bpo-36974: expand call protocol documentation (GH-13844)

CC @encukou 

I'm also adding Petr Viktorin as contributor for vectorcall in the "what's new" section.


https://bugs.python.org/issue36974



Automerge-Triggered-By: @encukou

Automerge-Triggered-By: @encukou

7 files changed, 462 insertions, 288 deletions

diff --git a/Doc/c-api/abstract.rst b/Doc/c-api/abstract.rst
index 0edd1d5..1823f9d 100644
--- a/Doc/c-api/abstract.rst
+++ b/Doc/c-api/abstract.rst
@@ -18,6 +18,7 @@ but whose items have not been set to some non-\ ``NULL`` value yet.
 .. toctree::
 
    object.rst
+   call.rst
    number.rst
    sequence.rst
    mapping.rst
diff --git a/Doc/c-api/call.rst b/Doc/c-api/call.rst
new file mode 100644
index 0000000..0833531
--- /dev/null
+++ b/Doc/c-api/call.rst
@@ -0,0 +1,411 @@
+.. highlight:: c
+
+.. _call:
+
+Call Protocol
+=============
+
+CPython supports two different calling protocols:
+*tp_call* and vectorcall.
+
+The *tp_call* Protocol
+----------------------
+
+Instances of classes that set :c:member:`~PyTypeObject.tp_call` are callable.
+The signature of the slot is::
+
+    PyObject *tp_call(PyObject *callable, PyObject *args, PyObject *kwargs);
+
+A call is made using a tuple for the positional arguments
+and a dict for the keyword arguments, similarly to
+``callable(*args, **kwargs)`` in Python code.
+*args* must be non-NULL (use an empty tuple if there are no arguments)
+but *kwargs* may be *NULL* if there are no keyword arguments.
+
+This convention is not only used by *tp_call*:
+:c:member:`~PyTypeObject.tp_new` and :c:member:`~PyTypeObject.tp_init`
+also pass arguments this way.
+
+To call an object, use :c:func:`PyObject_Call` or other
+:ref:`call API <capi-call>`.
+
+
+.. _vectorcall:
+
+The Vectorcall Protocol
+-----------------------
+
+.. versionadded:: 3.8
+
+The vectorcall protocol was introduced in :pep:`590` as an additional protocol
+for making calls more efficient.
+
+.. warning::
+
+   The vectorcall API is provisional and expected to become public in
+   Python 3.9, with a different names and, possibly, changed semantics.
+   If you use the it, plan for updating your code for Python 3.9.
+
+As rule of thumb, CPython will prefer the vectorcall for internal calls
+if the callable supports it. However, this is not a hard rule.
+Additionally, some third-party extensions use *tp_call* directly
+(rather than using :c:func:`PyObject_Call`).
+Therefore, a class supporting vectorcall must also implement
+:c:member:`~PyTypeObject.tp_call`.
+Moreover, the callable must behave the same
+regardless of which protocol is used.
+The recommended way to achieve this is by setting
+:c:member:`~PyTypeObject.tp_call` to :c:func:`PyVectorcall_Call`.
+This bears repeating:
+
+.. warning::
+
+   A class supporting vectorcall **must** also implement
+   :c:member:`~PyTypeObject.tp_call` with the same semantics.
+
+A class should not implement vectorcall if that would be slower
+than *tp_call*. For example, if the callee needs to convert
+the arguments to an args tuple and kwargs dict anyway, then there is no point
+in implementing vectorcall.
+
+Classes can implement the vectorcall protocol by enabling the
+:const:`_Py_TPFLAGS_HAVE_VECTORCALL` flag and setting
+:c:member:`~PyTypeObject.tp_vectorcall_offset` to the offset inside the
+object structure where a *vectorcallfunc* appears.
+This is a pointer to a function with the following signature:
+
+.. c:type:: PyObject *(*vectorcallfunc)(PyObject *callable, PyObject *const *args, size_t nargsf, PyObject *kwnames)
+
+- *callable* is the object being called.
+- *args* is a C array consisting of the positional arguments followed by the
+   values of the keyword arguments.
+   This can be *NULL* if there are no arguments.
+- *nargsf* is the number of positional arguments plus possibly the
+   :const:`PY_VECTORCALL_ARGUMENTS_OFFSET` flag.
+   To get the actual number of positional arguments from *nargsf*,
+   use :c:func:`PyVectorcall_NARGS`.
+- *kwnames* is a tuple containing the names of the keyword arguments;
+   in other words, the keys of the kwargs dict.
+   These names must be strings (instances of ``str`` or a subclass)
+   and they must be unique.
+   If there are no keyword arguments, then *kwnames* can instead be *NULL*.
+
+.. c:var:: PY_VECTORCALL_ARGUMENTS_OFFSET
+
+   If this flag is set in a vectorcall *nargsf* argument, the callee is allowed
+   to temporarily change ``args[-1]``. In other words, *args* points to
+   argument 1 (not 0) in the allocated vector.
+   The callee must restore the value of ``args[-1]`` before returning.
+
+   For :c:func:`_PyObject_VectorcallMethod`, this flag means instead that
+   ``args[0]`` may be changed.
+
+   Whenever they can do so cheaply (without additional allocation), callers
+   are encouraged to use :const:`PY_VECTORCALL_ARGUMENTS_OFFSET`.
+   Doing so will allow callables such as bound methods to make their onward
+   calls (which include a prepended *self* argument) very efficiently.
+
+To call an object that implements vectorcall, use a :ref:`call API <capi-call>`
+function as with any other callable.
+:c:func:`_PyObject_Vectorcall` will usually be most efficient.
+
+
+Recursion Control
+.................
+
+When using *tp_call*, callees do not need to worry about
+:ref:`recursion <recursion>`: CPython uses
+:c:func:`Py_EnterRecursiveCall` and :c:func:`Py_LeaveRecursiveCall`
+for calls made using *tp_call*.
+
+For efficiency, this is not the case for calls done using vectorcall:
+the callee should use *Py_EnterRecursiveCall* and *Py_LeaveRecursiveCall*
+if needed.
+
+
+Vectorcall Support API
+......................
+
+.. c:function:: Py_ssize_t PyVectorcall_NARGS(size_t nargsf)
+
+   Given a vectorcall *nargsf* argument, return the actual number of
+   arguments.
+   Currently equivalent to::
+
+      (Py_ssize_t)(nargsf & ~PY_VECTORCALL_ARGUMENTS_OFFSET)
+
+   However, the function ``PyVectorcall_NARGS`` should be used to allow
+   for future extensions.
+
+   .. versionadded:: 3.8
+
+.. c:function:: vectorcallfunc _PyVectorcall_Function(PyObject *op)
+
+   If *op* does not support the vectorcall protocol (either because the type
+   does not or because the specific instance does not), return *NULL*.
