@rowstile/cli-win32-x64 0.1.0-alpha.2

package/python/Lib/multiprocessing/forkserver.py

@@ -0,0 +1,373 @@
+import atexit
+import errno
+import os
+import selectors
+import signal
+import socket
+import struct
+import sys
+import threading
+import warnings
+
+from . import connection
+from . import process
+from .context import reduction
+from . import resource_tracker
+from . import spawn
+from . import util
+
+__all__ = ['ensure_running', 'get_inherited_fds', 'connect_to_new_process',
+           'set_forkserver_preload']
+
+#
+#
+#
+
+MAXFDS_TO_SEND = 256
+SIGNED_STRUCT = struct.Struct('q')     # large enough for pid_t
+
+#
+# Forkserver class
+#
+
+class ForkServer(object):
+
+    def __init__(self):
+        self._forkserver_address = None
+        self._forkserver_alive_fd = None
+        self._forkserver_pid = None
+        self._inherited_fds = None
+        self._lock = threading.Lock()
+        self._preload_modules = ['__main__']
+
+    def _stop(self):
+        # Method used by unit tests to stop the server
+        with self._lock:
+            self._stop_unlocked()
+
+    def _stop_unlocked(self):
+        if self._forkserver_pid is None:
+            return
+
+        # close the "alive" file descriptor asks the server to stop
+        os.close(self._forkserver_alive_fd)
+        self._forkserver_alive_fd = None
+
+        os.waitpid(self._forkserver_pid, 0)
+        self._forkserver_pid = None
+
+        if not util.is_abstract_socket_namespace(self._forkserver_address):
+            os.unlink(self._forkserver_address)
+        self._forkserver_address = None
+
+    def set_forkserver_preload(self, modules_names):
+        '''Set list of module names to try to load in forkserver process.'''
+        if not all(type(mod) is str for mod in modules_names):
+            raise TypeError('module_names must be a list of strings')
+        self._preload_modules = modules_names
+
+    def get_inherited_fds(self):
+        '''Return list of fds inherited from parent process.
+
+        This returns None if the current process was not started by fork
+        server.
+        '''
+        return self._inherited_fds
+
+    def connect_to_new_process(self, fds):
+        '''Request forkserver to create a child process.
+
+        Returns a pair of fds (status_r, data_w).  The calling process can read
+        the child process's pid and (eventually) its returncode from status_r.
+        The calling process should write to data_w the pickled preparation and
+        process data.
+        '''
+        self.ensure_running()
+        if len(fds) + 4 >= MAXFDS_TO_SEND:
+            raise ValueError('too many fds')
+        with socket.socket(socket.AF_UNIX) as client:
+            client.connect(self._forkserver_address)
+            parent_r, child_w = os.pipe()
+            child_r, parent_w = os.pipe()
+            allfds = [child_r, child_w, self._forkserver_alive_fd,
+                      resource_tracker.getfd()]
+            allfds += fds
+            try:
+                reduction.sendfds(client, allfds)
+                return parent_r, parent_w
+            except:
+                os.close(parent_r)
+                os.close(parent_w)
+                raise
+            finally:
+                os.close(child_r)
+                os.close(child_w)
+
+    def ensure_running(self):
+        '''Make sure that a fork server is running.
+
+        This can be called from any process.  Note that usually a child
+        process will just reuse the forkserver started by its parent, so
+        ensure_running() will do nothing.
+        '''
+        with self._lock:
+            resource_tracker.ensure_running()
+            if self._forkserver_pid is not None:
+                # forkserver was launched before, is it still running?
+                pid, status = os.waitpid(self._forkserver_pid, os.WNOHANG)
+                if not pid:
+                    # still alive
+                    return
+                # dead, launch it again
+                os.close(self._forkserver_alive_fd)
+                self._forkserver_address = None
+                self._forkserver_alive_fd = None
+                self._forkserver_pid = None
+
+            # gh-144503: sys_argv is passed as real argv elements after the
+            # ``-c cmd`` rather than repr'd into main_kws so that a large
+            # parent sys.argv cannot push the single ``-c`` command string
+            # over the OS per-argument length limit (MAX_ARG_STRLEN on Linux).
