FFI for PEP 590 Vectorcall

https://www.python.org/dev/peps/pep-0590/

This was tested on 3.7 using _PyCFunctionFast and 3.8
using PyObject_Vectorcall. Extending pyo3-derive-backend
to generate code using vectorcall is not done here.

This exposes PyObject_Vectorcall with a link_name to the
underscored name on 3.8 because it is expected to be stabilized
on 3.9.

This fixes the "fast" objects being new in 3.7, not 3.6.

CHANGELOG.md

@@ -5,6 +5,12 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/)
 and this project adheres to [Semantic Versioning](http://semver.org/spec/v2.0.0.html).
 
+## Unreleased
+
+### Added
+
+* FFI compatibility for PEP 590 Vectorcall.
+
 ## [0.8.1]
 
 ### Added

src/ffi/methodobject.rs

@@ -15,7 +15,39 @@ pub unsafe fn PyCFunction_Check(op: *mut PyObject) -> c_int {
 pub type PyCFunction =
     unsafe extern "C" fn(slf: *mut PyObject, args: *mut PyObject) -> *mut PyObject;
 
-#[cfg(all(Py_3_6, not(Py_LIMITED_API)))]
+#[cfg(all(Py_3_8, not(Py_LIMITED_API)))]
+#[cfg_attr(Py_3_8, link_name = "_PyObject_Vectorcall")]
+pub type PyObject_Vectorcall = unsafe extern "C" fn(
+    slf: *mut PyObject,
+    // positional and keyword arguments
+    args: *const *mut PyObject,
+    // number of position arguments in args, after which values are kwargs
+    nargs: crate::ffi::pyport::Py_ssize_t,
+    // tuple of kwargs, if given, or null
+    kwnames: *mut PyObject,
+) -> *mut PyObject;
+
+#[cfg(all(Py_3_8, not(Py_LIMITED_API)))]
+#[cfg_attr(Py_3_8, link_name = "PyVectorcall_Call")]
+pub type PyVectorcall_Call = unsafe extern "C" fn(
+    obj: *mut PyObject,
+    tuple: *mut PyObject,
+    dict: *mut PyObject,
+) -> *mut PyObject;
+
+#[cfg(all(Py_3_7, not(Py_LIMITED_API)))]
+const PY_VECTORCALL_ARGUMENTS_OFFSET: crate::ffi::pyport::Py_ssize_t =
+    1 << (8 * std::mem::size_of::<usize>() - 1);
+
+#[cfg(all(Py_3_7, not(Py_LIMITED_API)))]
+#[inline(always)]
+pub unsafe fn PyVectorcall_NARGS(
+    n: crate::ffi::pyport::Py_ssize_t,
+) -> crate::ffi::pyport::Py_ssize_t {
+    n & !PY_VECTORCALL_ARGUMENTS_OFFSET
+}
+
+#[cfg(all(Py_3_7, not(Py_LIMITED_API)))]
 pub type _PyCFunctionFast = unsafe extern "C" fn(
     slf: *mut PyObject,
     args: *mut *mut PyObject,
@@ -102,8 +134,10 @@ slot like sq_contains. */
 
 pub const METH_COEXIST: c_int = 0x0040;
 
-#[cfg(all(Py_3_6, not(Py_LIMITED_API)))]
-pub const METHOD_FASTCALL: c_int = 0x0080;
+/* METH_FASTCALL indicates the PEP 590 Vectorcall calling format. It may
+be specified alone or with METH_KEYWORDS. */
+#[cfg(all(Py_3_7, not(Py_LIMITED_API)))]
+pub const METH_FASTCALL: c_int = 0x0080;
 
 #[cfg_attr(windows, link(name = "pythonXY"))]
 extern "C" {

src/ffi/object.rs

@@ -463,7 +463,10 @@ mod typeobject {
         pub tp_basicsize: Py_ssize_t,
         pub tp_itemsize: Py_ssize_t,
         pub tp_dealloc: Option<object::destructor>,
+        #[cfg(not(Py_3_8))]
         pub tp_print: Option<object::printfunc>,
+        #[cfg(Py_3_8)]
+        pub tp_vectorcall_offset: Py_ssize_t,
         pub tp_getattr: Option<object::getattrfunc>,
         pub tp_setattr: Option<object::setattrfunc>,
         pub tp_as_async: *mut PyAsyncMethods,
@@ -529,7 +532,10 @@ mod typeobject {
                     tp_basicsize: 0,
                     tp_itemsize: 0,
                     tp_dealloc: None,
+                    #[cfg(not(Py_3_8))]
                     tp_print: None,
+                    #[cfg(Py_3_8)]
+                    tp_vectorcall_offset: 0,
                     tp_getattr: None,
                     tp_setattr: None,
                     $tp_as_async: ptr::null_mut(),
@@ -847,6 +853,10 @@ pub const Py_TPFLAGS_HEAPTYPE: c_ulong = (1 << 9);
 /// Set if the type allows subclassing
 pub const Py_TPFLAGS_BASETYPE: c_ulong = (1 << 10);
 
+/// Set if the type implements the vectorcall protocol (PEP 590)
+#[cfg(all(Py_3_8, not(Py_LIMITED_API)))]
+pub const _Py_TPFLAGS_HAVE_VECTORCALL: c_ulong = (1 << 11);
+
 /// Set if the type is 'ready' -- fully initialized
 pub const Py_TPFLAGS_READY: c_ulong = (1 << 12);