+   Otherwise, return the vectorcall function pointer stored in *op*.
+   This function never raises an exception.
+
+   This is mostly useful to check whether or not *op* supports vectorcall,
+   which can be done by checking ``_PyVectorcall_Function(op) != NULL``.
+
+   .. versionadded:: 3.8
+
+.. c:function:: PyObject* PyVectorcall_Call(PyObject *callable, PyObject *tuple, PyObject *dict)
+
+   Call *callable*'s :c:type:`vectorcallfunc` with positional and keyword
+   arguments given in a tuple and dict, respectively.
+
+   This is a specialized function, intended to be put in the
+   :c:member:`~PyTypeObject.tp_call` slot or be used in an implementation of ``tp_call``.
+   It does not check the :const:`_Py_TPFLAGS_HAVE_VECTORCALL` flag
+   and it does not fall back to ``tp_call``.
+
+   .. versionadded:: 3.8
+
+
+.. _capi-call:
+
+Object Calling API
+------------------
+
+Various functions are available for calling a Python object.
+Each converts its arguments to a convention supported by the called object –
+either *tp_call* or vectorcall.
+In order to do as litle conversion as possible, pick one that best fits
+the format of data you have available.
+
+The following table summarizes the available functions;
+please see individual documentation for details.
+
++------------------------------------------+------------------+--------------------+---------------+
+| Function                                 | callable         | args               | kwargs        |
++==========================================+==================+====================+===============+
+| :c:func:`PyObject_Call`                  | ``PyObject *``   | tuple              | dict/``NULL`` |
++------------------------------------------+------------------+--------------------+---------------+
+| :c:func:`PyObject_CallNoArgs`            | ``PyObject *``   | ---                | ---           |
++------------------------------------------+------------------+--------------------+---------------+
+| :c:func:`_PyObject_CallOneArg`           | ``PyObject *``   | 1 object           | ---           |
++------------------------------------------+------------------+--------------------+---------------+
+| :c:func:`PyObject_CallObject`            | ``PyObject *``   | tuple/``NULL``     | ---           |
++------------------------------------------+------------------+--------------------+---------------+
+| :c:func:`PyObject_CallFunction`          | ``PyObject *``   | format             | ---           |
++------------------------------------------+------------------+--------------------+---------------+
+| :c:func:`PyObject_CallMethod`            | obj + ``char*``  | format             | ---           |
++------------------------------------------+------------------+--------------------+---------------+
+| :c:func:`PyObject_CallFunctionObjArgs`   | ``PyObject *``   | variadic           | ---           |
++------------------------------------------+------------------+--------------------+---------------+
+| :c:func:`PyObject_CallMethodObjArgs`     | obj + name       | variadic           | ---           |
++------------------------------------------+------------------+--------------------+---------------+
+| :c:func:`_PyObject_CallMethodNoArgs`     | obj + name       | ---                | ---           |
++------------------------------------------+------------------+--------------------+---------------+
+| :c:func:`_PyObject_CallMethodOneArg`     | obj + name       | 1 object           | ---           |
++------------------------------------------+------------------+--------------------+---------------+
+| :c:func:`_PyObject_Vectorcall`           | ``PyObject *``   | vectorcall         | vectorcall    |
++------------------------------------------+------------------+--------------------+---------------+
+| :c:func:`_PyObject_FastCallDict`         | ``PyObject *``   | vectorcall         | dict/``NULL`` |
++------------------------------------------+------------------+--------------------+---------------+
+| :c:func:`_PyObject_VectorcallMethod`     | arg + name       | vectorcall         | vectorcall    |
++------------------------------------------+------------------+--------------------+---------------+
+
+
+.. c:function:: PyObject* PyObject_Call(PyObject *callable, PyObject *args, PyObject *kwargs)
+
+   Call a callable Python object *callable*, with arguments given by the
+   tuple *args*, and named arguments given by the dictionary *kwargs*.
+
+   *args* must not be *NULL*; use an empty tuple if no arguments are needed.
+   If no named arguments are needed, *kwargs* can be *NULL*.
+
+   Return the result of the call on success, or raise an exception and return
+   *NULL* on failure.
+
+   This is the equivalent of the Python expression:
+   ``callable(*args, **kwargs)``.
+
+
+.. c:function:: PyObject* PyObject_CallNoArgs(PyObject *callable)
+
+   Call a callable Python object *callable* without any arguments. It is the
+   most efficient way to call a callable Python object without any argument.
+
+   Return the result of the call on success, or raise an exception and return
+   *NULL* on failure.
+
+   .. versionadded:: 3.9
+
+
+.. c:function:: PyObject* _PyObject_CallOneArg(PyObject *callable, PyObject *arg)
+
+   Call a callable Python object *callable* with exactly 1 positional argument
+   *arg* and no keyword arguments.
+
+   Return the result of the call on success, or raise an exception and return
+   *NULL* on failure.
+
+   .. versionadded:: 3.9
+
+
+.. c:function:: PyObject* PyObject_CallObject(PyObject *callable, PyObject *args)
+
+   Call a callable Python object *callable*, with arguments given by the
+   tuple *args*.  If no arguments are needed, then *args* can be *NULL*.
+
+   Return the result of the call on success, or raise an exception and return
+   *NULL* on failure.
+
+   This is the equivalent of the Python expression: ``callable(*args)``.
+
+
+.. c:function:: PyObject* PyObject_CallFunction(PyObject *callable, const char *format, ...)
+
+   Call a callable Python object *callable*, with a variable number of C arguments.
+   The C arguments are described using a :c:func:`Py_BuildValue` style format
