Don't try to use -MMD / .o.d depfiles for ninja with .s files.

test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py

Index: test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py
diff --git a/test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py b/test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py
new file mode 100755
index 0000000000000000000000000000000000000000..5df5b45f9d401016f69e43ccd3164fbdbb5992f1
--- /dev/null
+++ b/test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py
@@ -0,0 +1,42 @@
+#!/usr/bin/env python
+
+# Copyright (c) 2012 Google Inc. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+"""
+Verify that .s files don't always trigger a rebuild, as would happen if depfiles
+were used for them (since clang & gcc ignore -MMD when building .s->.o).

[Nico, 2012/10/18 18:09:39]

"when building .s->o on linux" (doesn't happen on os x)

[Ami GONE FROM CHROMIUM, 2012/10/18 18:14:38]

Are you sure?  The interwebz tell me otherwise.

(does this test pass for you on osx without the corresponding change to
ninja.py?)

[Nico, 2012/10/18 18:30:21]

Pretty sure. See the ninja-build thread on this, where I checked.
Yes:

hummer:gyp-rw thakis$ python gyptest.py
test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py 
PYTHONPATH=/Volumes/MacintoshHD2/src/gyp-rw/test/lib
TESTGYP_FORMAT=make
/usr/bin/python test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py
Invalid test for 'make' format; skipping test.
NO RESULT
TESTGYP_FORMAT=ninja
/usr/bin/python test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py
PASSED
TESTGYP_FORMAT=xcode
/usr/bin/python test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py
Invalid test for 'xcode' format; skipping test.
NO RESULT

No result from the following 2 tests:
	test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py
	test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py
hummer:gyp-rw thakis$ svn revert pylib/gyp/generator/ninja.py
Reverted 'pylib/gyp/generator/ninja.py'
^P^Phummer:gyp-rw thakpython gyptest.py
test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py -f ninja
PYTHONPATH=/Volumes/MacintoshHD2/src/gyp-rw/test/lib
TESTGYP_FORMAT=ninja
/usr/bin/python test/ninja/s-needs-no-depfiles/gyptest-s-needs-no-depfiles.py
PASSED
hummer:gyp-rw thakis$ 

+"""
+
+import os
+import sys
+import TestCommon
+import TestGyp
+
+# NOTE(fischman): Each generator uses depfiles (or not) differently, so this is
+# a ninja-specific test.
+test = TestGyp.TestGyp(formats=['ninja'])
+
+if sys.platform == 'win32' or sys.platform == 'win64':
+  # This test is about clang/gcc vs. depfiles; VS gets a pass.
+  test.pass_test()
+  sys.exit(0)
+
+test.run_gyp('s-needs-no-depfiles.gyp')
+
+# Build the library, grab its timestamp, rebuild the library, ensure timestamp
+# hasn't changed.
+test.write(os.path.join(test.workdir, 'empty.s'), '')

[Nico, 2012/10/18 18:09:39]

Can't you just check in an empty 'empty.s'?

[Ami GONE FROM CHROMIUM, 2012/10/18 18:14:38]

I did that with an earlier revision, but it made git try cry b/c one of the
patching steps apparently can't cope with empty files...

[Nico, 2012/10/18 18:30:21]

"# This file intentionally left empty." Regular bots can handle empty files too
I think, so maybe that's not even necessary.

+test.build('s-needs-no-depfiles.gyp', 'empty')
+empty_dll = test.built_file_path('empty', test.SHARED_LIB)
+test.built_file_must_exist(empty_dll)
+pre_stat = os.stat(test.built_file_path(empty_dll))
+test.sleep()
+test.build('s-needs-no-depfiles.gyp', 'empty')
+post_stat = os.stat(test.built_file_path(empty_dll))
+
+if pre_stat.st_mtime != post_stat.st_mtime:
+  test.fail_test()
+else:
+  test.pass_test()