+            # The child sees them as sys.argv[1:].
+            cmd = ('import sys; '
+                   'from multiprocessing.forkserver import main; '
+                   'main(%d, %d, %r, sys_argv=sys.argv[1:], **%r)')
+
+            main_kws = {}
+            sys_argv = None
+            if self._preload_modules:
+                data = spawn.get_preparation_data('ignore')
+                if 'sys_path' in data:
+                    main_kws['sys_path'] = data['sys_path']
+                if 'init_main_from_path' in data:
+                    main_kws['main_path'] = data['init_main_from_path']
+                if 'sys_argv' in data:
+                    sys_argv = data['sys_argv']
+
+            with socket.socket(socket.AF_UNIX) as listener:
+                address = connection.arbitrary_address('AF_UNIX')
+                listener.bind(address)
+                if not util.is_abstract_socket_namespace(address):
+                    os.chmod(address, 0o600)
+                listener.listen()
+
+                # all client processes own the write end of the "alive" pipe;
+                # when they all terminate the read end becomes ready.
+                alive_r, alive_w = os.pipe()
+                try:
+                    fds_to_pass = [listener.fileno(), alive_r]
+                    cmd %= (listener.fileno(), alive_r, self._preload_modules,
+                            main_kws)
+                    exe = spawn.get_executable()
+                    args = [exe] + util._args_from_interpreter_flags()
+                    args += ['-c', cmd]
+                    if sys_argv is not None:
+                        args += sys_argv
+                    pid = util.spawnv_passfds(exe, args, fds_to_pass)
+                except:
+                    os.close(alive_w)
+                    raise
+                finally:
+                    os.close(alive_r)
+                self._forkserver_address = address
+                self._forkserver_alive_fd = alive_w
+                self._forkserver_pid = pid
+
+#
+#
+#
+
+def main(listener_fd, alive_r, preload, main_path=None, sys_path=None,
+         *, sys_argv=None):
+    '''Run forkserver.'''
+    if preload:
+        if sys_argv is not None:
+            sys.argv[:] = sys_argv
+        if sys_path is not None:
+            sys.path[:] = sys_path
+        if '__main__' in preload and main_path is not None:
+            process.current_process()._inheriting = True
+            try:
+                spawn.import_main_path(main_path)
+            finally:
+                del process.current_process()._inheriting
+        for modname in preload:
+            try:
+                __import__(modname)
+            except ImportError:
+                pass
+
+        # gh-135335: flush stdout/stderr in case any of the preloaded modules
+        # wrote to them, otherwise children might inherit buffered data
+        util._flush_std_streams()
+
+    util._close_stdin()
+
+    sig_r, sig_w = os.pipe()
+    os.set_blocking(sig_r, False)
+    os.set_blocking(sig_w, False)
+
+    def sigchld_handler(*_unused):
+        # Dummy signal handler, doesn't do anything
+        pass
+
+    handlers = {
+        # unblocking SIGCHLD allows the wakeup fd to notify our event loop
+        signal.SIGCHLD: sigchld_handler,
+        # protect the process from ^C
+        signal.SIGINT: signal.SIG_IGN,
+        }
+    old_handlers = {sig: signal.signal(sig, val)
+                    for (sig, val) in handlers.items()}
+
+    # calling os.write() in the Python signal handler is racy
+    signal.set_wakeup_fd(sig_w)
+
+    # map child pids to client fds
+    pid_to_fd = {}
+
+    with socket.socket(socket.AF_UNIX, fileno=listener_fd) as listener, \
+         selectors.DefaultSelector() as selector:
+        _forkserver._forkserver_address = listener.getsockname()
+
+        selector.register(listener, selectors.EVENT_READ)
+        selector.register(alive_r, selectors.EVENT_READ)
+        selector.register(sig_r, selectors.EVENT_READ)
+
+        while True:
+            try:
+                while True:
+                    rfds = [key.fileobj for (key, events) in selector.select()]
+                    if rfds:
+                        break
+
+                if alive_r in rfds:
+                    # EOF because no more client processes left
+                    assert os.read(alive_r, 1) == b'', "Not at EOF?"