+   string.  The format can be *NULL*, indicating that no arguments are provided.
+
+   Return the result of the call on success, or raise an exception and return
+   *NULL* on failure.
+
+   This is the equivalent of the Python expression: ``callable(*args)``.
+
+   Note that if you only pass :c:type:`PyObject \*` args,
+   :c:func:`PyObject_CallFunctionObjArgs` is a faster alternative.
+
+   .. versionchanged:: 3.4
+      The type of *format* was changed from ``char *``.
+
+
+.. c:function:: PyObject* PyObject_CallMethod(PyObject *obj, const char *name, const char *format, ...)
+
+   Call the method named *name* of object *obj* with a variable number of C
+   arguments.  The C arguments are described by a :c:func:`Py_BuildValue` format
+   string that should produce a tuple.
+
+   The format can be *NULL*, indicating that no arguments are provided.
+
+   Return the result of the call on success, or raise an exception and return
+   *NULL* on failure.
+
+   This is the equivalent of the Python expression:
+   ``obj.name(arg1, arg2, ...)``.
+
+   Note that if you only pass :c:type:`PyObject \*` args,
+   :c:func:`PyObject_CallMethodObjArgs` is a faster alternative.
+
+   .. versionchanged:: 3.4
+      The types of *name* and *format* were changed from ``char *``.
+
+
+.. c:function:: PyObject* PyObject_CallFunctionObjArgs(PyObject *callable, ..., NULL)
+
+   Call a callable Python object *callable*, with a variable number of
+   :c:type:`PyObject \*` arguments.  The arguments are provided as a variable number
+   of parameters followed by *NULL*.
+
+   Return the result of the call on success, or raise an exception and return
+   *NULL* on failure.
+
+   This is the equivalent of the Python expression:
+   ``callable(arg1, arg2, ...)``.
+
+
+.. c:function:: PyObject* PyObject_CallMethodObjArgs(PyObject *obj, PyObject *name, ..., NULL)
+
+   Call a method of the Python object *obj*, where the name of the method is given as a
+   Python string object in *name*.  It is called with a variable number of
+   :c:type:`PyObject \*` arguments.  The arguments are provided as a variable number
+   of parameters followed by *NULL*.
+
+   Return the result of the call on success, or raise an exception and return
+   *NULL* on failure.
+
+
+.. c:function:: PyObject* _PyObject_CallMethodNoArgs(PyObject *obj, PyObject *name)
+
+   Call a method of the Python object *obj* without arguments,
+   where the name of the method is given as a Python string object in *name*.
+
+   Return the result of the call on success, or raise an exception and return
+   *NULL* on failure.
+
+   .. versionadded:: 3.9
+
+
+.. c:function:: PyObject* _PyObject_CallMethodOneArg(PyObject *obj, PyObject *name, PyObject *arg)
+
+   Call a method of the Python object *obj* with a single positional argument
+   *arg*, where the name of the method is given as a Python string object in
+   *name*.
+
+   Return the result of the call on success, or raise an exception and return
+   *NULL* on failure.
+
+   .. versionadded:: 3.9
+
+
+.. c:function:: PyObject* _PyObject_Vectorcall(PyObject *callable, PyObject *const *args, size_t nargsf, PyObject *kwnames)
+
+   Call a callable Python object *callable*.
+   The arguments are the same as for :c:type:`vectorcallfunc`.
+   If *callable* supports vectorcall_, this directly calls
+   the vectorcall function stored in *callable*.
+
+   Return the result of the call on success, or raise an exception and return
+   *NULL* on failure.
+
+   .. note::
+
+      This function is provisional and expected to become public in Python 3.9,
+      with a different name and, possibly, changed semantics.
+      If you use the function, plan for updating your code for Python 3.9.
+
+   .. versionadded:: 3.8
+
+.. c:function:: PyObject* _PyObject_FastCallDict(PyObject *callable, PyObject *const *args, size_t nargsf, PyObject *kwdict)
+
+   Call *callable* with positional arguments passed exactly as in the vectorcall_ protocol,
+   but with keyword arguments passed as a dictionary *kwdict*.
+   The *args* array contains only the positional arguments.
+
+   Regardless of which protocol is used internally,
+   a conversion of arguments needs to be done.
+   Therefore, this function should only be used if the caller
+   already has a dictionary ready to use for the keyword arguments,
+   but not a tuple for the positional arguments.
+
+   .. note::
+
+      This function is provisional and expected to become public in Python 3.9,
+      with a different name and, possibly, changed semantics.
+      If you use the function, plan for updating your code for Python 3.9.
+
+   .. versionadded:: 3.8
+
+.. c:function:: PyObject* _PyObject_VectorcallMethod(PyObject *name, PyObject *const *args, size_t nargsf, PyObject *kwnames)
+
+   Call a method using the vectorcall calling convention. The name of the method
+   is given as a Python string *name*. The object whose method is called is
+   *args[0]*, and the *args* array starting at *args[1]* represents the arguments
+   of the call. There must be at least one positional argument.
+   *nargsf* is the number of positional arguments including *args[0]*,
+   plus :const:`PY_VECTORCALL_ARGUMENTS_OFFSET` if the value of ``args[0]`` may
+   temporarily be changed. Keyword arguments can be passed just like in
+   :c:func:`_PyObject_Vectorcall`.
+
+   If the object has the :const:`Py_TPFLAGS_METHOD_DESCRIPTOR` feature,
+   this will call the unbound method object with the full
+   *args* vector as arguments.
+
+   Return the result of the call on success, or raise an exception and return
+   *NULL* on failure.
+
+   .. versionadded:: 3.9
+
+
+Call Support API
+----------------
+
+.. c:function:: int PyCallable_Check(PyObject *o)
+
+   Determine if the object *o* is callable.  Return ``1`` if the object is callable
+   and ``0`` otherwise.  This function always succeeds.
diff --git a/Doc/c-api/exceptions.rst b/Doc/c-api/exceptions.rst
index cd6df00..2edcbf7 100644
--- a/Doc/c-api/exceptions.rst
+++ b/Doc/c-api/exceptions.rst
@@ -697,6 +697,8 @@ The following functions are used to create and modify Unicode exceptions from C.
    ``0`` on success, ``-1`` on failure.
 
