Making the browser tests work on Unix

Porting the browser tests to Unix.
The browser tests are an alternative to UI tests.
They provide a way to exercise the browser from within the test (without having the test and the browser running in different processes).
In order to ensure atexit hanlders are run after each tests and static initializers start fresh for each test, each test is run in a new process (on Linux and Mac). On Windows, a DLL containing the test is loaded/unloaded for each tests.

BUG=None
TEST=Run the browser tests. 


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=17781

[Amanda Walker, 2009-06-01 18:29:00]

LGTM!  Only one minor nit.

http://codereview.chromium.org/115896/diff/1031/1033
File webkit/glue/plugins/plugin_lib_mac.mm (right):

http://codereview.chromium.org/115896/diff/1031/1033#newcode303
Line 303: // not leaked.
This comment is hard to parse--is there an extra "not" in here, perhaps?

[tony, 2009-06-01 19:21:20]

drive-by wstring/filepath nits

http://codereview.chromium.org/115896/diff/75/1089
File base/native_library.h (right):

http://codereview.chromium.org/115896/diff/75/1089#newcode66
Line 66: std::wstring GetNativeLibraryName(const std::wstring& name);
Nit: We're trying to get rid of wstring in the code base.  A simple fix would be
to return a string16 instead.  However, it may be better to just return a
FilePath (the input arg could be ASCII (std::string)).

http://codereview.chromium.org/115896/diff/75/1091
File chrome/test/browser/browser_test_launcher_in_proc.cc (right):

http://codereview.chromium.org/115896/diff/75/1091#newcode53
Line 53: LOG(INFO) << "Loading '" <<  lib_path.ToWStringHack() << "'";
Nit: Can you use .value() instead of ToWStringHack?

[jcampan, 2009-06-03 00:02:49]

http://codereview.chromium.org/115896/diff/75/1089
File base/native_library.h (right):

http://codereview.chromium.org/115896/diff/75/1089#newcode66
Line 66: std::wstring GetNativeLibraryName(const std::wstring& name);
On 2009/06/01 19:21:20, tony wrote:
> Nit: We're trying to get rid of wstring in the code base.  A simple fix would
be
> to return a string16 instead.  However, it may be better to just return a
> FilePath (the input arg could be ASCII (std::string)).
Would using a FilePath really make sense as this is not a path?
I changed to use string16.

http://codereview.chromium.org/115896/diff/75/1091
File chrome/test/browser/browser_test_launcher_in_proc.cc (right):

http://codereview.chromium.org/115896/diff/75/1091#newcode53
Line 53: LOG(INFO) << "Loading '" <<  lib_path.ToWStringHack() << "'";
On 2009/06/01 19:21:20, tony wrote:
> Nit: Can you use .value() instead of ToWStringHack?

Done.

[tony, 2009-06-03 23:22:47]

LGTM! (sorry, didn't realize you were waiting on me)

[sgk, 2009-06-05 21:59:09]

http://codereview.chromium.org/115896/diff/1123/1139
File chrome/chrome.gyp (right):

http://codereview.chromium.org/115896/diff/1123/1139#newcode3626
Line 3626: # IMPORTANT NOTE: you must also put them in browser_tests_dll
So people don't have to duplicate, let's put these in a variable defined up in
the top 'variables' section:

'browser_test_sources': [
  'browser/ssl/ssl_browser_tests.cc',
  etc.
],

And then you can include the variable in both this and the sources list below as
follows (including a comment to direct people where to look):

  # <@(browser_tests_sources) defined in 'variables'
  # at the top of the file
  '<@(browser_tests_sources)',

http://codereview.chromium.org/115896/diff/1123/1139#newcode4033
Line 4033: 'dependencies': [
I'd expect this to have a dependency on 'browser_tests_dll'.  Yes, it's not
necessary for actally building the .exe, but since the .dll is (I presume)
necessary to actually *run* browser_tests.exe and have it do something, I think
it makes sense to guarantee that it's there.  We make 'chrome.exe' (in the 'app'
target) depend on 'chrome.dll' for similar reasons.