bpo-35813: Added shared_memory submodule of multiprocessing. (#11664)

Added shared_memory submodule to multiprocessing in time for first alpha with cross-platform tests soon to follow.

4 files changed, 1310 insertions, 0 deletions

diff --git a/Lib/multiprocessing/shared_memory.py b/Lib/multiprocessing/shared_memory.py
new file mode 100644
index 0000000..11eac4b
--- /dev/null
+++ b/Lib/multiprocessing/shared_memory.py
@@ -0,0 +1,573 @@
+"Provides shared memory for direct access across processes."
+
+
+__all__ = [ 'SharedMemory', 'PosixSharedMemory', 'WindowsNamedSharedMemory',
+            'ShareableList', 'shareable_wrap',
+            'SharedMemoryServer', 'SharedMemoryManager', 'SharedMemoryTracker' ]
+
+
+from functools import reduce
+import mmap
+from .managers import DictProxy, SyncManager, Server
+from . import util
+import os
+import random
+import struct
+import sys
+try:
+    from _posixshmem import _PosixSharedMemory, Error, ExistentialError, O_CREX
+except ImportError as ie:
+    if os.name != "nt":
+        # On Windows, posixshmem is not required to be available.
+        raise ie
+    else:
+        _PosixSharedMemory = object
+        class ExistentialError(BaseException): pass
+        class Error(BaseException): pass
+        O_CREX = -1
+
+
+class WindowsNamedSharedMemory:
+
+    def __init__(self, name, flags=None, mode=None, size=None, read_only=False):
+        if name is None:
+            name = f'wnsm_{os.getpid()}_{random.randrange(100000)}'
+
+        self._mmap = mmap.mmap(-1, size, tagname=name)
+        self.buf = memoryview(self._mmap)
+        self.name = name
+        self.size = size
+
+    def __repr__(self):
+        return f'{self.__class__.__name__}({self.name!r}, size={self.size})'
+
+    def close(self):
+        self.buf.release()
+        self._mmap.close()
+
+    def unlink(self):
+        """Windows ensures that destruction of the last reference to this
+        named shared memory block will result in the release of this memory."""
+        pass
+
+
+class PosixSharedMemory(_PosixSharedMemory):
+
+    def __init__(self, name, flags=None, mode=None, size=None, read_only=False):
+        if name and (flags is None):
+            _PosixSharedMemory.__init__(self, name)
+        else:
+            if name is None:
+                name = f'psm_{os.getpid()}_{random.randrange(100000)}'
+            _PosixSharedMemory.__init__(self, name, flags=O_CREX, size=size)
+
+        self._mmap = mmap.mmap(self.fd, self.size)
+        self.buf = memoryview(self._mmap)
+
+    def __repr__(self):
+        return f'{self.__class__.__name__}({self.name!r}, size={self.size})'
+
+    def close(self):
+        self.buf.release()
+        self._mmap.close()
+        self.close_fd()
+
+
+class SharedMemory:
+
+    def __new__(cls, *args, **kwargs):
+        if os.name == 'nt':
+            cls = WindowsNamedSharedMemory
+        else:
+            cls = PosixSharedMemory
+        return cls(*args, **kwargs)
+
+
+def shareable_wrap(
+    existing_obj=None,
+    shmem_name=None,
+    cls=None,
+    shape=(0,),
+    strides=None,
+    dtype=None,
+    format=None,
+    **kwargs
+):
+    augmented_kwargs = dict(kwargs)
+    extras = dict(shape=shape, strides=strides, dtype=dtype, format=format)
+    for key, value in extras.items():
+        if value is not None:
+            augmented_kwargs[key] = value
+
+    if existing_obj is not None:
+        existing_type = getattr(
+            existing_obj,
+            "_proxied_type",
+            type(existing_obj)
+        )
+
+        #agg = existing_obj.itemsize
+        #size = [ agg := i * agg for i in existing_obj.shape ][-1]
+        # TODO: replace use of reduce below with above 2 lines once available
+        size = reduce(
+            lambda x, y: x * y,
+            existing_obj.shape,
+            existing_obj.itemsize
+        )
+
+    else:
+        assert shmem_name is not None
+        existing_type = cls
+        size = 1
+
+    shm = SharedMemory(shmem_name, size=size)
+
+    class CustomShareableProxy(existing_type):
+
+        def __init__(self, *args, buffer=None, **kwargs):
+            # If copy method called, prevent recursion from replacing _shm.
+            if not hasattr(self, "_shm"):
+                self._shm = shm
+                self._proxied_type = existing_type
+            else:
+                # _proxied_type only used in pickling.
