Propagate unknown --tool_flags to Dr. Memory as we do for Valgrind

tools/valgrind/valgrind_test.py

Index: tools/valgrind/valgrind_test.py
===================================================================
--- tools/valgrind/valgrind_test.py	(revision 95803)
+++ tools/valgrind/valgrind_test.py	(working copy)
@@ -712,21 +712,16 @@
                       help="Monitor python child processes.  If off, neither "
                       "python children nor any children of python children "
                       "will be monitored.")
-    parser._parser.add_option("", "--indirect", action="store_true",

[Timur Iskhodzhanov, 2011/08/08 13:38:04]

oops - this was copy-paste mistake

+    parser.add_option("", "--indirect", action="store_true",
                       default=False,
                       help="set BROWSER_WRAPPER rather than "
                            "running Dr. Memory directly on the harness")
     parser.add_option("", "--use_debug", action="store_true",
                       default=False, dest="use_debug",
                       help="Run Dr. Memory debug build")
-    # TODO(bruening): I want to add --extraops that can take extra
-    # args that are passed through to Dr. Memory, but the

[Timur Iskhodzhanov, 2011/08/08 13:38:04]

This is what --tool_flags was made for.
I've just forgotten to propagate it in the Dr. Memory runner

-    # chrome_tests.bat and chrome_tests.py layers combined w/
-    # chrome_tests.py parsing makes it not work out in practice.  We
-    # should change chrome_tests.py to pass all unknown options
-    # through to valgrind_test.py so we don't need to use --tool_flags

[Timur Iskhodzhanov, 2011/08/08 13:38:04]

re: to use --tool_flags or not to use
If one makes a typo in some chrome_tests flag (e.g. --t0ol), the suggested
approach would propagate --t0ol to the tool and it would fail on load. This is
not very debuggable when you start up a large test like ui_tests with
BROWSER_WRAPPER where you don't see the tool output at all...
We've had this --tool_flags for 2+ years for Valgrind-based tools and looks like
only Chrome-Valgrind team ever needed it :)

You can re-formulate and put back the TODO if you still think we should do w/o
--tool_flags.

-    # and quote it, and ditto w/ valgrind_test.py passing unknown
-    # options through to its tool.
+    parser.add_option("", "--trace_children", action="store_true",
+                            default=True,
+                            help="TODO: default value differs from Valgrind")
 
   def ToolCommand(self):
     """Get the tool command to run."""
@@ -805,7 +800,7 @@
 
     # Increase some Dr. Memory constants
     proc += ["-redzone_size", "16"]
-    proc += ["-callstack_max_frames", "30"]
+    proc += ["-callstack_max_frames", "40"]
 
     # Un-comment to ignore uninitialized accesses
     #proc += ["-no_check_uninitialized"]
@@ -813,6 +808,8 @@
     # Un-comment to ignore leaks
     #proc += ["-no_check_leaks", "-no_count_leaks"]
 
+    proc += self._tool_flags
+
     # Dr.Memory requires -- to separate tool flags from the executable name.
     proc += ["--"]