SetString(tstate, PyExc_SystemError,
+                         "null argument to internal routine");
+    }
     return NULL;
 }
 
@@ -81,17 +83,18 @@ _Py_CheckFunctionResult(PyThreadState *tstate, PyObject *callable,
 PyObject *
 PyObject_CallNoArgs(PyObject *func)
 {
-    return _PyObject_CallNoArg(func);
+    PyThreadState *tstate = _PyThreadState_GET();
+    return _PyObject_CallNoArgTstate(tstate, func);
 }
 
 
 PyObject *
-_PyObject_FastCallDict(PyObject *callable, PyObject *const *args,
-                       size_t nargsf, PyObject *kwargs)
+_PyObject_FastCallDictTstate(PyThreadState *tstate, PyObject *callable,
+                             PyObject *const *args, size_t nargsf,
+                             PyObject *kwargs)
 {
     assert(callable != NULL);
 
-    PyThreadState *tstate = _PyThreadState_GET();
     /* _PyObject_FastCallDict() must not be called with an exception set,
        because it can clear it (directly or indirectly) and so the
        caller loses its exception */
@@ -130,10 +133,23 @@ _PyObject_FastCallDict(PyObject *callable, PyObject *const *args,
 
 
 PyObject *
+_PyObject_FastCallDict(PyObject *callable, PyObject *const *args,
+                       size_t nargsf, PyObject *kwargs)
+{
+    PyThreadState *tstate = _PyThreadState_GET();
+    return _PyObject_FastCallDictTstate(tstate, callable, args, nargsf, kwargs);
+}
+
+
+PyObject *
 _PyObject_MakeTpCall(PyThreadState *tstate, PyObject *callable,
                      PyObject *const *args, Py_ssize_t nargs,
                      PyObject *keywords)
 {
+    assert(nargs >= 0);
+    assert(nargs == 0 || args != NULL);
+    assert(keywords == NULL || PyTuple_Check(keywords) || PyDict_Check(keywords));
+
     /* Slow path: build a temporary tuple for positional arguments and a
      * temporary dictionary for keyword arguments (if any) */
     ternaryfunc call = Py_TYPE(callable)->tp_call;
@@ -144,9 +160,6 @@ _PyObject_MakeTpCall(PyThreadState *tstate, PyObject *callable,
         return NULL;
     }
 
-    assert(nargs >= 0);
-    assert(nargs == 0 || args != NULL);
-    assert(keywords == NULL || PyTuple_Check(keywords) || PyDict_Check(keywords));
     PyObject *argstuple = _PyTuple_FromArray(args, nargs);
     if (argstuple == NULL) {
         return NULL;
@@ -182,8 +195,7 @@ _PyObject_MakeTpCall(PyThreadState *tstate, PyObject *callable,
         Py_DECREF(kwdict);
     }
 
-    result = _Py_CheckFunctionResult(tstate, callable, result, NULL);
-    return result;
+    return _Py_CheckFunctionResult(tstate, callable, result, NULL);
 }
 
 
@@ -228,14 +240,15 @@ PyVectorcall_Call(PyObject *callable, PyObject *tuple, PyObject *kwargs)
     PyObject *result = func(callable, args,
                             nargs | PY_VECTORCALL_ARGUMENTS_OFFSET, kwnames);
     _PyStack_UnpackDict_Free(args, nargs, kwnames);
+
     return _Py_CheckFunctionResult(tstate, callable, result, NULL);
 }
 
 
 PyObject *
-PyObject_Call(PyObject *callable, PyObject *args, PyObject *kwargs)
+_PyObject_Call(PyThreadState *tstate, PyObject *callable,
+               PyObject *args, PyObject *kwargs)
 {
-    PyThreadState *tstate = _PyThreadState_GET();
     ternaryfunc call;
     PyObject *result;
 
@@ -270,11 +283,19 @@ PyObject_Call(PyObject *callable, PyObject *args, PyObject *kwargs)
     }
 }
 
+PyObject *
+PyObject_Call(PyObject *callable, PyObject *args, PyObject *kwargs)
+{
+    PyThreadState *tstate = _PyThreadState_GET();
+    return _PyObject_Call(tstate, callable, args, kwargs);
+}
+
 