+                assert hasattr(self, "_proxied_type")
+            try:
+                existing_type.__init__(self, *args, **kwargs)
+            except:
+                pass
+
+        def __repr__(self):
+            if not hasattr(self, "_shm"):
+                return existing_type.__repr__(self)
+            formatted_pairs = (
+                "%s=%r" % kv for kv in self._build_state(self).items()
+            )
+            return f"{self.__class__.__name__}({', '.join(formatted_pairs)})"
+
+        #def __getstate__(self):
+        #    if not hasattr(self, "_shm"):
+        #        return existing_type.__getstate__(self)
+        #    state = self._build_state(self)
+        #    return state
+
+        #def __setstate__(self, state):
+        #    self.__init__(**state)
+
+        def __reduce__(self):
+            return (
+                shareable_wrap,
+                (
+                    None,
+                    self._shm.name,
+                    self._proxied_type,
+                    self.shape,
+                    self.strides,
+                    self.dtype.str if hasattr(self, "dtype") else None,
+                    getattr(self, "format", None),
+                ),
+            )
+
+        def copy(self):
+            dupe = existing_type.copy(self)
+            if not hasattr(dupe, "_shm"):
+                dupe = shareable_wrap(dupe)
+            return dupe
+
+        @staticmethod
+        def _build_state(existing_obj, generics_only=False):
+            state = {
+                "shape": existing_obj.shape,
+                "strides": existing_obj.strides,
+            }
+            try:
+                state["dtype"] = existing_obj.dtype
+            except AttributeError:
+                try:
+                    state["format"] = existing_obj.format
+                except AttributeError:
+                    pass
+            if not generics_only:
+                try:
+                    state["shmem_name"] = existing_obj._shm.name
+                    state["cls"] = existing_type
+                except AttributeError:
+                    pass
+            return state
+
+    proxy_type = type(
+        f"{existing_type.__name__}Shareable",
+        CustomShareableProxy.__bases__,
+        dict(CustomShareableProxy.__dict__),
+    )
+
+    if existing_obj is not None:
+        try:
+            proxy_obj = proxy_type(
+                buffer=shm.buf,
+                **proxy_type._build_state(existing_obj)
+            )
+        except Exception:
+            proxy_obj = proxy_type(
+                buffer=shm.buf,
+                **proxy_type._build_state(existing_obj, True)
+            )
+
+        mveo = memoryview(existing_obj)
+        proxy_obj._shm.buf[:mveo.nbytes] = mveo.tobytes()
+
+    else:
+        proxy_obj = proxy_type(buffer=shm.buf, **augmented_kwargs)
+
+    return proxy_obj
+
+
+encoding = "utf8"
+
+class ShareableList:
+    """Pattern for a mutable list-like object shareable via a shared
+    memory block.  It differs from the built-in list type in that these
+    lists can not change their overall length (i.e. no append, insert,
+    etc.)
+
+    Because values are packed into a memoryview as bytes, the struct
+    packing format for any storable value must require no more than 8
+    characters to describe its format."""
+
+    # TODO: Adjust for discovered word size of machine.
+    types_mapping = {
+        int: "q",
+        float: "d",
+        bool: "xxxxxxx?",
+        str: "%ds",
+        bytes: "%ds",
+        None.__class__: "xxxxxx?x",
+    }
+    alignment = 8
+    back_transform_codes = {
+        0: lambda value: value,                   # int, float, bool
+        1: lambda value: value.rstrip(b'\x00').decode(encoding),  # str
+        2: lambda value: value.rstrip(b'\x00'),   # bytes
+        3: lambda _value: None,                   # None
+    }
+
+    @staticmethod
+    def _extract_recreation_code(value):
+        """Used in concert with back_transform_codes to convert values
+        into the appropriate Python objects when retrieving them from
+        the list as well as when storing them."""
+        if not isinstance(value, (str, bytes, None.__class__)):
+            return 0
+        elif isinstance(value, str):
+            return 1
+        elif isinstance(value, bytes):
+            return 2
+        else:
+            return 3  # NoneType
+
+    def __init__(self, iterable=None, name=None):
+        if iterable is not None:
+            _formats = [
+                self.types_mapping[type(item)]
+                    if not isinstance(item, (str, bytes))
+                    else self.types_mapping[type(item)] % (
+                        self.alignment * (len(item) // self.alignment + 1),
+                    )
+                for item in iterable
+            ]
+            self._list_len = len(_formats)
+            assert sum(len(fmt) <= 8 for fmt in _formats) == self._list_len
+            self._allocated_bytes = tuple(
+                    self.alignment if fmt[-1] != "s" else int(fmt[:-1])
+                    for fmt in _formats
+            )
+            _back_transform_codes = [
+                self._extract_recreation_code(item) for item in iterable
+            ]
+            requested_size = struct.calcsize(
+                "q" + self._format_size_metainfo + "".join(_formats)
+            )
+
+        else:
+            requested_size = 1  # Some platforms require > 0.
+
+        self.shm = SharedMemory(name, size=requested_size)
+
+        if iterable is not None:
+            _enc = encoding
+            struct.pack_into(
+                "q" + self._format_size_metainfo,
+                self.shm.buf,
+                0,
+                self._list_len,
+                *(self._allocated_bytes)
+            )
+            struct.pack_into(
+                "".join(_formats),
+                self.shm.buf,
+                self._offset_data_start,
+                *(v.encode(_enc) if isinstance(v, str) else v for v in iterable)
+            )
+            struct.pack_into(
+                self._format_packing_metainfo,
+                self.shm.buf,
+                self._offset_packing_formats,
+                *(v.encode(_enc) for v in _formats)
+            )
+            struct.pack_into(
+                self._format_back_transform_codes,
+                self.shm.buf,
+                self._offset_back_transform_codes,
+                *(_back_transform_codes)
+            )
+
+        else:
+            self._list_len = len(self)  # Obtains size from offset 0 in buffer.
+            self._allocated_bytes = struct.unpack_from(
+                self._format_size_metainfo,
+                self.shm.buf,
+                1 * 8
+            )
+
+    def _get_packing_format(self, position):
+        "Gets the packing format for a single value stored in the list."
+        position = position if position >= 0 else position + self._list_len
+        if (position >= self._list_len) or (self._list_len < 0):
+            raise IndexError("Requested position out of range.")
