Make run_command() use subprocess.Popen() rather than os.popen().

This lets us capture the command's stderr as well as stdout. I've been
perplexed recently by errors from Mac SetFile commands that didn't report
stderr.

indra/lib/python/indra/util/llmanifest.py

@@ -39,6 +39,7 @@ import shutil
 import sys
 import tarfile
 import errno
+import subprocess
 
 def path_ancestors(path):
     drive, path = os.path.splitdrive(os.path.normpath(path))
@@ -366,20 +367,23 @@ class LLManifest(object):
 
     def run_command(self, command):
         """ Runs an external command, and returns the output.  Raises
-        an exception if the command reurns a nonzero status code.  For
-        debugging/informational purpoases, prints out the command's
+        an exception if the command returns a nonzero status code.  For
+        debugging/informational purposes, prints out the command's
         output as it is received."""
         print "Running command:", command
-        fd = os.popen(command, 'r')
+        sys.stdout.flush()
+        child = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.STDOUT,
+                                 shell=True)
         lines = []
         while True:
-            lines.append(fd.readline())
+            lines.append(child.stdout.readline())
             if lines[-1] == '':
                 break
             else:
                 print lines[-1],
         output = ''.join(lines)
-        status = fd.close()
+        child.stdout.close()
+        status = child.wait()
         if status:
             raise RuntimeError(
                 "Command %s returned non-zero status (%s) \noutput:\n%s"