 PyObject *
 PyCFunction_Call(PyObject *callable, PyObject *args, PyObject *kwargs)
 {
-    return PyObject_Call(callable, args, kwargs);
+    PyThreadState *tstate = _PyThreadState_GET();
+    return _PyObject_Call(tstate, callable, args, kwargs);
 }
 
 
@@ -385,30 +406,31 @@ PyObject *
 PyEval_CallObjectWithKeywords(PyObject *callable,
                               PyObject *args, PyObject *kwargs)
 {
+    PyThreadState *tstate = _PyThreadState_GET();
 #ifdef Py_DEBUG
     /* PyEval_CallObjectWithKeywords() must not be called with an exception
        set. It raises a new exception if parameters are invalid or if
        PyTuple_New() fails, and so the original exception is lost. */
-    assert(!PyErr_Occurred());
+    assert(!_PyErr_Occurred(tstate));
 #endif
 
     if (args != NULL && !PyTuple_Check(args)) {
-        PyErr_SetString(PyExc_TypeError,
-                        "argument list must be a tuple");
+        _PyErr_SetString(tstate, PyExc_TypeError,
+                         "argument list must be a tuple");
         return NULL;
     }
 
     if (kwargs != NULL && !PyDict_Check(kwargs)) {
-        PyErr_SetString(PyExc_TypeError,
-                        "keyword list must be a dictionary");
+        _PyErr_SetString(tstate, PyExc_TypeError,
+                         "keyword list must be a dictionary");
         return NULL;
     }
 
     if (args == NULL) {
-        return _PyObject_FastCallDict(callable, NULL, 0, kwargs);
+        return _PyObject_FastCallDictTstate(tstate, callable, NULL, 0, kwargs);
     }
     else {
-        return PyObject_Call(callable, args, kwargs);
+        return _PyObject_Call(tstate, callable, args, kwargs);
     }
 }
 
@@ -416,32 +438,31 @@ PyEval_CallObjectWithKeywords(PyObject *callable,
 PyObject *
 PyObject_CallObject(PyObject *callable, PyObject *args)
 {
-    assert(!PyErr_Occurred());
+    PyThreadState *tstate = _PyThreadState_GET();
+    assert(!_PyErr_Occurred(tstate));
     if (args == NULL) {
-        return _PyObject_CallNoArg(callable);
+        return _PyObject_CallNoArgTstate(tstate, callable);
     }
     if (!PyTuple_Check(args)) {
-        PyErr_SetString(PyExc_TypeError,
-                        "argument list must be a tuple");
+        _PyErr_SetString(tstate, PyExc_TypeError,
+                         "argument list must be a tuple");
         return NULL;
     }
-    return PyObject_Call(callable, args, NULL);
+    return _PyObject_Call(tstate, callable, args, NULL);
 }
 
 
 /* Call callable(obj, *args, **kwargs). */
 PyObject *
-_PyObject_Call_Prepend(PyObject *callable,
+_PyObject_Call_Prepend(PyThreadState *tstate, PyObject *callable,
                        PyObject *obj, PyObject *args, PyObject *kwargs)
 {
+    assert(PyTuple_Check(args));
+
     PyObject *small_stack[_PY_FASTCALL_SMALL_STACK];
     PyObject **stack;
-    Py_ssize_t argcount;
-    PyObject *result;
 
-    assert(PyTuple_Check(args));
-
-    argcount = PyTuple_GET_SIZE(args);
+    Py_ssize_t argcount = PyTuple_GET_SIZE(args);
     if (argcount + 1 <= (Py_ssize_t)Py_ARRAY_LENGTH(small_stack)) {
         stack = small_stack;
     }
@@ -459,9 +480,9 @@ _PyObject_Call_Prepend(PyObject *callable,
            _PyTuple_ITEMS(args),
            argcount * sizeof(PyObject *));
 
-    result = _PyObject_FastCallDict(callable,
-                                    stack, argcount + 1,
-                                    kwargs);
+    PyObject *result = _PyObject_FastCallDictTstate(tstate, callable,
+                                                    stack, argcount + 1,
+                                                    kwargs);
     if (stack != small_stack) {
         PyMem_Free(stack);
     }
@@ -472,8 +493,8 @@ _PyObject_Call_Prepend(PyObject *callable,
 /* --- Call with a format string ---------------------------------- */
 