+
+        v = struct.unpack_from(
+            "8s",
+            self.shm.buf,
+            self._offset_packing_formats + position * 8
+        )[0]
+        fmt = v.rstrip(b'\x00')
+        fmt_as_str = fmt.decode(encoding)
+
+        return fmt_as_str
+
+    def _get_back_transform(self, position):
+        "Gets the back transformation function for a single value."
+
+        position = position if position >= 0 else position + self._list_len
+        if (position >= self._list_len) or (self._list_len < 0):
+            raise IndexError("Requested position out of range.")
+
+        transform_code = struct.unpack_from(
+            "b",
+            self.shm.buf,
+            self._offset_back_transform_codes + position
+        )[0]
+        transform_function = self.back_transform_codes[transform_code]
+
+        return transform_function
+
+    def _set_packing_format_and_transform(self, position, fmt_as_str, value):
+        """Sets the packing format and back transformation code for a
+        single value in the list at the specified position."""
+
+        position = position if position >= 0 else position + self._list_len
+        if (position >= self._list_len) or (self._list_len < 0):
+            raise IndexError("Requested position out of range.")
+
+        struct.pack_into(
+            "8s",
+            self.shm.buf,
+            self._offset_packing_formats + position * 8,
+            fmt_as_str.encode(encoding)
+        )
+
+        transform_code = self._extract_recreation_code(value)
+        struct.pack_into(
+            "b",
+            self.shm.buf,
+            self._offset_back_transform_codes + position,
+            transform_code
+        )
+
+    def __getitem__(self, position):
+        try:
+            offset = self._offset_data_start \
+                     + sum(self._allocated_bytes[:position])
+            (v,) = struct.unpack_from(
+                self._get_packing_format(position),
+                self.shm.buf,
+                offset
+            )
+        except IndexError:
+            raise IndexError("index out of range")
+
+        back_transform = self._get_back_transform(position)
+        v = back_transform(v)
+
+        return v
+
+    def __setitem__(self, position, value):
+        try:
+            offset = self._offset_data_start \
+                     + sum(self._allocated_bytes[:position])
+            current_format = self._get_packing_format(position)
+        except IndexError:
+            raise IndexError("assignment index out of range")
+
+        if not isinstance(value, (str, bytes)):
+            new_format = self.types_mapping[type(value)]
+        else:
+            if len(value) > self._allocated_bytes[position]:
+                raise ValueError("exceeds available storage for existing str")
+            if current_format[-1] == "s":
+                new_format = current_format
+            else:
+                new_format = self.types_mapping[str] % (
+                    self._allocated_bytes[position],
+                )
+
+        self._set_packing_format_and_transform(
+            position,
+            new_format,
+            value
+        )
+        value = value.encode(encoding) if isinstance(value, str) else value
+        struct.pack_into(new_format, self.shm.buf, offset, value)
+
+    def __len__(self):
+        return struct.unpack_from("q", self.shm.buf, 0)[0]
+
+    @property
+    def format(self):
+        "The struct packing format used by all currently stored values."
+        return "".join(self._get_packing_format(i) for i in range(self._list_len))
+
+    @property
+    def _format_size_metainfo(self):
+        "The struct packing format used for metainfo on storage sizes."
+        return f"{self._list_len}q"
+
+    @property
+    def _format_packing_metainfo(self):
+        "The struct packing format used for the values' packing formats."
+        return "8s" * self._list_len
+
+    @property
+    def _format_back_transform_codes(self):
+        "The struct packing format used for the values' back transforms."
+        return "b" * self._list_len
+
+    @property
+    def _offset_data_start(self):
+        return (self._list_len + 1) * 8  # 8 bytes per "q"
+
+    @property
+    def _offset_packing_formats(self):
+        return self._offset_data_start + sum(self._allocated_bytes)
+
+    @property
+    def _offset_back_transform_codes(self):
+        return self._offset_packing_formats + self._list_len * 8
+
+    @classmethod
+    def copy(cls, self):
+        "L.copy() -> ShareableList -- a shallow copy of L."
+
+        return cls(self)
+
+    def count(self, value):
+        "L.count(value) -> integer -- return number of occurrences of value."
+
+        return sum(value == entry for entry in self)
+
+    def index(self, value):
+        """L.index(value) -> integer -- return first index of value.
+        Raises ValueError if the value is not present."""
+
+        for position, entry in enumerate(self):
+            if value == entry:
+                return position
+        else:
+            raise ValueError(f"{value!r} not in this container")
+
+
+class SharedMemoryTracker:
+    "Manages one or more shared memory segments."
