Speedup Process.children() (#1185)

* update HISTORY

* update doc

* #1183: speedup Process.children() by 2.2x

* fix windows err

* #1083 / #1084: implement linux-specific ppid_map() function speending things up from 2x to 2.4x

* add ESRCH to err handling

* update doc

4 files changed, 72 insertions, 47 deletions

diff --git a/HISTORY.rst b/HISTORY.rst
index bfaa2568..8bc3f784 100644
--- a/HISTORY.rst
+++ b/HISTORY.rst
@@ -10,6 +10,7 @@
 - 1173_: introduced PSUTIL_DEBUG environment variable which can be set in order
   to print useful debug messages on stderr (useful in case of nasty errors).
 - 1177_: added support for sensors_battery() on OSX.  (patch by Arnon Yaari)
+- 1183_: Process.children() is 2x faster on UNIX and 2.4x faster on Linux.
 
 **Bug fixes**
 
@@ -17,8 +18,8 @@
 - 1169_: [Linux] users() "hostname" returns username instead.  (patch by
   janderbrain)
 - 1172_: [Windows] `make test` does not work.
-- 1179_: [Linux] Process.cmdline() correctly splits cmdline args for
-  misbehaving processes who overwrite /proc/pid/cmdline by using spaces
+- 1179_: [Linux] Process.cmdline() is now able to splits cmdline args for
+  misbehaving processes which overwrite /proc/pid/cmdline and use spaces
   instead of null bytes as args separator.
 - 1181_: [OSX] Process.memory_maps() may raise ENOENT.
 
diff --git a/docs/index.rst b/docs/index.rst
index d34a1c45..ea2384bb 100644
--- a/docs/index.rst
+++ b/docs/index.rst
@@ -1716,9 +1716,9 @@ Process class
 
   .. method:: children(recursive=False)
 
-    Return the children of this process as a list of :Class:`Process` objects,
-    preemptively checking whether PID has been reused. If recursive is `True`
-    return all the parent descendants.
+    Return the children of this process as a list of :class:`Process`
+    instances.
+    If recursive is `True` return all the parent descendants.
     Pseudo code example assuming *A == this process*:
     ::
 
@@ -1738,7 +1738,7 @@ Process class
     Note that in the example above if process X disappears process Y won't be
     returned either as the reference to process A is lost.
     This concept is well summaried by this
-    `unit test <https://github.com/giampaolo/psutil/blob/fb9ae861cf3cf175c3da4a3cd4e558c6cbd6af91/psutil/tests/test_process.py#L1236-L1247>`__.
+    `unit test <https://github.com/giampaolo/psutil/blob/65a52341b55faaab41f68ebc4ed31f18f0929754/psutil/tests/test_process.py#L1064-L1075>`__.
     See also how to `kill a process tree <#kill-process-tree>`__ and
     `terminate my children <#terminate-my-children>`__.
 
diff --git a/psutil/__init__.py b/psutil/__init__.py
index c8da0a3d..a8447973 100644
--- a/psutil/__init__.py
+++ b/psutil/__init__.py
@@ -250,10 +250,31 @@ if (int(__version__.replace('.', '')) !=
 
 
 # =====================================================================
-# --- Process class
+# --- Utils
 # =====================================================================
 
 
+if hasattr(_psplatform, 'ppid_map'):
+    # Faster version (Windows and Linux).
+    _ppid_map = _psplatform.ppid_map
+else:
+    def _ppid_map():
+        """Return a {pid: ppid, ...} dict for all running processes in
+        one shot. Used to speed up Process.children().
+        """
+        ret = {}
+        for pid in pids():
+            try:
+                proc = _psplatform.Process(pid)
+                ppid = proc.ppid()
+            except (NoSuchProcess, AccessDenied):
+                # Note: AccessDenied is unlikely to happen.
+                pass
+            else:
+                ret[pid] = ppid
+        return ret
+
+
 def _assert_pid_not_reused(fun):
     """Decorator which raises NoSuchProcess in case a process is no
     longer running or its PID has been reused.
@@ -266,6 +287,11 @@ def _assert_pid_not_reused(fun):
     return wrapper
 
