Upstream Android's PathUtils implementation.

base/base.gyp

Index: base/base.gyp
diff --git a/base/base.gyp b/base/base.gyp
index 0b200278da6de2121b11f55f03453a05b8521bd1..e3ccc3403c2be988d79a9b3a95dd383e1419d54d 100644
--- a/base/base.gyp
+++ b/base/base.gyp
@@ -115,6 +115,29 @@
       ],
     },
     {
+      'target_name': 'base_jni_headers',
+      'type': 'none',
+      'actions': [
+        {
+          'action_name': 'generate_jni_headers',

[Nico, 2012/04/03 18:51:05]

This isn't done conditionally, so a `make all` creates jni headers even when not
building for android. It seems to be fairly fast so maybe that's ok, but it's a
bit surprising.

+          'inputs': [
+            'android/jni_generator/jni_generator.py',
+            'android/java/org/chromium/base/PathUtils.java',
+          ],
+          'outputs': [
+            '<(SHARED_INTERMEDIATE_DIR)/base/jni/path_utils_jni.h',
+          ],
+          'action': [
+            'python',
+            'android/jni_generator/jni_generator.py',
+            '-o',
+            '<@(_inputs)',

[Mark Mentovai, 2012/03/13 20:16:21]

Do you really want to pass jni_generator.py back into jni_generator.py?

[Peter Beverloo, 2012/03/13 20:18:50]

As answered in PS3, per bulach@ this indeed is the way to invoke the generator.
I agree that it's odd, and will address this while looking at using a template
instead of bare actions.

+            '<@(_outputs)',
+          ],
+        }
+      ],
+    },
+    {
       'target_name': 'base_unittests',
       'type': 'executable',
       'sources': [