First pass at content shell for the Mac. Doesn't work yet.

content/content_shell.gypi

Index: content/content_shell.gypi
diff --git a/content/content_shell.gypi b/content/content_shell.gypi
index 72820bc7483464f866f4ecb815decc23f487132e..b5b1767dfcad983b7f382ca2b0dd0427eccb1657 100644
--- a/content/content_shell.gypi
+++ b/content/content_shell.gypi
@@ -3,6 +3,9 @@
 # found in the LICENSE file.
 
 {
+  'variables': {
+    'content_shell_product_name': 'Content Shell',
+  },
   'targets': [
     {
       'target_name': 'content_shell_lib',
@@ -114,7 +117,19 @@
               },
             },
           },
-        }],
+        }],  # OS=="win"
+        ['OS=="mac"', {
+          'type': 'shared_library',
+          'product_name': '<(content_shell_product_name) Framework',

[Mark Mentovai, 2012/02/28 20:12:40]

No Info.plist?

[Avi (use Gerrit), 2012/02/29 19:46:50]

Do I need one?

+          'mac_bundle': 1,
+          'mac_bundle_resources': [
+            'shell/mac/English.lproj/MainMenu.xib',
+            '<(SHARED_INTERMEDIATE_DIR)/content/shell_resources.pak'
+          ],
+          'xcode_settings': {
+            'EXPORTED_SYMBOLS_FILE': 'shell/mac/exported_symbols.exp',

[Mark Mentovai, 2012/02/28 20:12:40]

That name’s kinda redundant, no? Following the example in Chrome, I’d expect the
name to have something to do with “framework”, and the Info.plist to be
framework-Info.plist, to disambiguate from the names that have to do with the
app or other bundles or linker output.

I don’t think that this is even necessary now that you’re using
__attribute__((visibility("default"))).

[Avi (use Gerrit), 2012/02/29 19:46:50]

mooted.

+          },
+        }],  # OS=="mac"
       ],
     },
     {
@@ -194,6 +209,7 @@
     {
       'target_name': 'content_shell',
       'type': 'executable',
+      'mac_bundle': 1,
       'defines!': ['CONTENT_IMPLEMENTATION'],
       'variables': {
         'chromium_code': 1,
@@ -209,6 +225,19 @@
         'app/startup_helper_win.cc',
         'shell/shell_main.cc',

[Mark Mentovai, 2012/02/28 20:12:40]

In this file:

#if defined(OS_MACOSX)
__attribute__((visibility("default")))
int main(int argc, const char* argv[]) {
#else
int main(int argc, const char** argv) {
#endif  // OS_MACOSX

1. The “const” is technically illegal, and

2. There’s no reason that non-OS_MACOSX can’t use “char* argv[]” like the
OS_MACOSX branch.

As I will mention elsewhere, exposing the delegate from the framework seems kind
of unclean to me. I think that the best approach would be to allow ContentMain
to accept a NULL delegate; in the case of a NULL delegate passed in, it should
use its own built-in ContentShellDelegate as the default. This will eliminate
the need to pass the default delegate in and therefore eliminate the need to
export these extra symbols.

[Avi (use Gerrit), 2012/02/29 19:46:50]

Simplified.
Now that we have shell_browser_main, we can do that there. Otherwise it would be
a gross layering violation.

       ],
+      'mac_bundle_resources': [
+        'shell/mac/app.icns',
+        'shell/mac/Info.plist',

[Mark Mentovai, 2012/02/28 20:12:40]

app-Info.plist. Again on 236 and 239.

[Avi (use Gerrit), 2012/02/29 19:46:50]

Done.

+      ],
+      # TODO(mark): Come up with a fancier way to do this.  It should only
+      # be necessary to list app-Info.plist once, not the three times it is
+      # listed here.
+      'mac_bundle_resources!': [
+        'shell/mac/Info.plist',
+      ],
+      'xcode_settings': {
+        'INFOPLIST_FILE': 'shell/mac/Info.plist',
+      },
       'msvs_settings': {
         'VCLinkerTool': {
           'SubSystem': '2',  # Set /SUBSYSTEM:WINDOWS
@@ -230,17 +259,40 @@
               },
             },
           },
-        }],
+        }],  # OS=="win"
         ['OS == "win" or (toolkit_uses_gtk == 1 and selinux == 0)', {
           'dependencies': [
             '../sandbox/sandbox.gyp:sandbox',
           ],
-        }],
+        }],  # OS=="win" or (toolkit_uses_gtk == 1 and selinux == 0)
         ['toolkit_uses_gtk == 1', {
           'dependencies': [
             '<(DEPTH)/build/linux/system.gyp:gtk',
           ],
-        }],
+        }],  # toolkit_uses_gtk
+        ['OS=="mac"', {
+          'product_name': '<(content_shell_product_name)',
+          'postbuilds': [

[Mark Mentovai, 2012/02/28 20:12:40]

It’d be nice to run tweak_info_plist just to get the SVN keys set. Some things
in base that do version number lookups actually care about that. You’d run
tweak_info_plist with -s1 and the other things false.

[Mark Mentovai, 2012/02/28 20:12:40]

The “verify” build step from chrome_exe would be very nice to have here. There
should be no Objective-C in the main executable. See bug 88697.

[Avi (use Gerrit), 2012/02/29 19:46:50]

Done.

[Avi (use Gerrit), 2012/02/29 19:46:50]

Done.

+            {
+              'postbuild_name': 'Copy <(content_shell_product_name) Framework.framework',
+              'action': [
+                '../build/mac/copy_framework_unversioned',
+                '${BUILT_PRODUCTS_DIR}/<(content_shell_product_name) Framework.framework',
+                '${BUILT_PRODUCTS_DIR}/${CONTENTS_FOLDER_PATH}/Frameworks',
+              ],
+            },
+            {
+              'postbuild_name': 'Fix Framework Link',
+              'action': [
+                'install_name_tool',
+                '-change',
+                '/Library/Frameworks/<(content_shell_product_name) Framework.framework/Versions/A/<(content_shell_product_name) Framework',
+                '@executable_path/../Frameworks/<(content_shell_product_name) Framework.framework/<(content_shell_product_name) Framework',
+                '${BUILT_PRODUCTS_DIR}/${EXECUTABLE_PATH}'
+              ],
+            },
+          ],
+        }],  # OS=="mac"
       ],
     },
   ],