 
+.. _recursion:
+
 Recursion Control
 =================
 
@@ -704,6 +706,8 @@ These two functions provide a way to perform safe recursive calls at the C
 level, both in the core and in extension modules.  They are needed if the
 recursive code does not necessarily invoke Python code (which tracks its
 recursion depth automatically).
+They are also not needed for *tp_call* implementations
+because the :ref:`call protocol <call>` takes care of recursion handling.
 
 .. c:function:: int Py_EnterRecursiveCall(const char *where)
 
diff --git a/Doc/c-api/object.rst b/Doc/c-api/object.rst
index 7d7a3be..ca9db1a 100644
--- a/Doc/c-api/object.rst
+++ b/Doc/c-api/object.rst
@@ -248,246 +248,6 @@ Object Protocol
    of base classes).
 
 
-.. c:function:: int PyCallable_Check(PyObject *o)
-
-   Determine if the object *o* is callable.  Return ``1`` if the object is callable
-   and ``0`` otherwise.  This function always succeeds.
-
-
-.. c:function:: PyObject* PyObject_CallNoArgs(PyObject *callable)
-
-   Call a callable Python object *callable* without any arguments. It is the
-   most efficient way to call a callable Python object without any argument.
-
-   Return the result of the call on success, or raise an exception and return
-   ``NULL`` on failure.
-
-   .. versionadded:: 3.9
-
-
-.. c:function:: PyObject* _PyObject_CallOneArg(PyObject *callable, PyObject *arg)
-
-   Call a callable Python object *callable* with exactly 1 positional argument
-   *arg* and no keyword arguments.
-
-   Return the result of the call on success, or raise an exception and return
-   ``NULL`` on failure.
-
-   .. versionadded:: 3.9
-
-
-.. c:function:: PyObject* PyObject_Call(PyObject *callable, PyObject *args, PyObject *kwargs)
-
-   Call a callable Python object *callable*, with arguments given by the
-   tuple *args*, and named arguments given by the dictionary *kwargs*.
-
-   *args* must not be ``NULL``, use an empty tuple if no arguments are needed.
-   If no named arguments are needed, *kwargs* can be ``NULL``.
-
-   Return the result of the call on success, or raise an exception and return
-   ``NULL`` on failure.
-
-   This is the equivalent of the Python expression:
-   ``callable(*args, **kwargs)``.
-
-
-.. c:function:: PyObject* PyObject_CallObject(PyObject *callable, PyObject *args)
-
-   Call a callable Python object *callable*, with arguments given by the
-   tuple *args*.  If no arguments are needed, then *args* can be ``NULL``.
-
-   Return the result of the call on success, or raise an exception and return
-   ``NULL`` on failure.
-
-   This is the equivalent of the Python expression: ``callable(*args)``.
-
-
-.. c:function:: PyObject* PyObject_CallFunction(PyObject *callable, const char *format, ...)
-
-   Call a callable Python object *callable*, with a variable number of C arguments.
-   The C arguments are described using a :c:func:`Py_BuildValue` style format
-   string.  The format can be ``NULL``, indicating that no arguments are provided.
-
-   Return the result of the call on success, or raise an exception and return
-   ``NULL`` on failure.
-
-   This is the equivalent of the Python expression: ``callable(*args)``.
-
-   Note that if you only pass :c:type:`PyObject \*` args,
-   :c:func:`PyObject_CallFunctionObjArgs` is a faster alternative.
-
-   .. versionchanged:: 3.4
-      The type of *format* was changed from ``char *``.
-
-
-.. c:function:: PyObject* PyObject_CallMethod(PyObject *obj, const char *name, const char *format, ...)
-
-   Call the method named *name* of object *obj* with a variable number of C
-   arguments.  The C arguments are described by a :c:func:`Py_BuildValue` format
-   string that should  produce a tuple.
-
-   The format can be ``NULL``, indicating that no arguments are provided.
-
-   Return the result of the call on success, or raise an exception and return
-   ``NULL`` on failure.
-
-   This is the equivalent of the Python expression:
-   ``obj.name(arg1, arg2, ...)``.
-
-   Note that if you only pass :c:type:`PyObject \*` args,
-   :c:func:`PyObject_CallMethodObjArgs` is a faster alternative.
-
-   .. versionchanged:: 3.4
-      The types of *name* and *format* were changed from ``char *``.
-
-
-.. c:function:: PyObject* PyObject_CallFunctionObjArgs(PyObject *callable, ..., NULL)
-
-   Call a callable Python object *callable*, with a variable number of
-   :c:type:`PyObject\*` arguments.  The arguments are provided as a variable number
-   of parameters followed by ``NULL``.
-
-   Return the result of the call on success, or raise an exception and return
-   ``NULL`` on failure.
-
-   This is the equivalent of the Python expression:
-   ``callable(arg1, arg2, ...)``.
-
-
-.. c:function:: PyObject* PyObject_CallMethodObjArgs(PyObject *obj, PyObject *name, ..., NULL)
-
-   Calls a method of the Python object *obj*, where the name of the method is given as a
-   Python string object in *name*.  It is called with a variable number of
-   :c:type:`PyObject\*` arguments.  The arguments are provided as a variable number
-   of parameters followed by ``NULL``.
-
-   Return the result of the call on success, or raise an exception and return
-   ``NULL`` on failure.
-
-