+
+    def __init__(self, name, segment_names=[]):
+        self.shared_memory_context_name = name
+        self.segment_names = segment_names
+
+    def register_segment(self, segment):
+        util.debug(f"Registering segment {segment.name!r} in pid {os.getpid()}")
+        self.segment_names.append(segment.name)
+
+    def destroy_segment(self, segment_name):
+        util.debug(f"Destroying segment {segment_name!r} in pid {os.getpid()}")
+        self.segment_names.remove(segment_name)
+        segment = SharedMemory(segment_name, size=1)
+        segment.close()
+        segment.unlink()
+
+    def unlink(self):
+        for segment_name in self.segment_names[:]:
+            self.destroy_segment(segment_name)
+
+    def __del__(self):
+        util.debug(f"Called {self.__class__.__name__}.__del__ in {os.getpid()}")
+        self.unlink()
+
+    def __getstate__(self):
+        return (self.shared_memory_context_name, self.segment_names)
+
+    def __setstate__(self, state):
+        self.__init__(*state)
+
+    def wrap(self, obj_exposing_buffer_protocol):
+        wrapped_obj = shareable_wrap(obj_exposing_buffer_protocol)
+        self.register_segment(wrapped_obj._shm)
+        return wrapped_obj
+
+
+class SharedMemoryServer(Server):
+    def __init__(self, *args, **kwargs):
+        Server.__init__(self, *args, **kwargs)
+        self.shared_memory_context = \
+            SharedMemoryTracker(f"shmm_{self.address}_{os.getpid()}")
+        util.debug(f"SharedMemoryServer started by pid {os.getpid()}")
+
+    def create(self, c, typeid, *args, **kwargs):
+        # Unless set up as a shared proxy, don't make shared_memory_context
+        # a standard part of kwargs.  This makes things easier for supplying
+        # simple functions.
+        if hasattr(self.registry[typeid][-1], "_shared_memory_proxy"):
+            kwargs['shared_memory_context'] = self.shared_memory_context
+        return Server.create(self, c, typeid, *args, **kwargs)
+
+    def shutdown(self, c):
+        self.shared_memory_context.unlink()
+        return Server.shutdown(self, c)
+
+
+class SharedMemoryManager(SyncManager):
+    """Like SyncManager but uses SharedMemoryServer instead of Server.
+
+    TODO: Consider relocate/merge into managers submodule."""
+
+    _Server = SharedMemoryServer
+
+    def __init__(self, *args, **kwargs):
+        SyncManager.__init__(self, *args, **kwargs)
+        util.debug(f"{self.__class__.__name__} created by pid {os.getpid()}")
+
+    def __del__(self):
+        util.debug(f"{self.__class__.__name__} told die by pid {os.getpid()}")
+        pass
+
+    def get_server(self):
+        'Better than monkeypatching for now; merge into Server ultimately'
+        if self._state.value != State.INITIAL:
+            if self._state.value == State.STARTED:
+                raise ProcessError("Already started server")
+            elif self._state.value == State.SHUTDOWN:
+                raise ProcessError("Manager has shut down")
+            else:
+                raise ProcessError(
+                    "Unknown state {!r}".format(self._state.value))
+        return _Server(self._registry, self._address,
+                       self._authkey, self._serializer)
diff --git a/Misc/NEWS.d/next/Library/2019-01-23-22-44-37.bpo-35813.Yobj-Y.rst b/Misc/NEWS.d/next/Library/2019-01-23-22-44-37.bpo-35813.Yobj-Y.rst
new file mode 100644
index 0000000..714a24c
--- /dev/null
+++ b/Misc/NEWS.d/next/Library/2019-01-23-22-44-37.bpo-35813.Yobj-Y.rst
@@ -0,0 +1,2 @@
+Shared memory submodule added to multiprocessing to avoid need for
+serialization between processes
diff --git a/Modules/_multiprocessing/posixshmem.c b/Modules/_multiprocessing/posixshmem.c
new file mode 100644
index 0000000..7dd29f4
--- /dev/null
+++ b/Modules/_multiprocessing/posixshmem.c
@@ -0,0 +1,724 @@
+/*
+posixshmem - A Python module for accessing POSIX 1003.1b-1993 shared memory.
+
+Copyright (c) 2012, Philip Semanchuk
+Copyright (c) 2018, 2019, Davin Potts
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+    * Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in the
+      documentation and/or other materials provided with the distribution.
+    * Neither the name of posixshmem nor the names of its contributors may
+      be used to endorse or promote products derived from this software
+      without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY ITS CONTRIBUTORS ''AS IS'' AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL Philip Semanchuk BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#define PY_SSIZE_T_CLEAN
+
+#include <Python.h>
+#include "structmember.h"
+
+#include <time.h>
+#include <sys/time.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <stdio.h>
+
+// For shared memory stuff
+#include <sys/stat.h>
+#include <sys/mman.h>
+
+/* SEM_FAILED is defined as an int in Apple's headers, and this makes the
+compiler complain when I compare it to a pointer. Python faced the same
+problem (issue 9586) and I copied their solution here.
+ref: http://bugs.python.org/issue9586
+
+Note that in /Developer/SDKs/MacOSX10.4u.sdk/usr/include/sys/semaphore.h,
+SEM_FAILED is #defined as -1 and that's apparently the definition used by
+Python when building. In /usr/include/sys/semaphore.h, it's defined
+as ((sem_t *)-1).
+*/
+#ifdef __APPLE__
+    #undef  SEM_FAILED
+    #define SEM_FAILED ((sem_t *)-1)
+#endif
+
+/* POSIX says that a mode_t "shall be an integer type". To avoid the need
+for a specific get_mode function for each type, I'll just stuff the mode into
+a long and mention it in the Xxx_members list for each type.
+ref: http://www.opengroup.org/onlinepubs/000095399/basedefs/sys/types.h.html
+*/
+
+typedef struct {
+    PyObject_HEAD
+    char *name;
+    long mode;
+    int fd;
+} SharedMemory;
+
+
+// FreeBSD (and perhaps other BSDs) limit names to 14 characters. In the
+// code below, strings of this length are allocated on the stack, so
+// increase this gently or change that code to use malloc().
