bpo-36974: implement PEP 590 (GH-13185)


Co-authored-by: Jeroen Demeyer <J.Demeyer@UGent.be>
Co-authored-by: Mark Shannon <mark@hotpy.org>
diff --git a/Objects/call.c b/Objects/call.c
index b608492..183a5c2 100644
--- a/Objects/call.c
+++ b/Objects/call.c
@@ -5,6 +5,10 @@
 #include "frameobject.h"
 
 
+static PyObject *
+cfunction_call_varargs(PyObject *func, PyObject *args, PyObject *kwargs);
+
+
 int
 _PyObject_HasFastCall(PyObject *callable)
 {
@@ -83,131 +87,132 @@
 /* --- Core PyObject call functions ------------------------------- */
 
 PyObject *
-_PyObject_FastCallDict(PyObject *callable, PyObject *const *args, Py_ssize_t nargs,
-                       PyObject *kwargs)
+_PyObject_FastCallDict(PyObject *callable, PyObject *const *args,
+                       size_t nargsf, PyObject *kwargs)
 {
     /* _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 */
     assert(!PyErr_Occurred());
-
     assert(callable != NULL);
+
+    Py_ssize_t nargs = PyVectorcall_NARGS(nargsf);
     assert(nargs >= 0);
     assert(nargs == 0 || args != NULL);
     assert(kwargs == NULL || PyDict_Check(kwargs));
 
-    if (PyFunction_Check(callable)) {
-        return _PyFunction_FastCallDict(callable, args, nargs, kwargs);
+    vectorcallfunc func = _PyVectorcall_Function(callable);
+    if (func == NULL) {
+        /* Use tp_call instead */
+        return _PyObject_MakeTpCall(callable, args, nargs, kwargs);
     }
-    else if (PyCFunction_Check(callable)) {
-        return _PyCFunction_FastCallDict(callable, args, nargs, kwargs);
+
+    PyObject *res;
+    if (kwargs == NULL) {
+        res = func(callable, args, nargsf, NULL);
     }
     else {
-        PyObject *argstuple, *result;
-        ternaryfunc call;
-
-        /* Slow-path: build a temporary tuple */
-        call = callable->ob_type->tp_call;
-        if (call == NULL) {
-            PyErr_Format(PyExc_TypeError, "'%.200s' object is not callable",
-                         callable->ob_type->tp_name);
+        PyObject *kwnames;
+        PyObject *const *newargs;
+        if (_PyStack_UnpackDict(args, nargs, kwargs, &newargs, &kwnames) < 0) {
             return NULL;
         }
-
-        argstuple = _PyTuple_FromArray(args, nargs);
-        if (argstuple == NULL) {
-            return NULL;
+        res = func(callable, newargs, nargs, kwnames);
+        if (kwnames != NULL) {
+            Py_ssize_t i, n = PyTuple_GET_SIZE(kwnames) + nargs;
+            for (i = 0; i < n; i++) {
+                Py_DECREF(newargs[i]);
+            }
+            PyMem_Free((PyObject **)newargs);
+            Py_DECREF(kwnames);
         }
-
-        if (Py_EnterRecursiveCall(" while calling a Python object")) {
-            Py_DECREF(argstuple);
-            return NULL;
-        }
-
-        result = (*call)(callable, argstuple, kwargs);
-
-        Py_LeaveRecursiveCall();
-        Py_DECREF(argstuple);
-
-        result = _Py_CheckFunctionResult(callable, result, NULL);
-        return result;
     }
+    return _Py_CheckFunctionResult(callable, res, NULL);
 }
 