-.. c:function:: PyObject* _PyObject_CallMethodNoArgs(PyObject *obj, PyObject *name)
-
-   Call a method of the Python object *obj* without arguments,
-   where the name of the method is given as a Python string object in *name*.
-
-   Return the result of the call on success, or raise an exception and return
-   ``NULL`` on failure.
-
-   .. versionadded:: 3.9
-
-
-.. c:function:: PyObject* _PyObject_CallMethodOneArg(PyObject *obj, PyObject *name, PyObject *arg)
-
-   Call a method of the Python object *obj* with a single positional argument
-   *arg*, where the name of the method is given as a Python string object in
-   *name*.
-
-   Return the result of the call on success, or raise an exception and return
-   ``NULL`` on failure.
-
-   .. versionadded:: 3.9
-
-
-.. c:function:: PyObject* _PyObject_Vectorcall(PyObject *callable, PyObject *const *args, size_t nargsf, PyObject *kwnames)
-
-   Call a callable Python object *callable*, using
-   :c:data:`vectorcall <PyTypeObject.tp_vectorcall_offset>` if possible.
-
-   *args* is a C array with the positional arguments.
-
-   *nargsf* is the number of positional arguments plus optionally the flag
-   :const:`PY_VECTORCALL_ARGUMENTS_OFFSET` (see below).
-   To get actual number of arguments, use
-   :c:func:`PyVectorcall_NARGS(nargsf) <PyVectorcall_NARGS>`.
-
-   *kwnames* can be either ``NULL`` (no keyword arguments) or a tuple of keyword
-   names, which must be strings. In the latter case, the values of the keyword
-   arguments are stored in *args* after the positional arguments.
-   The number of keyword arguments does not influence *nargsf*.
-
-   *kwnames* must contain only objects of type ``str`` (not a subclass),
-   and all keys must be unique.
-
-   Return the result of the call on success, or raise an exception and return
-   ``NULL`` on failure.
-
-   This uses the vectorcall protocol if the callable supports it;
-   otherwise, the arguments are converted to use
-   :c:member:`~PyTypeObject.tp_call`.
-
-   .. note::
-
-      This function is provisional and expected to become public in Python 3.9,
-      with a different name and, possibly, changed semantics.
-      If you use the function, plan for updating your code for Python 3.9.
-
-   .. versionadded:: 3.8
-
-.. c:var:: PY_VECTORCALL_ARGUMENTS_OFFSET
-
-   If set in a vectorcall *nargsf* argument, the callee is allowed to
-   temporarily change ``args[-1]``. In other words, *args* points to
-   argument 1 (not 0) in the allocated vector.
-   The callee must restore the value of ``args[-1]`` before returning.
-
-   For :c:func:`_PyObject_VectorcallMethod`, this flag means instead that
-   ``args[0]`` may be changed.
-
-   Whenever they can do so cheaply (without additional allocation), callers
-   are encouraged to use :const:`PY_VECTORCALL_ARGUMENTS_OFFSET`.
-   Doing so will allow callables such as bound methods to make their onward
-   calls (which include a prepended *self* argument) cheaply.
-
-   .. versionadded:: 3.8
-
-.. c:function:: Py_ssize_t PyVectorcall_NARGS(size_t nargsf)
-
-   Given a vectorcall *nargsf* argument, return the actual number of
-   arguments.
-   Currently equivalent to ``nargsf & ~PY_VECTORCALL_ARGUMENTS_OFFSET``.
-
-   .. versionadded:: 3.8
-
-.. c:function:: PyObject* _PyObject_FastCallDict(PyObject *callable, PyObject *const *args, size_t nargsf, PyObject *kwdict)
-
-   Same as :c:func:`_PyObject_Vectorcall` except that the keyword arguments
-   are passed as a dictionary in *kwdict*. This may be ``NULL`` if there
-   are no keyword arguments.
-
-   For callables supporting :c:data:`vectorcall <PyTypeObject.tp_vectorcall_offset>`,
-   the arguments are internally converted to the vectorcall convention.
-   Therefore, this function adds some overhead compared to
-   :c:func:`_PyObject_Vectorcall`.
-   It should only be used if the caller already has a dictionary ready to use.
-
-   .. note::
-
-      This function is provisional and expected to become public in Python 3.9,
-      with a different name and, possibly, changed semantics.
-      If you use the function, plan for updating your code for Python 3.9.
-
-   .. versionadded:: 3.8
-
-.. c:function:: PyObject* _PyObject_VectorcallMethod(PyObject *name, PyObject *const *args, size_t nargsf, PyObject *kwnames)
-
-   Call a method using the vectorcall calling convention. The name of the method
-   is given as Python string *name*. The object whose method is called is
-   *args[0]* and the *args* array starting at *args[1]* represents the arguments
-   of the call. There must be at least one positional argument.
-   *nargsf* is the number of positional arguments including *args[0]*,
-   plus :const:`PY_VECTORCALL_ARGUMENTS_OFFSET` if the value of ``args[0]`` may
-   temporarily be changed. Keyword arguments can be passed just like in
-   :c:func:`_PyObject_Vectorcall`.
-
-   If the object has the :const:`Py_TPFLAGS_METHOD_DESCRIPTOR` feature,
-   this will actually call the unbound method object with the full
-   *args* vector as arguments.
-
-   Return the result of the call on success, or raise an exception and return
-   ``NULL`` on failure.
-
-   .. versionadded:: 3.9
-
 .. c:function:: Py_hash_t PyObject_Hash(PyObject *o)
 