+#define MAX_SAFE_NAME_LENGTH  14
+
+
+/* Struct to contain an IPC object name which can be None */
+typedef struct {
+    int is_none;
+    char *name;
+} NoneableName;
+
+
+/*
+      Exceptions for this module
+*/
+
+static PyObject *pBaseException;
+static PyObject *pPermissionsException;
+static PyObject *pExistentialException;
+
+
+#ifdef POSIX_IPC_DEBUG
+#define DPRINTF(fmt, args...) fprintf(stderr, "+++ " fmt, ## args)
+#else
+#define DPRINTF(fmt, args...)
+#endif
+
+static char *
+bytes_to_c_string(PyObject* o, int lock) {
+/* Convert a bytes object to a char *. Optionally lock the buffer if it is a
+   bytes array.
+   This code swiped directly from Python 3.1's posixmodule.c by Philip S.
+   The name there is bytes2str().
+*/
+    if (PyBytes_Check(o))
+        return PyBytes_AsString(o);
+    else if (PyByteArray_Check(o)) {
+        if (lock && PyObject_GetBuffer(o, NULL, 0) < 0)
+            /* On a bytearray, this should not fail. */
+            PyErr_BadInternalCall();
+        return PyByteArray_AsString(o);
+    } else {
+        /* The FS converter should have verified that this
+           is either bytes or bytearray. */
+        Py_FatalError("bad object passed to bytes2str");
+        /* not reached. */
+        return "";
+    }
+}
+
+static void
+release_bytes(PyObject* o)
+    /* Release the lock, decref the object.
+   This code swiped directly from Python 3.1's posixmodule.c by Philip S.
+   */
+{
+    if (PyByteArray_Check(o))
+        o->ob_type->tp_as_buffer->bf_releasebuffer(NULL, 0);
+    Py_DECREF(o);
+}
+
+
+static int
+random_in_range(int min, int max) {
+    // returns a random int N such that min <= N <= max
+    int diff = (max - min) + 1;
+
+    // ref: http://www.c-faq.com/lib/randrange.html
+    return ((int)((double)rand() / ((double)RAND_MAX + 1) * diff)) + min;
+}
+
+
+static
+int create_random_name(char *name) {
+    // The random name is always lowercase so that this code will work
+    // on case-insensitive file systems. It always starts with a forward
+    // slash.
+    int length;
+    char *alphabet = "abcdefghijklmnopqrstuvwxyz";
+    int i;
+
+    // Generate a random length for the name. I subtract 1 from the
+    // MAX_SAFE_NAME_LENGTH in order to allow for the name's leading "/".
+    length = random_in_range(6, MAX_SAFE_NAME_LENGTH - 1);
+
+    name[0] = '/';
+    name[length] = '\0';
+    i = length;
+    while (--i)
+        name[i] = alphabet[random_in_range(0, 25)];
+
+    return length;
+}
+
+
+static int
+convert_name_param(PyObject *py_name_param, void *checked_name) {
+    /* Verifies that the py_name_param is either None or a string.
+    If it's a string, checked_name->name points to a PyMalloc-ed buffer
+    holding a NULL-terminated C version of the string when this function
+    concludes. The caller is responsible for releasing the buffer.
+    */
+    int rc = 0;
+    NoneableName *p_name = (NoneableName *)checked_name;
+    PyObject *py_name_as_bytes = NULL;
+    char *p_name_as_c_string = NULL;
+
+    DPRINTF("inside convert_name_param\n");
+    DPRINTF("PyBytes_Check() = %d \n", PyBytes_Check(py_name_param));
+    DPRINTF("PyString_Check() = %d \n", PyString_Check(py_name_param));
+    DPRINTF("PyUnicode_Check() = %d \n", PyUnicode_Check(py_name_param));
+
+    p_name->is_none = 0;
+
+    // The name can be None or a Python string
+    if (py_name_param == Py_None) {
+        DPRINTF("name is None\n");
+        rc = 1;
+        p_name->is_none = 1;
+    }
+    else if (PyUnicode_Check(py_name_param) || PyBytes_Check(py_name_param)) {
+        DPRINTF("name is Unicode or bytes\n");
+        // The caller passed me a Unicode string or a byte array; I need a
+        // char *. Getting from one to the other takes a couple steps.
+
+        if (PyUnicode_Check(py_name_param)) {
+            DPRINTF("name is Unicode\n");
+            // PyUnicode_FSConverter() converts the Unicode object into a
+            // bytes or a bytearray object. (Why can't it be one or the other?)
+            PyUnicode_FSConverter(py_name_param, &py_name_as_bytes);
+        }
+        else {
+            DPRINTF("name is bytes\n");
+            // Make a copy of the name param.
+            py_name_as_bytes = PyBytes_FromObject(py_name_param);
+        }
+
+        // bytes_to_c_string() returns a pointer to the buffer.
+        p_name_as_c_string = bytes_to_c_string(py_name_as_bytes, 0);
+
+        // PyMalloc memory and copy the user-supplied name to it.
+        p_name->name = (char *)PyMem_Malloc(strlen(p_name_as_c_string) + 1);
+        if (p_name->name) {
+            rc = 1;
+            strcpy(p_name->name, p_name_as_c_string);
+        }
+        else
+            PyErr_SetString(PyExc_MemoryError, "Out of memory");
+
+        // The bytes version of the name isn't useful to me, and per the
+        // documentation for PyUnicode_FSConverter(), I am responsible for
+        // releasing it when I'm done.