 
 PyObject *
-_PyObject_FastCallKeywords(PyObject *callable, PyObject *const *stack, Py_ssize_t nargs,
-                           PyObject *kwnames)
+_PyObject_MakeTpCall(PyObject *callable, PyObject *const *args, Py_ssize_t nargs, PyObject *keywords)
 {
-    /* _PyObject_FastCallKeywords() must not be called with an exception set,
-       because it can clear it (directly or indirectly) and so the
-       caller loses its exception */
-    assert(!PyErr_Occurred());
+    /* 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;
+    if (call == NULL) {
+        PyErr_Format(PyExc_TypeError, "'%.200s' object is not callable",
+                     Py_TYPE(callable)->tp_name);
+        return NULL;
+    }
 
     assert(nargs >= 0);
-    assert(kwnames == NULL || PyTuple_CheckExact(kwnames));
-
-    /* kwnames must only contains str strings, no subclass, and all keys must
-       be unique: these checks are implemented in Python/ceval.c and
-       _PyArg_ParseStackAndKeywords(). */
-
-    if (PyFunction_Check(callable)) {
-        return _PyFunction_FastCallKeywords(callable, stack, nargs, kwnames);
+    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;
     }
-    if (PyCFunction_Check(callable)) {
-        return _PyCFunction_FastCallKeywords(callable, stack, nargs, kwnames);
+
+    PyObject *kwdict;
+    if (keywords == NULL || PyDict_Check(keywords)) {
+        kwdict = keywords;
     }
     else {
-        /* Slow-path: build a temporary tuple for positional arguments and a
-           temporary dictionary for keyword arguments (if any) */
-
-        ternaryfunc call;
-        PyObject *argstuple;
-        PyObject *kwdict, *result;
-        Py_ssize_t nkwargs;
-
-        nkwargs = (kwnames == NULL) ? 0 : PyTuple_GET_SIZE(kwnames);
-        assert((nargs == 0 && nkwargs == 0) || stack != NULL);
-
-        call = callable->ob_type->tp_call;
-        if (call == NULL) {
-            PyErr_Format(PyExc_TypeError, "'%.200s' object is not callable",
-                         callable->ob_type->tp_name);
-            return NULL;
-        }
-
-        argstuple = _PyTuple_FromArray(stack, nargs);
-        if (argstuple == NULL) {
-            return NULL;
-        }
-
-        if (nkwargs > 0) {
-            kwdict = _PyStack_AsDict(stack + nargs, kwnames);
+        if (PyTuple_GET_SIZE(keywords)) {
+            assert(args != NULL);
+            kwdict = _PyStack_AsDict(args + nargs, keywords);
             if (kwdict == NULL) {
                 Py_DECREF(argstuple);
                 return NULL;
             }
         }
         else {
-            kwdict = NULL;
+            keywords = kwdict = NULL;
         }
-
-        if (Py_EnterRecursiveCall(" while calling a Python object")) {
-            Py_DECREF(argstuple);
-            Py_XDECREF(kwdict);
-            return NULL;
-        }
-
-        result = (*call)(callable, argstuple, kwdict);
-
-        Py_LeaveRecursiveCall();
-
-        Py_DECREF(argstuple);
-        Py_XDECREF(kwdict);
-
-        result = _Py_CheckFunctionResult(callable, result, NULL);
-        return result;
     }
+
+    PyObject *result = NULL;
+    if (Py_EnterRecursiveCall(" while calling a Python object") == 0)
+    {
+        result = call(callable, argstuple, kwdict);
+        Py_LeaveRecursiveCall();
+    }
+
+    Py_DECREF(argstuple);
+    if (kwdict != keywords) {
+        Py_DECREF(kwdict);
+    }
+
+    result = _Py_CheckFunctionResult(callable, result, NULL);
+    return result;
+}
+
+
+PyObject *
+PyVectorcall_Call(PyObject *callable, PyObject *tuple, PyObject *kwargs)
+{
+    vectorcallfunc func = _PyVectorcall_Function(callable);
+    if (func == NULL) {
+        PyErr_Format(PyExc_TypeError, "'%.200s' object does not support vectorcall",
+                     Py_TYPE(callable)->tp_name);
+        return NULL;
+    }
+    PyObject *const *args;
+    Py_ssize_t nargs = PyTuple_GET_SIZE(tuple);
+    PyObject *kwnames;
+    if (_PyStack_UnpackDict(_PyTuple_ITEMS(tuple), nargs,
+        kwargs, &args, &kwnames) < 0) {
+        return NULL;
+    }
+    PyObject *result = func(callable, args, nargs, kwnames);
+    if (kwnames != NULL) {
+        Py_ssize_t i, n = PyTuple_GET_SIZE(kwnames) + nargs;
+        for (i = 0; i < n; i++) {
+            Py_DECREF(args[i]);
+        }
+        PyMem_Free((PyObject **)args);
+        Py_DECREF(kwnames);
+    }
+
+    return result;
 }
 