    .. index:: builtin: hash
diff --git a/Doc/c-api/structures.rst b/Doc/c-api/structures.rst
index 1035277..1bd769f 100644
--- a/Doc/c-api/structures.rst
+++ b/Doc/c-api/structures.rst
@@ -208,7 +208,8 @@ also keyword arguments.  So there are a total of 6 calling conventions:
 
    Extension of :const:`METH_FASTCALL` supporting also keyword arguments,
    with methods of type :c:type:`_PyCFunctionFastWithKeywords`.
-   Keyword arguments are passed the same way as in the vectorcall protocol:
+   Keyword arguments are passed the same way as in the
+   :ref:`vectorcall protocol <vectorcall>`:
    there is an additional fourth :c:type:`PyObject\*` parameter
    which is a tuple representing the names of the keyword arguments
    (which are guaranteed to be strings)
diff --git a/Doc/c-api/type.rst b/Doc/c-api/type.rst
index b1b2df9..41956b7 100644
--- a/Doc/c-api/type.rst
+++ b/Doc/c-api/type.rst
@@ -193,6 +193,7 @@ The following functions and structs are used to create
         (see :ref:`PyMemberDef <pymemberdef-offsets>`)
       * :c:member:`~PyTypeObject.tp_dictoffset`
         (see :ref:`PyMemberDef <pymemberdef-offsets>`)
+      * :c:member:`~PyTypeObject.tp_vectorcall_offset`
       * :c:member:`~PyBufferProcs.bf_getbuffer`
       * :c:member:`~PyBufferProcs.bf_releasebuffer`
 