 static PyObject *
-_PyObject_CallFunctionVa(PyObject *callable, const char *format,
-                         va_list va, int is_size_t)
+_PyObject_CallFunctionVa(PyThreadState *tstate, PyObject *callable,
+                         const char *format, va_list va, int is_size_t)
 {
     PyObject* small_stack[_PY_FASTCALL_SMALL_STACK];
     const Py_ssize_t small_stack_len = Py_ARRAY_LENGTH(small_stack);
@@ -482,11 +503,11 @@ _PyObject_CallFunctionVa(PyObject *callable, const char *format,
     PyObject *result;
 
     if (callable == NULL) {
-        return null_error();
+        return null_error(tstate);
     }
 
     if (!format || !*format) {
-        return _PyObject_CallNoArg(callable);
+        return _PyObject_CallNoArgTstate(tstate, callable);
     }
 
     if (is_size_t) {
@@ -507,12 +528,14 @@ _PyObject_CallFunctionVa(PyObject *callable, const char *format,
            - PyObject_CallFunction(func, "(OOO)", arg1, arg2, arg3) calls
              func(*(arg1, arg2, arg3)): func(arg1, arg2, arg3) */
         PyObject *args = stack[0];
-        result = _PyObject_FastCall(callable,
-                                    _PyTuple_ITEMS(args),
-                                    PyTuple_GET_SIZE(args));
+        result = _PyObject_VectorcallTstate(tstate, callable,
+                                            _PyTuple_ITEMS(args),
+                                            PyTuple_GET_SIZE(args),
+                                            NULL);
     }
     else {
-        result = _PyObject_FastCall(callable, stack, nargs);
+        result = _PyObject_VectorcallTstate(tstate, callable,
+                                            stack, nargs, NULL);
     }
 
     for (i = 0; i < nargs; ++i) {
@@ -530,9 +553,10 @@ PyObject_CallFunction(PyObject *callable, const char *format, ...)
 {
     va_list va;
     PyObject *result;
+    PyThreadState *tstate = _PyThreadState_GET();
 
     va_start(va, format);
-    result = _PyObject_CallFunctionVa(callable, format, va, 0);
+    result = _PyObject_CallFunctionVa(tstate, callable, format, va, 0);
     va_end(va);
 
     return result;
@@ -547,9 +571,10 @@ PyEval_CallFunction(PyObject *callable, const char *format, ...)
 {
     va_list va;
     PyObject *result;
+    PyThreadState *tstate = _PyThreadState_GET();
 
     va_start(va, format);
-    result = _PyObject_CallFunctionVa(callable, format, va, 0);
+    result = _PyObject_CallFunctionVa(tstate, callable, format, va, 0);
     va_end(va);
 
     return result;
@@ -559,11 +584,11 @@ PyEval_CallFunction(PyObject *callable, const char *format, ...)
 PyObject *
 _PyObject_CallFunction_SizeT(PyObject *callable, const char *format, ...)
 {
-    va_list va;
-    PyObject *result;
+    PyThreadState *tstate = _PyThreadState_GET();
 
+    va_list va;
     va_start(va, format);
-    result = _PyObject_CallFunctionVa(callable, format, va, 1);
+    PyObject *result = _PyObject_CallFunctionVa(tstate, callable, format, va, 1);
     va_end(va);
 
     return result;
@@ -571,37 +596,37 @@ _PyObject_CallFunction_SizeT(PyObject *callable, const char *format, ...)
 
 
 static PyObject*
-callmethod(PyObject* callable, const char *format, va_list va, int is_size_t)
+callmethod(PyThreadState *tstate, PyObject* callable, const char *format, va_list va, int is_size_t)
 {
     assert(callable != NULL);
-
     if (!PyCallable_Check(callable)) {
-        PyErr_Format(PyExc_TypeError,
-                     "attribute of type '%.200s' is not callable",
-                     Py_TYPE(callable)->tp_name);
+        _PyErr_Format(tstate, PyExc_TypeError,
+                      "attribute of type '%.200s' is not callable",
+                      Py_TYPE(callable)->tp_name);
         return NULL;
     }
 
-    return _PyObject_CallFunctionVa(callable, format, va, is_size_t);
+    return _PyObject_CallFunctionVa(tstate, callable, format, va, is_size_t);
 }
 
 
 PyObject *
 PyObject_CallMethod(PyObject *obj, const char *name, const char *format, ...)
 {
-    va_list va;
-    PyObject *callable, *retval;
+    PyThreadState *tstate = _PyThreadState_GET();
 
     if (obj == NULL || name == NULL) {
-        return null_error();
+        return null_error(tstate);
     }
 
