Yet another fix for concurrent calls to expand.exe.

build/extract_from_cab.py

 #!/usr/bin/env python
 # Copyright (c) 2011 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 """Extracts a single file from a CAB archive."""
 
 import os
 import subprocess
 import sys
 import tempfile
 
-lock_file = os.path.join(tempfile.gettempdir(), 'expand.lock')
-
-
-def acquire_lock():
-  while True:
-    try:
-      fd = os.open(lock_file, os.O_CREAT | os.O_EXCL | os.O_RDWR)
-      return fd
-    except OSError as e:
-      if e.errno != errno.EEXIST:
-        raise
-      print 'Cab extraction could not get exclusive lock. Retrying in 100ms...'
-      time.sleep(0.1)
-
-
-def release_lock(fd):
-  os.close(fd)
-  os.unlink(lock_file)
-
 
 def main():
   if len(sys.argv) != 4:
     print 'Usage: extract_from_cab.py cab_path archived_file output_dir'
     return 1
 
   [cab_path, archived_file, output_dir] = sys.argv[1:]
 
-  lock_fd = acquire_lock()
+  # Expand.exe does its work in a fixed-named temporary directory created within
+  # the given output directory. This is a problem for concurrent extractions, so
+  # create a unique temp dir within the desired output directory to work around
+  # this limitation.
+  temp_dir = tempfile.mkdtemp(dir=output_dir)
+
   try:
+    # Setup the environment in case other versions of expand.exe do their work
+    # in TEMP/TMP instead of or in addition to the output directory.
+    subprocess_env = os.environ.copy()
+    subprocess_env['TEMP'] = temp_dir

[M-A Ruel, 2011/12/13 19:35:54]

Can you pinpoint which of the two it's using and only setting this one? I'd bet
on TEMP.

[grt (UTC plus 2), 2011/12/13 19:47:24]

The Win7 version I tested uses neither; it uses the output directory given on
the command line.  I thought it would be more robust to set these as well since
other versions could conceivably use TEMP/TMP.  I'm comfortable removing them if
they bring you displeasure.

[M-A Ruel, 2011/12/13 19:51:35]

Well, if the code has no effect, I'd rather remove it. At worst, you could RDP
into a XP slave with procmon just to verify.

[grt (UTC plus 2), 2011/12/13 20:06:44]

Looks like TEMP/TMP aren't used in XP, so I've removed this.

+    subprocess_env['TMP'] = temp_dir
     # Invoke the Windows expand utility to extract the file.
     level = subprocess.call(
-        ['expand', cab_path, '-F:' + archived_file, output_dir])
-    if level != 0:
-      return level
+        ['expand', cab_path, '-F:' + archived_file, temp_dir],
+        env=subprocess_env)
+    if level == 0:
+      # Move the output file into place, preserving expand.exe's behavior of
+      # paving over any preexisting file.
+      output_file = os.path.join(output_dir, archived_file)
+      try:
+        os.remove(output_file)
+      except OSError:
+        pass
+      os.rename(os.path.join(temp_dir, archived_file), output_file)
   finally:
-    release_lock(lock_fd)
+    try:
+      os.rmdir(temp_dir)

[M-A Ruel, 2011/12/13 19:35:54]

Yes, shutil.rmtree(temp_dir, True) is better.

[grt (UTC plus 2), 2011/12/13 19:47:24]

Done.

+    except OSError:
+      print 'Warning: unable to delete temporary directory ' + temp_dir
+
+  if level != 0:
+    return level
 
   # The expand utility preserves the modification date and time of the archived
   # file. Touch the extracted file. This helps build systems that compare the
   # modification times of input and output files to determine whether to do an
   # action.
   os.utime(os.path.join(output_dir, archived_file), None)
   return 0
 
 
 if __name__ == '__main__':
   sys.exit(main())