Enable blink_gc_plugin tests on Windows.

tools/clang/blink_gc_plugin/tests/test.py

 #!/usr/bin/env python
 # Copyright 2015 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.
 
 
 import glob
 import os
 import subprocess
 import sys
@@ skipping 24 matching lines @@
     try:
       actual = subprocess.check_output(
           ['python', '../process-graph.py', '-c',
            '%s.graph.json' % test_base_name],
           stderr=subprocess.STDOUT)
     except subprocess.CalledProcessError, e:
       # The graph processing script returns a failure exit code if the graph is
       # 'bad' (e.g. it has a cycle). The output still needs to be captured in
       # that case, since the expected results capture the errors.
       actual = e.output
+    finally:
+      # Clean up the .graph.json file to prevent false passes from stale results
+      # from a previous run.
+      os.remove('%s.graph.json' % test_base_name)

[dcheng, 2015/05/14 05:28:18]

This caused me to miss that -fsyntax-only isn't actually good enough.

 
   # TODO(dcheng): Remove the rstrip() and just rebaseline the tests to match.
   actual = actual.rstrip()
 
+  # On Windows, clang emits CRLF as the end of line marker. Normalize it to LF
+  # to match posix systems.
+  actual = actual.replace('\r\n', '\n')
+
   try:
     expected = open('%s.txt' % test_base_name).read().rstrip()
   except IOError:
-    open('%s.txt' % test_base_name, 'w').write(actual)
+    open('%s.txt.actual' % test_base_name, 'w').write(actual)

[dcheng, 2015/05/14 05:28:18]

Somehow I checked in the wrong version of this file

     return 'no expected file found'
 
   if expected != actual:
-    open('%s.txt' % test_base_name, 'w').write(actual)
+    open('%s.txt.actual' % test_base_name, 'w').write(actual)
     return 'expected and actual differed'
 
 
 def run_tests(clang_path, plugin_path):
   """Runs the tests.
 
   Args:
     clang_path: The path to the clang binary to be tested.
     plugin_path: An optional path to the plugin to test. This may be None, if
                  plugin is built directly into clang, like on Windows.
 
   Returns:
     (passing, failing): Two lists containing the base names of the passing and
                         failing tests respectively.
   """
   passing = []
   failing = []
 
-  base_cmd = [clang_path, '-fsyntax-only', '-std=c++11']
+  # The plugin option to dump the object graph is incompatible with
+  # -fsyntax-only. It generates the .graph.json file based on the name of the
+  # output file, but there is no output filename with -fsyntax-only.
+  base_cmd = [clang_path, '-c', '-std=c++11']
   base_cmd.extend(['-Wno-inaccessible-base'])
   if plugin_path:
     base_cmd.extend(['-Xclang', '-load', '-Xclang', plugin_path])
   base_cmd.extend(['-Xclang', '-add-plugin', '-Xclang', 'blink-gc-plugin'])
 
   tests = glob.glob('*.cpp')
   for test in tests:
     sys.stdout.write('Testing %s... ' % test)
     test_base_name, _ = os.path.splitext(test)
 
@@ skipping 25 matching lines @@
   clang_path = sys.argv[1]
   plugin_path = sys.argv[2] if len(sys.argv) > 2 else None
   print 'Using clang %s...' % clang_path
   print 'Using plugin %s...' % plugin_path
 
   passing, failing = run_tests(clang_path, plugin_path)
   print 'Ran %d tests: %d succeeded, %d failed' % (
       len(passing) + len(failing), len(passing), len(failing))
   for test in failing:
     print '    %s' % test
+  return len(failing)
 
 
 if __name__ == '__main__':
   sys.exit(main())