 
@@ -224,14 +229,13 @@
     assert(PyTuple_Check(args));
     assert(kwargs == NULL || PyDict_Check(kwargs));
 
-    if (PyFunction_Check(callable)) {
-        return _PyFunction_FastCallDict(callable,
-                                        _PyTuple_ITEMS(args),
-                                        PyTuple_GET_SIZE(args),
-                                        kwargs);
+    if (_PyVectorcall_Function(callable) != NULL) {
+        return PyVectorcall_Call(callable, args, kwargs);
     }
     else if (PyCFunction_Check(callable)) {
-        return PyCFunction_Call(callable, args, kwargs);
+        /* This must be a METH_VARARGS function, otherwise we would be
+         * in the previous case */
+        return cfunction_call_varargs(callable, args, kwargs);
     }
     else {
         call = callable->ob_type->tp_call;
@@ -384,9 +388,10 @@
     return result;
 }
 
+
 PyObject *
-_PyFunction_FastCallKeywords(PyObject *func, PyObject *const *stack,
-                             Py_ssize_t nargs, PyObject *kwnames)
+_PyFunction_FastCallKeywords(PyObject *func, PyObject* const* stack,
+                             size_t nargsf, PyObject *kwnames)
 {
     PyCodeObject *co = (PyCodeObject *)PyFunction_GET_CODE(func);
     PyObject *globals = PyFunction_GET_GLOBALS(func);
@@ -397,6 +402,7 @@
     Py_ssize_t nd;
 
     assert(PyFunction_Check(func));
+    Py_ssize_t nargs = PyVectorcall_NARGS(nargsf);
     assert(nargs >= 0);
     assert(kwnames == NULL || PyTuple_CheckExact(kwnames));
     assert((nargs == 0 && nkwargs == 0) || stack != NULL);
@@ -725,13 +731,14 @@
 
 PyObject *
 _PyCFunction_FastCallKeywords(PyObject *func,
-                              PyObject *const *args, Py_ssize_t nargs,
+                              PyObject *const *args, size_t nargsf,
                               PyObject *kwnames)
 {
     PyObject *result;
 
     assert(func != NULL);
     assert(PyCFunction_Check(func));
+    Py_ssize_t nargs = PyVectorcall_NARGS(nargsf);
 
     result = _PyMethodDef_RawFastCallKeywords(((PyCFunctionObject*)func)->m_ml,
                                               PyCFunction_GET_SELF(func),
@@ -751,6 +758,7 @@
     PyObject *self = PyCFunction_GET_SELF(func);
     PyObject *result;
 
+    assert(PyCFunction_GET_FLAGS(func) & METH_VARARGS);
     if (PyCFunction_GET_FLAGS(func) & METH_KEYWORDS) {
         if (Py_EnterRecursiveCall(" while calling a Python object")) {
             return NULL;
@@ -783,18 +791,12 @@
 PyObject *
 PyCFunction_Call(PyObject *func, PyObject *args, PyObject *kwargs)
 {
-    /* first try METH_VARARGS to pass directly args tuple unchanged.
-       _PyMethodDef_RawFastCallDict() creates a new temporary tuple
-       for METH_VARARGS. */
+    /* For METH_VARARGS, we cannot use vectorcall as the vectorcall pointer
+     * is NULL. This is intentional, since vectorcall would be slower. */
     if (PyCFunction_GET_FLAGS(func) & METH_VARARGS) {
         return cfunction_call_varargs(func, args, kwargs);
     }
-    else {
-        return _PyCFunction_FastCallDict(func,
-                                         _PyTuple_ITEMS(args),
-                                         PyTuple_GET_SIZE(args),
-                                         kwargs);
-    }
+    return PyVectorcall_Call(func, args, kwargs);
 }