Fix build when emma_coverage changes from false->true

build/android/gyp/emma_instr.py

Index: build/android/gyp/emma_instr.py
diff --git a/build/android/gyp/emma_instr.py b/build/android/gyp/emma_instr.py
index c6ef6183935dd40d2412088eee58eb94238b3d12..f63aef6f7a330358be768bcc7297eda7eadab876 100755
--- a/build/android/gyp/emma_instr.py
+++ b/build/android/gyp/emma_instr.py
@@ -181,7 +181,11 @@ def _RunInstrumentCommand(_command, options, _, option_parser):
       print('Error: multiple output files in: %s' % (temp_jar_dir))
       return 1
 
-    shutil.copy(os.path.join(temp_jar_dir, jars[0]), options.output_path)
+    # Delete output_path first to avoid modifying input_path in the case where
+    # input_path is a hardlink to output_path. http://crbug.com/571642

[jbudorick, 2015/12/22 19:34:44]

I'm confused about how this fixes 571642. That failure appears to appear at emma
time up at line 176?

[agrieve, 2015/12/22 20:10:00]

1. Build once without coverage enabled, and there's a GN copy() target with
src=input_path dst=output_path. This results in a hardlink
2. Build once with coverage enabled. It succeeds, but since output_path is the
same inode as input_path, writing to output_path actually causes input_path to
get updated as well (they still point to the same inode after this runs)
3. Build again, and emma complains that input_path is already instrumented
(since it was wrongly modified by the previous run)

[jbudorick, 2015/12/22 20:42:35]

ah, ok. lgtm

+    if os.path.exists(options.output_path):
+      os.unlink(options.output_path)
+    shutil.move(os.path.join(temp_jar_dir, jars[0]), options.output_path)
   finally:
     shutil.rmtree(temp_dir)