Add a base::ParseJson() function to help tests construct Values.

base/test/values_test_util.h

Index: base/test/values_test_util.h
diff --git a/base/test/values_test_util.h b/base/test/values_test_util.h
index f2b5abdcf7af1b317ba289f1ef5000e11457a352..1f1853caf9d46ad5454c56e98721fa5f5aa4b7f6 100644
--- a/base/test/values_test_util.h
+++ b/base/test/values_test_util.h
@@ -7,10 +7,14 @@
 
 #include <string>
 
+#include "base/memory/scoped_ptr.h"
+#include "base/string_piece.h"
+
 namespace base {
 class DictionaryValue;
 class ListValue;
 class StringValue;
+class Value;
 
 // All the functions below expect that the value for the given key in
 // the given dictionary equals the given expected value.
@@ -39,6 +43,11 @@ void ExpectDictStringValue(const std::string& expected_value,
 void ExpectStringValue(const std::string& expected_str,
                        StringValue* actual);
 
+// Parses |json| as JSON, allowing trailing commas, and returns the

[jamesr, 2012/12/13 23:24:41]

I'm not sure I understand the trailing commas bit.  Is that just the most common
mode in our test code?

[Jeffrey Yasskin, 2012/12/13 23:29:27]

No, I tend to find that allowing trailing commas makes it easier to maintain
code, and when we're constructing a Value in tests, we'd prefer maintainable
code over portable input data. If we had an 'accept-single-quotes' mode, I'd
want to turn that on too.

Your example from layer_tree_host_perftest.cc of reading a file and passing it
through ParseJson may argue the other direction, though. I haven't counted
possible uses to figure out which is more common.

+// resulting value.  If the json fails to parse, causes an EXPECT
+// failure and returns the Null Value (but never a NULL pointer).
+scoped_ptr<Value> ParseJson(base::StringPiece json);
+
 }  // namespace base
 
 #endif  // BASE_TEST_VALUES_TEST_UTIL_H_