 
+# =====================================================================
+# --- Process class
+# =====================================================================
+
+
 class Process(object):
     """Represents an OS process with the given PID.
     If PID is omitted current process PID (os.getpid()) is used.
@@ -848,55 +874,29 @@ class Process(object):
         process Y won't be listed as the reference to process A
         is lost.
         """
-        if hasattr(_psplatform, 'ppid_map'):
-            # Windows only: obtain a {pid:ppid, ...} dict for all running
-            # processes in one shot (faster).
-            ppid_map = _psplatform.ppid_map()
-        else:
-            ppid_map = None
-
+        ppid_map = _ppid_map()
         ret = []
         if not recursive:
-            if ppid_map is None:
-                # 'slow' version, common to all platforms except Windows
-                for p in process_iter():
+            for pid, ppid in ppid_map.items():
+                if ppid == self.pid:
                     try:
-                        if p.ppid() == self.pid:
-                            # if child happens to be older than its parent
-                            # (self) it means child's PID has been reused
-                            if self.create_time() <= p.create_time():
-                                ret.append(p)
+                        child = Process(pid)
+                        # if child happens to be older than its parent
+                        # (self) it means child's PID has been reused
+                        if self.create_time() <= child.create_time():
+                            ret.append(child)
                     except (NoSuchProcess, ZombieProcess):
                         pass
-            else:  # pragma: no cover
-                # Windows only (faster)
-                for pid, ppid in ppid_map.items():
-                    if ppid == self.pid:
-                        try:
-                            child = Process(pid)
-                            # if child happens to be older than its parent
-                            # (self) it means child's PID has been reused
-                            if self.create_time() <= child.create_time():
-                                ret.append(child)
-                        except (NoSuchProcess, ZombieProcess):
-                            pass
         else:
             # construct a dict where 'values' are all the processes
             # having 'key' as their parent
             table = collections.defaultdict(list)
-            if ppid_map is None:
-                for p in process_iter():
-                    try:
-                        table[p.ppid()].append(p)
-                    except (NoSuchProcess, ZombieProcess):
-                        pass
-            else:  # pragma: no cover
-                for pid, ppid in ppid_map.items():
-                    try:
-                        p = Process(pid)
-                        table[ppid].append(p)
-                    except (NoSuchProcess, ZombieProcess):
-                        pass
+            for pid, ppid in ppid_map.items():
+                try:
+                    p = Process(pid)
+                    table[ppid].append(p)
+                except (NoSuchProcess, ZombieProcess):
+                    pass
             # At this point we have a mapping table where table[self.pid]
             # are the current process' children.
             # Below, we look for all descendants recursively, similarly
diff --git a/psutil/_pslinux.py b/psutil/_pslinux.py
index a5a3fd89..b57adb34 100644
--- a/psutil/_pslinux.py
+++ b/psutil/_pslinux.py
@@ -1356,6 +1356,30 @@ def pid_exists(pid):
             return pid in pids()
 
 
+def ppid_map():
+    """Obtain a {pid: ppid, ...} dict for all running processes in
+    one shot. Used to speed up Process.children().
+    """
+    ret = {}
+    procfs_path = get_procfs_path()
+    for pid in pids():
+        try:
+            with open_binary("%s/%s/stat" % (procfs_path, pid)) as f:
+                data = f.read()
+        except EnvironmentError as err:
+            # Note: we should be able to access /stat for all processes
+            # so we won't bump into EPERM, which is good.
+            if err.errno not in (errno.ENOENT, errno.ESRCH,
+                                 errno.EPERM, errno.EACCES):
+                raise
+        else:
+            rpar = data.rfind(b')')
+            dset = data[rpar + 2:].split()
+            ppid = int(dset[1])
+            ret[pid] = ppid
+    return ret
+
+
 def wrap_exceptions(fun):
     """Decorator which translates bare OSError and IOError exceptions
     into NoSuchProcess and AccessDenied.