+        release_bytes(py_name_as_bytes);
+    }
+    else
+        PyErr_SetString(PyExc_TypeError, "Name must be None or a string");
+
+    return rc;
+}
+
+
+
+/*   =====  Begin Shared Memory implementation functions ===== */
+
+static PyObject *
+shm_str(SharedMemory *self) {
+    return PyUnicode_FromString(self->name ? self->name : "(no name)");
+}
+
+static PyObject *
+shm_repr(SharedMemory *self) {
+    char mode[32];
+
+    sprintf(mode, "0%o", (int)(self->mode));
+
+    return PyUnicode_FromFormat("_posixshmem.SharedMemory(\"%s\", mode=%s)",
+                                self->name, mode);
+}
+
+static PyObject *
+my_shm_unlink(const char *name) {
+    DPRINTF("unlinking shm name %s\n", name);
+    if (-1 == shm_unlink(name)) {
+        switch (errno) {
+            case EACCES:
+                PyErr_SetString(pPermissionsException, "Permission denied");
+            break;
+
+            case ENOENT:
+                PyErr_SetString(pExistentialException,
+                    "No shared memory exists with the specified name");
+            break;
+
+            case ENAMETOOLONG:
+                PyErr_SetString(PyExc_ValueError, "The name is too long");
+            break;
+
+            default:
+                PyErr_SetFromErrno(PyExc_OSError);
+            break;
+        }
+
+        goto error_return;
+    }
+
+    Py_RETURN_NONE;
+
+    error_return:
+    return NULL;
+}
+
+
+static PyObject *
+SharedMemory_new(PyTypeObject *type, PyObject *args, PyObject *kwlist) {
+    SharedMemory *self;
+
+    self = (SharedMemory *)type->tp_alloc(type, 0);
+
+    return (PyObject *)self;
+}
+
+
+static int
+SharedMemory_init(SharedMemory *self, PyObject *args, PyObject *keywords) {
+    NoneableName name;
+    char temp_name[MAX_SAFE_NAME_LENGTH + 1];
+    unsigned int flags = 0;
+    unsigned long size = 0;
+    int read_only = 0;
+    static char *keyword_list[ ] = {"name", "flags", "mode", "size", "read_only", NULL};
+
+    // First things first -- initialize the self struct.
+    self->name = NULL;
+    self->fd = 0;
+    self->mode = 0600;
+
+    if (!PyArg_ParseTupleAndKeywords(args, keywords, "O&|Iiki", keyword_list,
+                                    &convert_name_param, &name, &flags,
+                                    &(self->mode), &size, &read_only))
+        goto error_return;
+
+    if ( !(flags & O_CREAT) && (flags & O_EXCL) ) {
+        PyErr_SetString(PyExc_ValueError,
+                "O_EXCL must be combined with O_CREAT");
+        goto error_return;
+    }
+
+    if (name.is_none && ((flags & O_EXCL) != O_EXCL)) {
+        PyErr_SetString(PyExc_ValueError,
+                "Name can only be None if O_EXCL is set");
+        goto error_return;
+    }
+
+    flags |= (read_only ? O_RDONLY : O_RDWR);
+
+    if (name.is_none) {
+        // (name == None) ==> generate a name for the caller
+        do {
+            errno = 0;
+            create_random_name(temp_name);
+
+            DPRINTF("calling shm_open, name=%s, flags=0x%x, mode=0%o\n",
+                        temp_name, flags, (int)self->mode);
+            self->fd = shm_open(temp_name, flags, (mode_t)self->mode);
+
+        } while ( (-1 == self->fd) && (EEXIST == errno) );
+
+        // PyMalloc memory and copy the randomly-generated name to it.
+        self->name = (char *)PyMem_Malloc(strlen(temp_name) + 1);
+        if (self->name)
+            strcpy(self->name, temp_name);
+        else {
+            PyErr_SetString(PyExc_MemoryError, "Out of memory");
+            goto error_return;
+        }
+    }
+    else {
+        // (name != None) ==> use name supplied by the caller. It was
+        // already converted to C by convert_name_param().
+        self->name = name.name;
+
+        DPRINTF("calling shm_open, name=%s, flags=0x%x, mode=0%o\n",
+                    self->name, flags, (int)self->mode);
+        self->fd = shm_open(self->name, flags, (mode_t)self->mode);
+    }
+
+    DPRINTF("shm fd = %d\n", self->fd);
+
+    if (-1 == self->fd) {
+        self->fd = 0;
+        switch (errno) {
+            case EACCES:
+                PyErr_Format(pPermissionsException,
+                                "No permission to %s this segment",
+                                (flags & O_TRUNC) ? "truncate" : "access"
+                                );
+            break;
+
+            case EEXIST:
+                PyErr_SetString(pExistentialException,
+                                "Shared memory with the specified name already exists");
+            break;
+
+            case ENOENT:
+                PyErr_SetString(pExistentialException,
+                                "No shared memory exists with the specified name");
+            break;
+
+            case EINVAL:
+                PyErr_SetString(PyExc_ValueError, "Invalid parameter(s)");
+            break;
+
+            case EMFILE:
+                PyErr_SetString(PyExc_OSError,
+                                 "This process already has the maximum number of files open");
+            break;
+
+            case ENFILE:
+                PyErr_SetString(PyExc_OSError,
+                                 "The system limit on the total number of open files has been reached");
+            break;
+
+            case ENAMETOOLONG:
+                PyErr_SetString(PyExc_ValueError,
+                                 "The name is too long");
+            break;
+
+            default:
+                PyErr_SetFromErrno(PyExc_OSError);
+            break;
+        }
+
+        goto error_return;
+    }
+    else {
+        if (size) {
+            DPRINTF("calling ftruncate, fd = %d, size = %ld\n", self->fd, size);
+            if (-1 == ftruncate(self->fd, (off_t)size)) {
+                // The code below will raise a Python error. Since that error
+                // is raised during __init__(), it will look to the caller
+                // as if object creation failed entirely. Here I clean up
+                // the system object I just created.