-    callable = PyObject_GetAttrString(obj, name);
-    if (callable == NULL)
+    PyObject *callable = PyObject_GetAttrString(obj, name);
+    if (callable == NULL) {
         return NULL;
+    }
 
+    va_list va;
     va_start(va, format);
-    retval = callmethod(callable, format, va, 0);
+    PyObject *retval = callmethod(tstate, callable, format, va, 0);
     va_end(va);
 
     Py_DECREF(callable);
@@ -615,19 +640,19 @@ PyObject_CallMethod(PyObject *obj, const char *name, const char *format, ...)
 PyObject *
 PyEval_CallMethod(PyObject *obj, const char *name, const char *format, ...)
 {
-    va_list va;
-    PyObject *callable, *retval;
-
+    PyThreadState *tstate = _PyThreadState_GET();
     if (obj == NULL || name == NULL) {
-        return null_error();
+        return null_error(tstate);
     }
 
-    callable = PyObject_GetAttrString(obj, name);
-    if (callable == NULL)
+    PyObject *callable = PyObject_GetAttrString(obj, name);
+    if (callable == NULL) {
         return NULL;
+    }
 
+    va_list va;
     va_start(va, format);
-    retval = callmethod(callable, format, va, 0);
+    PyObject *retval = callmethod(tstate, callable, format, va, 0);
     va_end(va);
 
     Py_DECREF(callable);
@@ -639,19 +664,19 @@ PyObject *
 _PyObject_CallMethodId(PyObject *obj, _Py_Identifier *name,
                        const char *format, ...)
 {
-    va_list va;
-    PyObject *callable, *retval;
-
+    PyThreadState *tstate = _PyThreadState_GET();
     if (obj == NULL || name == NULL) {
-        return null_error();
+        return null_error(tstate);
     }
 
-    callable = _PyObject_GetAttrId(obj, name);
-    if (callable == NULL)
+    PyObject *callable = _PyObject_GetAttrId(obj, name);
+    if (callable == NULL) {
         return NULL;
+    }
 
+    va_list va;
     va_start(va, format);
-    retval = callmethod(callable, format, va, 0);
+    PyObject *retval = callmethod(tstate, callable, format, va, 0);
     va_end(va);
 
     Py_DECREF(callable);
@@ -663,19 +688,19 @@ PyObject *
 _PyObject_CallMethod_SizeT(PyObject *obj, const char *name,
                            const char *format, ...)
 {
-    va_list va;
-    PyObject *callable, *retval;
-
+    PyThreadState *tstate = _PyThreadState_GET();
     if (obj == NULL || name == NULL) {
-        return null_error();
+        return null_error(tstate);
     }
 
-    callable = PyObject_GetAttrString(obj, name);
-    if (callable == NULL)
+    PyObject *callable = PyObject_GetAttrString(obj, name);
+    if (callable == NULL) {
         return NULL;
+    }
 
+    va_list va;
     va_start(va, format);
-    retval = callmethod(callable, format, va, 1);
+    PyObject *retval = callmethod(tstate, callable, format, va, 1);
     va_end(va);
 
     Py_DECREF(callable);
@@ -687,20 +712,19 @@ PyObject *
 _PyObject_CallMethodId_SizeT(PyObject *obj, _Py_Identifier *name,
                              const char *format, ...)
 {
-    va_list va;
-    PyObject *callable, *retval;
-
+    PyThreadState *tstate = _PyThreadState_GET();
     if (obj == NULL || name == NULL) {
-        return null_error();
+        return null_error(tstate);
     }
 
-    callable = _PyObject_GetAttrId(obj, name);
+    PyObject *callable = _PyObject_GetAttrId(obj, name);
     if (callable == NULL) {
         return NULL;
     }
 
+    va_list va;
     va_start(va, format);
-    retval = callmethod(callable, format, va, 1);
+    PyObject *retval = callmethod(tstate, callable, format, va, 1);
     va_end(va);
 