+                    raise SystemExit
+
+                if sig_r in rfds:
+                    # Got SIGCHLD
+                    os.read(sig_r, 65536)  # exhaust
+                    while True:
+                        # Scan for child processes
+                        try:
+                            pid, sts = os.waitpid(-1, os.WNOHANG)
+                        except ChildProcessError:
+                            break
+                        if pid == 0:
+                            break
+                        child_w = pid_to_fd.pop(pid, None)
+                        if child_w is not None:
+                            returncode = os.waitstatus_to_exitcode(sts)
+
+                            # Send exit code to client process
+                            try:
+                                write_signed(child_w, returncode)
+                            except BrokenPipeError:
+                                # client vanished
+                                pass
+                            os.close(child_w)
+                        else:
+                            # This shouldn't happen really
+                            warnings.warn('forkserver: waitpid returned '
+                                          'unexpected pid %d' % pid)
+
+                if listener in rfds:
+                    # Incoming fork request
+                    with listener.accept()[0] as s:
+                        # Receive fds from client
+                        fds = reduction.recvfds(s, MAXFDS_TO_SEND + 1)
+                        if len(fds) > MAXFDS_TO_SEND:
+                            raise RuntimeError(
+                                "Too many ({0:n}) fds to send".format(
+                                    len(fds)))
+                        child_r, child_w, *fds = fds
+                        s.close()
+                        pid = os.fork()
+                        if pid == 0:
+                            # Child
+                            code = 1
+                            try:
+                                listener.close()
+                                selector.close()
+                                unused_fds = [alive_r, child_w, sig_r, sig_w]
+                                unused_fds.extend(pid_to_fd.values())
+                                atexit._clear()
+                                atexit.register(util._exit_function)
+                                code = _serve_one(child_r, fds,
+                                                  unused_fds,
+                                                  old_handlers)
+                            except Exception:
+                                sys.excepthook(*sys.exc_info())
+                                sys.stderr.flush()
+                            finally:
+                                atexit._run_exitfuncs()
+                                os._exit(code)
+                        else:
+                            # Send pid to client process
+                            try:
+                                write_signed(child_w, pid)
+                            except BrokenPipeError:
+                                # client vanished
+                                pass
+                            pid_to_fd[pid] = child_w
+                            os.close(child_r)
+                            for fd in fds:
+                                os.close(fd)
+
+            except OSError as e:
+                if e.errno != errno.ECONNABORTED:
+                    raise
+
+
+def _serve_one(child_r, fds, unused_fds, handlers):
+    # close unnecessary stuff and reset signal handlers
+    signal.set_wakeup_fd(-1)
+    for sig, val in handlers.items():
+        signal.signal(sig, val)
+    for fd in unused_fds:
+        os.close(fd)
+
+    (_forkserver._forkserver_alive_fd,
+     resource_tracker._resource_tracker._fd,
+     *_forkserver._inherited_fds) = fds
+
+    # Run process object received over pipe
+    parent_sentinel = os.dup(child_r)
+    code = spawn._main(child_r, parent_sentinel)
+
+    return code
+
+
+#
+# Read and write signed numbers
+#
+
+def read_signed(fd):
+    data = b''
+    length = SIGNED_STRUCT.size
+    while len(data) < length:
+        s = os.read(fd, length - len(data))
+        if not s:
+            raise EOFError('unexpected EOF')
+        data += s
+    return SIGNED_STRUCT.unpack(data)[0]
+
+def write_signed(fd, n):
+    msg = SIGNED_STRUCT.pack(n)
+    while msg:
+        nbytes = os.write(fd, msg)
+        if nbytes == 0:
+            raise RuntimeError('should not get here')
+        msg = msg[nbytes:]
+
+#
+#
+#
+
+_forkserver = ForkServer()
+ensure_running = _forkserver.ensure_running
+get_inherited_fds = _forkserver.get_inherited_fds
+connect_to_new_process = _forkserver.connect_to_new_process
+set_forkserver_preload = _forkserver.set_forkserver_preload

package/python/Lib/multiprocessing/heap.py

@@ -0,0 +1,337 @@
+#
+# Module which supports allocation of memory from an mmap
+#
+# multiprocessing/heap.py
+#
+# Copyright (c) 2006-2008, R Oudkerk
+# Licensed to PSF under a Contributor Agreement.