+                close(self->fd);
+                shm_unlink(self->name);
+
+                // ftruncate can return a ton of different errors, but most
+                // are not relevant or are extremely unlikely.
+                switch (errno) {
+                    case EINVAL:
+                        PyErr_SetString(PyExc_ValueError,
+                                        "The size is invalid or the memory is read-only");
+                    break;
+
+                    case EFBIG:
+                        PyErr_SetString(PyExc_ValueError,
+                                        "The size is too large");
+                    break;
+
+                    case EROFS:
+                    case EACCES:
+                        PyErr_SetString(pPermissionsException,
+                                        "The memory is read-only");
+                    break;
+
+                    default:
+                        PyErr_SetFromErrno(PyExc_OSError);
+                    break;
+                }
+
+                goto error_return;
+            }
+        }
+    }
+
+    return 0;
+
+    error_return:
+    return -1;
+}
+
+
+static void SharedMemory_dealloc(SharedMemory *self) {
+    DPRINTF("dealloc\n");
+    PyMem_Free(self->name);
+    self->name = NULL;
+
+    Py_TYPE(self)->tp_free((PyObject*)self);
+}
+
+
+PyObject *
+SharedMemory_getsize(SharedMemory *self, void *closure) {
+    struct stat fileinfo;
+    off_t size = -1;
+
+    if (0 == fstat(self->fd, &fileinfo))
+        size = fileinfo.st_size;
+    else {
+        switch (errno) {
+            case EBADF:
+            case EINVAL:
+                PyErr_SetString(pExistentialException,
+                                "The segment does not exist");
+            break;
+
+            default:
+                PyErr_SetFromErrno(PyExc_OSError);
+            break;
+        }
+
+        goto error_return;
+    }
+
+    return Py_BuildValue("k", (unsigned long)size);
+
+    error_return:
+    return NULL;
+}
+
+
+PyObject *
+SharedMemory_close_fd(SharedMemory *self) {
+    if (self->fd) {
+        if (-1 == close(self->fd)) {
+            switch (errno) {
+                case EBADF:
+                    PyErr_SetString(PyExc_ValueError,
+                                    "The file descriptor is invalid");
+                break;
+
+                default:
+                    PyErr_SetFromErrno(PyExc_OSError);
+                break;
+            }
+
+            goto error_return;
+        }
+    }
+
+    Py_RETURN_NONE;
+
+    error_return:
+    return NULL;
+}
+
+
+PyObject *
+SharedMemory_unlink(SharedMemory *self) {
+    return my_shm_unlink(self->name);
+}
+
+
+/*   =====  End Shared Memory functions =====           */
+
+
+/*
+ *
+ * Shared memory meta stuff for describing myself to Python
+ *
+ */
+
+
+static PyMemberDef SharedMemory_members[] = {
+    {   "name",
+        T_STRING,
+        offsetof(SharedMemory, name),
+        READONLY,
+        "The name specified in the constructor"
+    },
+    {   "fd",
+        T_INT,
+        offsetof(SharedMemory, fd),
+        READONLY,
+        "Shared memory segment file descriptor"
+    },
+    {   "mode",
+        T_LONG,
+        offsetof(SharedMemory, mode),
+        READONLY,
+        "The mode specified in the constructor"
+    },
+    {NULL} /* Sentinel */
+};
+
+
+static PyMethodDef SharedMemory_methods[] = {
+    {   "close_fd",
+        (PyCFunction)SharedMemory_close_fd,
+        METH_NOARGS,
+        "Closes the file descriptor associated with the shared memory."
+    },
+    {   "unlink",
+        (PyCFunction)SharedMemory_unlink,
+        METH_NOARGS,
+        "Unlink (remove) the shared memory."