     Py_DECREF(callable);
@@ -711,7 +735,8 @@ _PyObject_CallMethodId_SizeT(PyObject *obj, _Py_Identifier *name,
 /* --- Call with "..." arguments ---------------------------------- */
 
 static PyObject *
-object_vacall(PyObject *base, PyObject *callable, va_list vargs)
+object_vacall(PyThreadState *tstate, PyObject *base,
+              PyObject *callable, va_list vargs)
 {
     PyObject *small_stack[_PY_FASTCALL_SMALL_STACK];
     PyObject **stack;
@@ -721,7 +746,7 @@ object_vacall(PyObject *base, PyObject *callable, va_list vargs)
     va_list countva;
 
     if (callable == NULL) {
-        return null_error();
+        return null_error(tstate);
     }
 
     /* Count the number of arguments */
@@ -758,7 +783,7 @@ object_vacall(PyObject *base, PyObject *callable, va_list vargs)
     }
 
     /* Call the function */
-    result = _PyObject_FastCall(callable, stack, nargs);
+    result = _PyObject_VectorcallTstate(tstate, callable, stack, nargs, NULL);
 
     if (stack != small_stack) {
         PyMem_Free(stack);
@@ -804,8 +829,9 @@ _PyObject_VectorcallMethod(PyObject *name, PyObject *const *args,
 PyObject *
 PyObject_CallMethodObjArgs(PyObject *obj, PyObject *name, ...)
 {
+    PyThreadState *tstate = _PyThreadState_GET();
     if (obj == NULL || name == NULL) {
-        return null_error();
+        return null_error(tstate);
     }
 
     PyObject *callable = NULL;
@@ -817,7 +843,7 @@ PyObject_CallMethodObjArgs(PyObject *obj, PyObject *name, ...)
 
     va_list vargs;
     va_start(vargs, name);
-    PyObject *result = object_vacall(obj, callable, vargs);
+    PyObject *result = object_vacall(tstate, obj, callable, vargs);
     va_end(vargs);
 
     Py_DECREF(callable);
@@ -829,8 +855,9 @@ PyObject *
 _PyObject_CallMethodIdObjArgs(PyObject *obj,
                               struct _Py_Identifier *name, ...)
 {
+    PyThreadState *tstate = _PyThreadState_GET();
     if (obj == NULL || name == NULL) {
-        return null_error();
+        return null_error(tstate);
     }
 
     PyObject *oname = _PyUnicode_FromId(name); /* borrowed */
@@ -847,7 +874,7 @@ _PyObject_CallMethodIdObjArgs(PyObject *obj,
 
     va_list vargs;
     va_start(vargs, name);
-    PyObject *result = object_vacall(obj, callable, vargs);
+    PyObject *result = object_vacall(tstate, obj, callable, vargs);
     va_end(vargs);
 
     Py_DECREF(callable);
@@ -858,11 +885,12 @@ _PyObject_CallMethodIdObjArgs(PyObject *obj,
 PyObject *
 PyObject_CallFunctionObjArgs(PyObject *callable, ...)
 {
+    PyThreadState *tstate = _PyThreadState_GET();
     va_list vargs;
     PyObject *result;
 
     va_start(vargs, callable);
-    result = object_vacall(NULL, callable, vargs);
+    result = object_vacall(tstate, NULL, callable, vargs);
     va_end(vargs);
 
     return result;
diff --git a/Objects/typeobject.c b/Objects/typeobject.c
index 50a3c15..7203634 100644
--- a/Objects/typeobject.c
+++ b/Objects/typeobject.c
@@ -1,6 +1,7 @@
 /* Type object implementation */
 
 #include "Python.h"
+#include "pycore_call.h"
 #include "pycore_object.h"
 #include "pycore_pyerrors.h"
 #include "pycore_pystate.h"
@@ -6554,19 +6555,21 @@ slot_tp_hash(PyObject *self)
 static PyObject *
 slot_tp_call(PyObject *self, PyObject *args, PyObject *kwds)
 {
+    PyThreadState *tstate = _PyThreadState_GET();
     _Py_IDENTIFIER(__call__);
     int unbound;
-    PyObject *meth = lookup_method(self, &PyId___call__, &unbound);
-    PyObject *res;
 
-    if (meth == NULL)
+    PyObject *meth = lookup_method(self, &PyId___call__, &unbound);
+    if (meth == NULL) {
         return NULL;
+    }
 
+    PyObject *res;
     if (unbound) {
-        res = _PyObject_Call_Prepend(meth, self, args, kwds);
+        res = _PyObject_Call_Prepend(tstate, meth, self, args, kwds);
     }
     else {
-        res = PyObject_Call(meth, args, kwds);
+        res = _PyObject_Call(tstate, meth, args, kwds);
     }
 