diff --git a/Doc/c-api/typeobj.rst b/Doc/c-api/typeobj.rst
index bff5abf..b4ffd6b 100644
--- a/Doc/c-api/typeobj.rst
+++ b/Doc/c-api/typeobj.rst
@@ -49,7 +49,7 @@ Quick Reference
    +------------------------------------------------+-----------------------------------+-------------------+---+---+---+---+
    | :c:member:`~PyTypeObject.tp_dealloc`           | :c:type:`destructor`              |                   | X | X |   | X |
    +------------------------------------------------+-----------------------------------+-------------------+---+---+---+---+
-   | :c:member:`~PyTypeObject.tp_vectorcall_offset` | Py_ssize_t                        |                   |   |   |   | ? |
+   | :c:member:`~PyTypeObject.tp_vectorcall_offset` | Py_ssize_t                        |                   |   | X |   | X |
    +------------------------------------------------+-----------------------------------+-------------------+---+---+---+---+
    | (:c:member:`~PyTypeObject.tp_getattr`)         | :c:type:`getattrfunc`             | __getattribute__, |   |   |   | G |
    |                                                |                                   | __getattr__       |   |   |   |   |
@@ -145,6 +145,8 @@ Quick Reference
    +------------------------------------------------+-----------------------------------+-------------------+---+---+---+---+
    | :c:member:`~PyTypeObject.tp_finalize`          | :c:type:`destructor`              | __del__           |   |   |   | X |
    +------------------------------------------------+-----------------------------------+-------------------+---+---+---+---+
+   | :c:member:`~PyTypeObject.tp_vectorcall`        | :c:type:`vectorcallfunc`          |                   |   |   |   |   |
+   +------------------------------------------------+-----------------------------------+-------------------+---+---+---+---+
 
 If :const:`COUNT_ALLOCS` is defined then the following (internal-only)
 fields exist as well:
@@ -180,7 +182,7 @@ fields exist as well:
 
    .. code-block:: none
 
-      X - type slot is inherited via PyType_Ready if defined with a NULL value
+      X - type slot is inherited via *PyType_Ready* if defined with a *NULL* value
       % - the slots of the sub-struct are inherited individually
       G - inherited, but only in combination with other slots; see the slot's description
       ? - it's complicated; see the slot's description
@@ -687,42 +689,29 @@ and :c:type:`PyType_Type` effectively act as defaults.)
 .. c:member:: Py_ssize_t PyTypeObject.tp_vectorcall_offset
 
    An optional offset to a per-instance function that implements calling
-   the object using the *vectorcall* protocol, a more efficient alternative
+   the object using the :ref:`vectorcall protocol <vectorcall>`,
+   a more efficient alternative
    of the simpler :c:member:`~PyTypeObject.tp_call`.
 
    This field is only used if the flag :const:`_Py_TPFLAGS_HAVE_VECTORCALL`
    is set. If so, this must be a positive integer containing the offset in the
    instance of a :c:type:`vectorcallfunc` pointer.
-   The signature is the same as for :c:func:`_PyObject_Vectorcall`::
-
-        PyObject *vectorcallfunc(PyObject *callable, PyObject *const *args, size_t nargsf, PyObject *kwnames)
 
-   The *vectorcallfunc* pointer may be zero, in which case the instance behaves
+   The *vectorcallfunc* pointer may be ``NULL``, in which case the instance behaves
    as if :const:`_Py_TPFLAGS_HAVE_VECTORCALL` was not set: calling the instance
    falls back to :c:member:`~PyTypeObject.tp_call`.
 
    Any class that sets ``_Py_TPFLAGS_HAVE_VECTORCALL`` must also set
    :c:member:`~PyTypeObject.tp_call` and make sure its behaviour is consistent
    with the *vectorcallfunc* function.
-   This can be done by setting *tp_call* to ``PyVectorcall_Call``:
-
-   .. c:function:: PyObject *PyVectorcall_Call(PyObject *callable, PyObject *tuple, PyObject *dict)
+   This can be done by setting *tp_call* to :c:func:`PyVectorcall_Call`.
 
-      Call *callable*'s *vectorcallfunc* with positional and keyword
-      arguments given in a tuple and dict, respectively.
-
-      This function is intended to be used in the ``tp_call`` slot.
-      It does not fall back to ``tp_call`` and it currently does not check the
-      ``_Py_TPFLAGS_HAVE_VECTORCALL`` flag.
-      To call an object, use one of the :c:func:`PyObject_Call <PyObject_Call>`
-      functions instead.
-
-   .. note::
+   .. warning::
 
       It is not recommended for :ref:`heap types <heap-types>` to implement
       the vectorcall protocol.
-      When a user sets ``__call__`` in Python code, only ``tp_call`` is updated,
-      possibly making it inconsistent with the vectorcall function.
+      When a user sets :attr:`__call__` in Python code, only *tp_call* is updated,
+      likely making it inconsistent with the vectorcall function.
 