+    },
+    {NULL, NULL, 0, NULL} /* Sentinel */
+};
+
+
+static PyGetSetDef SharedMemory_getseters[] = {
+    // size is read-only
+    {   "size",
+        (getter)SharedMemory_getsize,
+        (setter)NULL,
+        "size",
+        NULL
+    },
+    {NULL} /* Sentinel */
+};
+
+
+static PyTypeObject SharedMemoryType = {
+    PyVarObject_HEAD_INIT(NULL, 0)
+    "_posixshmem._PosixSharedMemory",   // tp_name
+    sizeof(SharedMemory),               // tp_basicsize
+    0,                                  // tp_itemsize
+    (destructor) SharedMemory_dealloc,  // tp_dealloc
+    0,                                  // tp_print
+    0,                                  // tp_getattr
+    0,                                  // tp_setattr
+    0,                                  // tp_compare
+    (reprfunc) shm_repr,                // tp_repr
+    0,                                  // tp_as_number
+    0,                                  // tp_as_sequence
+    0,                                  // tp_as_mapping
+    0,                                  // tp_hash
+    0,                                  // tp_call
+    (reprfunc) shm_str,                 // tp_str
+    0,                                  // tp_getattro
+    0,                                  // tp_setattro
+    0,                                  // tp_as_buffer
+    Py_TPFLAGS_DEFAULT | Py_TPFLAGS_BASETYPE,
+                                        // tp_flags
+    "POSIX shared memory object",       // tp_doc
+    0,                                  // tp_traverse
+    0,                                  // tp_clear
+    0,                                  // tp_richcompare
+    0,                                  // tp_weaklistoffset
+    0,                                  // tp_iter
+    0,                                  // tp_iternext
+    SharedMemory_methods,               // tp_methods
+    SharedMemory_members,               // tp_members
+    SharedMemory_getseters,             // tp_getset
+    0,                                  // tp_base
+    0,                                  // tp_dict
+    0,                                  // tp_descr_get
+    0,                                  // tp_descr_set
+    0,                                  // tp_dictoffset
+    (initproc) SharedMemory_init,       // tp_init
+    0,                                  // tp_alloc
+    (newfunc) SharedMemory_new,         // tp_new
+    0,                                  // tp_free
+    0,                                  // tp_is_gc
+    0                                   // tp_bases
+};
+
+
+/*
+ *
+ * Module-level functions & meta stuff
+ *
+ */
+
+static PyObject *
+posixshmem_unlink_shared_memory(PyObject *self, PyObject *args) {
+    const char *name;
+
+    if (!PyArg_ParseTuple(args, "s", &name))
+        return NULL;
+    else
+        return my_shm_unlink(name);
+}
+
+
+static PyMethodDef module_methods[ ] = {
+    {   "unlink_shared_memory",
+        (PyCFunction)posixshmem_unlink_shared_memory,
+        METH_VARARGS,
+        "Unlink shared memory"
+    },
+    {NULL} /* Sentinel */
+};
+
+
+static struct PyModuleDef this_module = {
+    PyModuleDef_HEAD_INIT,  // m_base
+    "_posixshmem",          // m_name
+    "POSIX shared memory module",     // m_doc
+    -1,                     // m_size (space allocated for module globals)
+    module_methods,         // m_methods
+    NULL,                   // m_reload
+    NULL,                   // m_traverse
+    NULL,                   // m_clear
+    NULL                    // m_free
+};
+
+/* Module init function */
+PyMODINIT_FUNC
+PyInit__posixshmem(void) {
+    PyObject *module;
+    PyObject *module_dict;
+
+    // I call this in case I'm asked to create any random names.
+    srand((unsigned int)time(NULL));
+
+    module = PyModule_Create(&this_module);
+
+    if (!module)
+        goto error_return;
+
+    if (PyType_Ready(&SharedMemoryType) < 0)
+        goto error_return;
+
+    Py_INCREF(&SharedMemoryType);
+    PyModule_AddObject(module, "_PosixSharedMemory", (PyObject *)&SharedMemoryType);
+
+
+    PyModule_AddStringConstant(module, "__copyright__", "Copyright 2012 Philip Semanchuk, 2018-2019 Davin Potts");
+
+    PyModule_AddIntConstant(module, "O_CREAT", O_CREAT);
+    PyModule_AddIntConstant(module, "O_EXCL", O_EXCL);
+    PyModule_AddIntConstant(module, "O_CREX", O_CREAT | O_EXCL);
+    PyModule_AddIntConstant(module, "O_TRUNC", O_TRUNC);
+
+    if (!(module_dict = PyModule_GetDict(module)))
+        goto error_return;
+
+    // Exceptions
+    if (!(pBaseException = PyErr_NewException("_posixshmem.Error", NULL, NULL)))
+        goto error_return;
+    else
+        PyDict_SetItemString(module_dict, "Error", pBaseException);
+
+    if (!(pPermissionsException = PyErr_NewException("_posixshmem.PermissionsError", pBaseException, NULL)))
+        goto error_return;
+    else
+        PyDict_SetItemString(module_dict, "PermissionsError", pPermissionsException);
+
+    if (!(pExistentialException = PyErr_NewException("_posixshmem.ExistentialError", pBaseException, NULL)))
+        goto error_return;
+    else
+        PyDict_SetItemString(module_dict, "ExistentialError", pExistentialException);
+
+    return module;
+
+    error_return:
+    return NULL;
+}
diff --git a/setup.py b/setup.py
index 44a563b..d54bbe0 100644
--- a/setup.py
+++ b/setup.py
@@ -1592,6 +1592,17 @@ class PyBuildExt(build_ext):
             if (sysconfig.get_config_var('HAVE_SEM_OPEN') and not
                 sysconfig.get_config_var('POSIX_SEMAPHORES_NOT_ENABLED')):
                 multiprocessing_srcs.append('_multiprocessing/semaphore.c')
+            if (self.compiler.find_library_file(lib_dirs, 'rt') or
+                host_platform != 'cygwin'):
+                posixshmem_srcs = [ '_multiprocessing/posixshmem.c',
+                                  ]
+                libs = []
+                if self.compiler.find_library_file(lib_dirs, 'rt'):
+                    libs.append('rt')
+                exts.append( Extension('_posixshmem', posixshmem_srcs,
+                                       define_macros={},
+                                       libraries=libs,
+                                       include_dirs=["Modules/_multiprocessing"]))
 
         exts.append ( Extension('_multiprocessing', multiprocessing_srcs,
                                 define_macros=list(macros.items()),