     Py_DECREF(meth);
@@ -6688,17 +6691,16 @@ static PyObject *
 slot_tp_richcompare(PyObject *self, PyObject *other, int op)
 {
     PyThreadState *tstate = _PyThreadState_GET();
-    int unbound;
-    PyObject *func, *res;
 
-    func = lookup_maybe_method(self, &name_op[op], &unbound);
+    int unbound;
+    PyObject *func = lookup_maybe_method(self, &name_op[op], &unbound);
     if (func == NULL) {
         PyErr_Clear();
         Py_RETURN_NOTIMPLEMENTED;
     }
 
     PyObject *stack[2] = {self, other};
-    res = vectorcall_unbound(tstate, unbound, func, stack, 2);
+    PyObject *res = vectorcall_unbound(tstate, unbound, func, stack, 2);
     Py_DECREF(func);
     return res;
 }
@@ -6793,18 +6795,21 @@ slot_tp_descr_set(PyObject *self, PyObject *target, PyObject *value)
 static int
 slot_tp_init(PyObject *self, PyObject *args, PyObject *kwds)
 {
+    PyThreadState *tstate = _PyThreadState_GET();
+
     _Py_IDENTIFIER(__init__);
     int unbound;
     PyObject *meth = lookup_method(self, &PyId___init__, &unbound);
-    PyObject *res;
-
-    if (meth == NULL)
+    if (meth == NULL) {
         return -1;
+    }
+
+    PyObject *res;
     if (unbound) {
-        res = _PyObject_Call_Prepend(meth, self, args, kwds);
+        res = _PyObject_Call_Prepend(tstate, meth, self, args, kwds);
     }
     else {
-        res = PyObject_Call(meth, args, kwds);
+        res = _PyObject_Call(tstate, meth, args, kwds);
     }
     Py_DECREF(meth);
     if (res == NULL)
@@ -6823,6 +6828,7 @@ slot_tp_init(PyObject *self, PyObject *args, PyObject *kwds)
 static PyObject *
 slot_tp_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
 {
+    PyThreadState *tstate = _PyThreadState_GET();
     PyObject *func, *result;
 
     func = _PyObject_GetAttrId((PyObject *)type, &PyId___new__);
@@ -6830,7 +6836,7 @@ slot_tp_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
         return NULL;
     }
 
-    result = _PyObject_Call_Prepend(func, (PyObject *)type, args, kwds);
+    result = _PyObject_Call_Prepend(tstate, func, (PyObject *)type, args, kwds);
     Py_DECREF(func);
     return result;
 }
diff --git a/PCbuild/pythoncore.vcxproj b/PCbuild/pythoncore.vcxproj
index b724740..f5be8aa 100644
--- a/PCbuild/pythoncore.vcxproj
+++ b/PCbuild/pythoncore.vcxproj
@@ -160,6 +160,7 @@
     <ClInclude Include="..\Include\import.h" />
     <ClInclude Include="..\Include\internal\pycore_accu.h" />
     <ClInclude Include="..\Include\internal\pycore_atomic.h" />
+    <ClInclude Include="..\Include\internal\pycore_call.h" />
     <ClInclude Include="..\Include\internal\pycore_ceval.h" />
     <ClInclude Include="..\Include\internal\pycore_code.h" />
     <ClInclude Include="..\Include\internal\pycore_condvar.h" />
diff --git a/PCbuild/pythoncore.vcxproj.filters b/PCbuild/pythoncore.vcxproj.filters
index 8090813..ba1839f 100644
--- a/PCbuild/pythoncore.vcxproj.filters
+++ b/PCbuild/pythoncore.vcxproj.filters
@@ -183,6 +183,9 @@
     <ClInclude Include="..\Include\internal\pycore_atomic.h">
       <Filter>Include</Filter>
     </ClInclude>
+    <ClInclude Include="..\Include\internal\pycore_call.h">
+      <Filter>Include</Filter>
+    </ClInclude>
     <ClInclude Include="..\Include\internal\pycore_code.h">
       <Filter>Include</Filter>
     </ClInclude>
diff --git a/Python/_warnings.c b/Python/_warnings.c
index dd2b072..b8585d2 100644
--- a/Python/_warnings.c
+++ b/Python/_warnings.c
@@ -1,4 +1,5 @@
 #include "Python.h"
+#include "pycore_pyerrors.h"
 #include "pycore_pystate.h"
 #include "frameobject.h"
 #include "clinic/_warnings.c.h"
@@ -27,10 +28,11 @@ static struct PyModuleDef warningsmodule;
 static WarningsState *
 _Warnings_GetState()
 {
-    PyThreadState *tstate = PyThreadState_GET();
+    PyThreadState *tstate = _PyThreadState_GET();
     if (tstate == NULL) {
-        PyErr_SetString(PyExc_RuntimeError,
-                        "_Warnings_GetState: could not identify current interpreter");
+        _PyErr_SetString(tstate, PyExc_RuntimeError,
+                          "_Warnings_GetState: could not identify "
+                          "current interpreter");
         return NULL;
     }
     return &tstate->interp->warnings;
diff --git a/Python/bltinmodule.c b/Python/bltinmodule.c
index d79e254..4470310 100644
--- a/Python/bltinmodule.c
+++ b/Python/bltinmodule.c
@@ -4,6 +4,7 @@
 #include <ctype.h>
 #include "ast.h"
 #undef Yield   /* undefine macro conflicting with <winbase.h> */
+#include "pycore_pyerrors.h"
 #include "pycore_pystate.h"
 #include "pycore_tupleobject.h"
 