+#
+
+import bisect
+from collections import defaultdict
+import mmap
+import os
+import sys
+import tempfile
+import threading
+
+from .context import reduction, assert_spawning
+from . import util
+
+__all__ = ['BufferWrapper']
+
+#
+# Inheritable class which wraps an mmap, and from which blocks can be allocated
+#
+
+if sys.platform == 'win32':
+
+    import _winapi
+
+    class Arena(object):
+        """
+        A shared memory area backed by anonymous memory (Windows).
+        """
+
+        _rand = tempfile._RandomNameSequence()
+
+        def __init__(self, size):
+            self.size = size
+            for i in range(100):
+                name = 'pym-%d-%s' % (os.getpid(), next(self._rand))
+                buf = mmap.mmap(-1, size, tagname=name)
+                if _winapi.GetLastError() == 0:
+                    break
+                # We have reopened a preexisting mmap.
+                buf.close()
+            else:
+                raise FileExistsError('Cannot find name for new mmap')
+            self.name = name
+            self.buffer = buf
+            self._state = (self.size, self.name)
+
+        def __getstate__(self):
+            assert_spawning(self)
+            return self._state
+
+        def __setstate__(self, state):
+            self.size, self.name = self._state = state
+            # Reopen existing mmap
+            self.buffer = mmap.mmap(-1, self.size, tagname=self.name)
+            # XXX Temporarily preventing buildbot failures while determining
+            # XXX the correct long-term fix. See issue 23060
+            #assert _winapi.GetLastError() == _winapi.ERROR_ALREADY_EXISTS
+
+else:
+
+    class Arena(object):
+        """
+        A shared memory area backed by a temporary file (POSIX).
+        """
+
+        if sys.platform == 'linux':
+            _dir_candidates = ['/dev/shm']
+        else:
+            _dir_candidates = []
+
+        def __init__(self, size, fd=-1):
+            self.size = size
+            self.fd = fd
+            if fd == -1:
+                # Arena is created anew (if fd != -1, it means we're coming
+                # from rebuild_arena() below)
+                self.fd, name = tempfile.mkstemp(
+                     prefix='pym-%d-'%os.getpid(),
+                     dir=self._choose_dir(size))
+                os.unlink(name)
+                util.Finalize(self, os.close, (self.fd,))
+                os.ftruncate(self.fd, size)
+            self.buffer = mmap.mmap(self.fd, self.size)
+
+        def _choose_dir(self, size):
+            # Choose a non-storage backed directory if possible,
+            # to improve performance
+            for d in self._dir_candidates:
+                st = os.statvfs(d)
+                if st.f_bavail * st.f_frsize >= size:  # enough free space?