    .. note::
 
@@ -732,18 +721,19 @@ and :c:type:`PyType_Type` effectively act as defaults.)
 
    .. versionchanged:: 3.8
 
-      This slot was used for print formatting in Python 2.x.
-      In Python 3.0 to 3.7, it was reserved and named ``tp_print``.
+      Before version 3.8, this slot was named ``tp_print``.
+      In Python 2.x, it was used for printing to a file.
+      In Python 3.0 to 3.7, it was unused.
 
    **Inheritance:**
 
-   This field is inherited by subtypes together with
-   :c:member:`~PyTypeObject.tp_call`: a subtype inherits
-   :c:member:`~PyTypeObject.tp_vectorcall_offset` from its base type when
-   the subtype’s :c:member:`~PyTypeObject.tp_call` is ``NULL``.
-
-   Note that `heap types`_ (including subclasses defined in Python) do not
-   inherit the :const:`_Py_TPFLAGS_HAVE_VECTORCALL` flag.
+   This field is always inherited.
+   However, the :const:`_Py_TPFLAGS_HAVE_VECTORCALL` flag is not
+   always inherited. If it's not, then the subclass won't use
+   :ref:`vectorcall <vectorcall>`, except when
+   :c:func:`PyVectorcall_Call` is explicitly called.
+   This is in particular the case for `heap types`_
+   (including subclasses defined in Python).
 
 
 .. c:member:: getattrfunc PyTypeObject.tp_getattr
@@ -1171,18 +1161,17 @@ and :c:type:`PyType_Type` effectively act as defaults.)
          :c:member:`~PyTypeObject.tp_finalize` slot is always present in the
          type structure.
 
+
    .. data:: _Py_TPFLAGS_HAVE_VECTORCALL
 
-      This bit is set when the class implements the vectorcall protocol.
+      This bit is set when the class implements
+      the :ref:`vectorcall protocol <vectorcall>`.
       See :c:member:`~PyTypeObject.tp_vectorcall_offset` for details.
 
       **Inheritance:**
 
-      This bit is set on *static* subtypes if ``tp_flags`` is not overridden:
-      a subtype inherits ``_Py_TPFLAGS_HAVE_VECTORCALL`` from its base type
-      when the subtype’s :c:member:`~PyTypeObject.tp_call` is ``NULL``
-      and the subtype's ``Py_TPFLAGS_HEAPTYPE`` is not set.
-
+      This bit is inherited for *static* subtypes if
+      :c:member:`~PyTypeObject.tp_call` is also inherited.
       `Heap types`_ do not inherit ``_Py_TPFLAGS_HAVE_VECTORCALL``.
 
       .. note::
@@ -1715,9 +1704,9 @@ and :c:type:`PyType_Type` effectively act as defaults.)
 
       PyObject *tp_new(PyTypeObject *subtype, PyObject *args, PyObject *kwds);
 
-   The subtype argument is the type of the object being created; the *args* and
+   The *subtype* argument is the type of the object being created; the *args* and
    *kwds* arguments represent positional and keyword arguments of the call to the
-   type.  Note that subtype doesn't have to equal the type whose :c:member:`~PyTypeObject.tp_new`
+   type.  Note that *subtype* doesn't have to equal the type whose :c:member:`~PyTypeObject.tp_new`
    function is called; it may be a subtype of that type (but not an unrelated
    type).
 
@@ -1900,6 +1889,21 @@ and :c:type:`PyType_Type` effectively act as defaults.)
    .. seealso:: "Safe object finalization" (:pep:`442`)
 
 
+.. c:member:: vectorcallfunc PyTypeObject.tp_vectorcall
+
+   Vectorcall function to use for calls of this type object.
+   In other words, it is used to implement
+   :ref:`vectorcall <vectorcall>` for ``type.__call__``.
+   If ``tp_vectorcall`` is ``NULL``, the default call implementation
+   using :attr:`__new__` and :attr:`__init__` is used.
+
+   **Inheritance:**
+
+   This field is never inherited.
+
+   .. versionadded:: 3.9 (the field exists since 3.8 but it's only used since 3.9)
+
+
 The remaining fields are only defined if the feature test macro
 :const:`COUNT_ALLOCS` is defined, and are for internal use only. They are
 documented here for completeness.  None of these fields are inherited by
@@ -2369,14 +2373,6 @@ Slot Type typedefs
 
 .. c:type:: void (*destructor)(PyObject *)
 
-.. c:type:: PyObject *(*vectorcallfunc)(PyObject *callable, PyObject *const *args, size_t nargsf, PyObject *kwnames)
-
-   See :c:member:`~PyTypeObject.tp_vectorcall_offset`.
-
-   Arguments to ``vectorcallfunc`` are the same as for :c:func:`_PyObject_Vectorcall`.
-
-   .. versionadded:: 3.8
-
 .. c:type:: void (*freefunc)(void *)
 
    See :c:member:`~PyTypeObject.tp_free`.