My biggest gripe about g_test_trap_subprocess() is how it requires
changes all over the test case C file to get working. A test case needs
to know it's own path, and additional tests need to be added with
appropriate names. 

The above makes sense for pretty advanced cases, but for bog simple
this-should-produce-a-critical-warning test, it's completely overkill.

Luckily we can use g_test_subprocess() to rerun a test case itself
in a subprocess. Like so:

  static void
  test_build_null_string (void)
  {
  	GHashTable *attributes;
  
  	if (g_test_subprocess ()) {
  		attributes = secret_attributes_build (&MOCK_SCHEMA,
  		                                      "number", 4,
  		                                      "string", NULL,
  		                                      "even", TRUE,
  		                                      NULL);
  		g_assert (attributes == NULL);
  		return;
  	}
  	
  	g_test_trap_subprocess ("/attributes/build-null-string", 0, 
	                        G_TEST_SUBPROCESS_INHERIT_STDOUT);
  	g_test_trap_assert_failed ();
  	g_test_trap_assert_stderr ("*attribute*NULL*");
  }

As you can see one complication remains ... a test case must know its
own path. This not only adds complexity, but prevents a single such
testcase function being used with multiple different fixtures, at
different paths.

So attached patch changes g_test_trap_subprocess to accept NULL
as it's test name, and it'll just run the same test in a subprocess.

Allows elegant use of subprocesses.

Updated documentation to remove complex example and replace with
simple one.