Fixes issue #8052: The posix subprocess module's close_fds behavior was

suboptimal by closing all possible file descriptors rather than just
the open ones in the child process before exec().

It now closes only the open fds when it is possible to safely determine what
those are.

Lib/test/test_subprocess.py

@@ -1295,6 +1295,11 @@ class POSIXProcessTestCase(BaseTestCase):
         self.addCleanup(os.close, fds[1])
 
         open_fds = set(fds)
+        # add a bunch more fds
+        for _ in range(9):
+            fd = os.open("/dev/null", os.O_RDONLY)
+            self.addCleanup(os.close, fd)
+            open_fds.add(fd)
 
         p = subprocess.Popen([sys.executable, fd_status],
                              stdout=subprocess.PIPE, close_fds=False)
@@ -1313,6 +1318,19 @@ class POSIXProcessTestCase(BaseTestCase):
                          "Some fds were left open")
         self.assertIn(1, remaining_fds, "Subprocess failed")
 
+        # Keep some of the fd's we opened open in the subprocess.
+        # This tests _posixsubprocess.c's proper handling of fds_to_keep.
+        fds_to_keep = set(open_fds.pop() for _ in range(8))
+        p = subprocess.Popen([sys.executable, fd_status],
+                             stdout=subprocess.PIPE, close_fds=True,
+                             pass_fds=())
+        output, ignored = p.communicate()
+        remaining_fds = set(map(int, output.split(b',')))
+
+        self.assertFalse(remaining_fds & fds_to_keep & open_fds,
+                         "Some fds not in pass_fds were left open")
+        self.assertIn(1, remaining_fds, "Subprocess failed")
+
     # Mac OS X Tiger (10.4) has a kernel bug: sometimes, the file
     # descriptor of a pipe closed in the parent process is valid in the
     # child process according to fstat(), but the mode of the file