+                    return d
+            return util.get_temp_dir()
+
+    def reduce_arena(a):
+        if a.fd == -1:
+            raise ValueError('Arena is unpicklable because '
+                             'forking was enabled when it was created')
+        return rebuild_arena, (a.size, reduction.DupFd(a.fd))
+
+    def rebuild_arena(size, dupfd):
+        return Arena(size, dupfd.detach())
+
+    reduction.register(Arena, reduce_arena)
+
+#
+# Class allowing allocation of chunks of memory from arenas
+#
+
+class Heap(object):
+
+    # Minimum malloc() alignment
+    _alignment = 8
+
+    _DISCARD_FREE_SPACE_LARGER_THAN = 4 * 1024 ** 2  # 4 MB
+    _DOUBLE_ARENA_SIZE_UNTIL = 4 * 1024 ** 2
+
+    def __init__(self, size=mmap.PAGESIZE):
+        self._lastpid = os.getpid()
+        self._lock = threading.Lock()
+        # Current arena allocation size
+        self._size = size
+        # A sorted list of available block sizes in arenas
+        self._lengths = []
+
+        # Free block management:
+        # - map each block size to a list of `(Arena, start, stop)` blocks
+        self._len_to_seq = {}
+        # - map `(Arena, start)` tuple to the `(Arena, start, stop)` block
+        #   starting at that offset
+        self._start_to_block = {}
+        # - map `(Arena, stop)` tuple to the `(Arena, start, stop)` block
+        #   ending at that offset
+        self._stop_to_block = {}
+
+        # Map arenas to their `(Arena, start, stop)` blocks in use
+        self._allocated_blocks = defaultdict(set)
+        self._arenas = []
+
+        # List of pending blocks to free - see comment in free() below
+        self._pending_free_blocks = []
+
+        # Statistics
+        self._n_mallocs = 0
+        self._n_frees = 0
+
+    @staticmethod
+    def _roundup(n, alignment):
+        # alignment must be a power of 2
+        mask = alignment - 1
+        return (n + mask) & ~mask
+
+    def _new_arena(self, size):
+        # Create a new arena with at least the given *size*
+        length = self._roundup(max(self._size, size), mmap.PAGESIZE)
+        # We carve larger and larger arenas, for efficiency, until we
+        # reach a large-ish size (roughly L3 cache-sized)
+        if self._size < self._DOUBLE_ARENA_SIZE_UNTIL:
+            self._size *= 2
+        util.info('allocating a new mmap of length %d', length)
+        arena = Arena(length)
+        self._arenas.append(arena)
+        return (arena, 0, length)
+
+    def _discard_arena(self, arena):
+        # Possibly delete the given (unused) arena
+        length = arena.size
+        # Reusing an existing arena is faster than creating a new one, so
+        # we only reclaim space if it's large enough.
+        if length < self._DISCARD_FREE_SPACE_LARGER_THAN:
+            return
+        blocks = self._allocated_blocks.pop(arena)
+        assert not blocks
+        del self._start_to_block[(arena, 0)]
+        del self._stop_to_block[(arena, length)]
+        self._arenas.remove(arena)
+        seq = self._len_to_seq[length]
+        seq.remove((arena, 0, length))
+        if not seq:
+            del self._len_to_seq[length]
+            self._lengths.remove(length)
+
+    def _malloc(self, size):
+        # returns a large enough block -- it might be much larger
+        i = bisect.bisect_left(self._lengths, size)
+        if i == len(self._lengths):
+            return self._new_arena(size)
+        else:
+            length = self._lengths[i]
+            seq = self._len_to_seq[length]
+            block = seq.pop()
+            if not seq:
+                del self._len_to_seq[length], self._lengths[i]
+
+        (arena, start, stop) = block
+        del self._start_to_block[(arena, start)]
+        del self._stop_to_block[(arena, stop)]
+        return block
+
+    def _add_free_block(self, block):
+        # make block available and try to merge with its neighbours in the arena
+        (arena, start, stop) = block
+
+        try:
+            prev_block = self._stop_to_block[(arena, start)]
+        except KeyError:
+            pass
+        else:
+            start, _ = self._absorb(prev_block)
+
+        try:
+            next_block = self._start_to_block[(arena, stop)]
+        except KeyError:
+            pass
+        else:
+            _, stop = self._absorb(next_block)
+
+        block = (arena, start, stop)
+        length = stop - start
+
+        try:
+            self._len_to_seq[length].append(block)
+        except KeyError:
+            self._len_to_seq[length] = [block]
+            bisect.insort(self._lengths, length)
+
+        self._start_to_block[(arena, start)] = block
+        self._stop_to_block[(arena, stop)] = block
+
+    def _absorb(self, block):
+        # deregister this block so it can be merged with a neighbour
+        (arena, start, stop) = block
+        del self._start_to_block[(arena, start)]
+        del self._stop_to_block[(arena, stop)]
+
+        length = stop - start
+        seq = self._len_to_seq[length]
+        seq.remove(block)
+        if not seq:
+            del self._len_to_seq[length]
+            self._lengths.remove(length)
+
+        return start, stop
+
+    def _remove_allocated_block(self, block):
+        arena, start, stop = block
+        blocks = self._allocated_blocks[arena]
+        blocks.remove((start, stop))
+        if not blocks:
+            # Arena is entirely free, discard it from this process
+            self._discard_arena(arena)
+
+    def _free_pending_blocks(self):
+        # Free all the blocks in the pending list - called with the lock held.