@@ -1254,23 +1255,23 @@ map_next(mapobject *lz)
 {
     PyObject *small_stack[_PY_FASTCALL_SMALL_STACK];
     PyObject **stack;
-    Py_ssize_t niters, nargs, i;
     PyObject *result = NULL;
+    PyThreadState *tstate = _PyThreadState_GET();
 
-    niters = PyTuple_GET_SIZE(lz->iters);
+    const Py_ssize_t niters = PyTuple_GET_SIZE(lz->iters);
     if (niters <= (Py_ssize_t)Py_ARRAY_LENGTH(small_stack)) {
         stack = small_stack;
     }
     else {
         stack = PyMem_Malloc(niters * sizeof(stack[0]));
         if (stack == NULL) {
-            PyErr_NoMemory();
+            _PyErr_NoMemory(tstate);
             return NULL;
         }
     }
 
-    nargs = 0;
-    for (i=0; i < niters; i++) {
+    Py_ssize_t nargs = 0;
+    for (Py_ssize_t i=0; i < niters; i++) {
         PyObject *it = PyTuple_GET_ITEM(lz->iters, i);
         PyObject *val = Py_TYPE(it)->tp_iternext(it);
         if (val == NULL) {
@@ -1280,10 +1281,10 @@ map_next(mapobject *lz)
         nargs++;
     }
 
-    result = _PyObject_FastCall(lz->func, stack, nargs);
+    result = _PyObject_VectorcallTstate(tstate, lz->func, stack, nargs, NULL);
 
 exit:
-    for (i=0; i < nargs; i++) {
+    for (Py_ssize_t i=0; i < nargs; i++) {
         Py_DECREF(stack[i]);
     }
     if (stack != small_stack) {
diff --git a/Python/ceval.c b/Python/ceval.c
index 046cd69..11d25a5 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -10,6 +10,7 @@
 #define PY_LOCAL_AGGRESSIVE
 
 #include "Python.h"
+#include "pycore_call.h"
 #include "pycore_ceval.h"
 #include "pycore_code.h"
 #include "pycore_object.h"
@@ -4306,7 +4307,6 @@ fail: /* Jump here from prelude on failure */
        current Python frame (f), the associated C stack is still in use,
        so recursion_depth must be boosted for the duration.
     */
-    assert(tstate != NULL);
     if (Py_REFCNT(f) > 1) {
         Py_DECREF(f);
         _PyObject_GC_TRACK(f);
@@ -5024,10 +5024,11 @@ do_call_core(PyThreadState *tstate, PyObject *func, PyObject *callargs, PyObject
                 return NULL;
             }
 
-            C_TRACE(result, _PyObject_FastCallDict(func,
-                                                   &_PyTuple_ITEMS(callargs)[1],
-                                                   nargs - 1,
-                                                   kwdict));
+            C_TRACE(result, _PyObject_FastCallDictTstate(
+                                    tstate, func,
+                                    &_PyTuple_ITEMS(callargs)[1],
+                                    nargs - 1,
+                                    kwdict));
             Py_DECREF(func);
             return result;
         }