+        while True:
+            try:
+                block = self._pending_free_blocks.pop()
+            except IndexError:
+                break
+            self._add_free_block(block)
+            self._remove_allocated_block(block)
+
+    def free(self, block):
+        # free a block returned by malloc()
+        # Since free() can be called asynchronously by the GC, it could happen
+        # that it's called while self._lock is held: in that case,
+        # self._lock.acquire() would deadlock (issue #12352). To avoid that, a
+        # trylock is used instead, and if the lock can't be acquired
+        # immediately, the block is added to a list of blocks to be freed
+        # synchronously sometimes later from malloc() or free(), by calling
+        # _free_pending_blocks() (appending and retrieving from a list is not
+        # strictly thread-safe but under CPython it's atomic thanks to the GIL).
+        if os.getpid() != self._lastpid:
+            raise ValueError(
+                "My pid ({0:n}) is not last pid {1:n}".format(
+                    os.getpid(),self._lastpid))
+        if not self._lock.acquire(False):
+            # can't acquire the lock right now, add the block to the list of
+            # pending blocks to free
+            self._pending_free_blocks.append(block)
+        else:
+            # we hold the lock
+            try:
+                self._n_frees += 1
+                self._free_pending_blocks()
+                self._add_free_block(block)
+                self._remove_allocated_block(block)
+            finally:
+                self._lock.release()
+
+    def malloc(self, size):
+        # return a block of right size (possibly rounded up)
+        if size < 0:
+            raise ValueError("Size {0:n} out of range".format(size))
+        if sys.maxsize <= size:
+            raise OverflowError("Size {0:n} too large".format(size))
+        if os.getpid() != self._lastpid:
+            self.__init__()                     # reinitialize after fork
+        with self._lock:
+            self._n_mallocs += 1
+            # allow pending blocks to be marked available
+            self._free_pending_blocks()
+            size = self._roundup(max(size, 1), self._alignment)
+            (arena, start, stop) = self._malloc(size)
+            real_stop = start + size
+            if real_stop < stop:
+                # if the returned block is larger than necessary, mark
+                # the remainder available
+                self._add_free_block((arena, real_stop, stop))
+            self._allocated_blocks[arena].add((start, real_stop))
+            return (arena, start, real_stop)
+
+#
+# Class wrapping a block allocated out of a Heap -- can be inherited by child process
+#
+
+class BufferWrapper(object):
+
+    _heap = Heap()
+
+    def __init__(self, size):
+        if size < 0:
+            raise ValueError("Size {0:n} out of range".format(size))
+        if sys.maxsize <= size:
+            raise OverflowError("Size {0:n} too large".format(size))
+        block = BufferWrapper._heap.malloc(size)
+        self._state = (block, size)
+        util.Finalize(self, BufferWrapper._heap.free, args=(block,))
+
+    def create_memoryview(self):
+        (arena, start, stop), size = self._state
+        return memoryview(arena.buffer)[start:start+size]