Testing: Fixed printing FilePaths when a test fails comparing them.

base/files/file_path.h

Index: base/files/file_path.h
diff --git a/base/files/file_path.h b/base/files/file_path.h
index 5225b12ffd4c8cab478e5b73c8f6a9823c0c3f9c..e4c70d30fb40829a449ccb00dc354930e206e164 100644
--- a/base/files/file_path.h
+++ b/base/files/file_path.h
@@ -433,10 +433,10 @@ class BASE_EXPORT FilePath {
   StringType path_;
 };
 
-}  // namespace base
-
 // This is required by googletest to print a readable output on test failures.
-BASE_EXPORT extern void PrintTo(const base::FilePath& path, std::ostream* out);
+BASE_EXPORT extern void PrintTo(const FilePath& path, std::ostream* out);

[danakj, 2015/05/22 20:59:23]

While you're here, including std::ostream everywhere isn't free, and linking
this into production code for gtest isn't quite right.

James did something nice for our gfx types to support PrintTo in tests, but not
in prod:
https://chromium.googlesource.com/chromium/src/+/a109fd05fac9af35d219de6e4285...

Can we do something like that for this? (ie use isofwd only and move the defn to
a test build target.)

[Matt Giuca, 2015/05/25 03:40:15]

Done (moved to test_support_base), but...

On 2015/05/22 20:59:23, danakj wrote:
I'm confused about where std::ostream is coming from. I went to remove "#include
<iostream>" but it wasn't there... does that mean one of the other header files
is including it?

This means regardless of whether I do this change, won't we still be paying the
cost of iostream? (I'm unsure exactly what the cost is... is it having to
process the iostream headers in every file that includes file_path.h? Or is it
linking in the ostream code at all?)

[Matt Giuca, 2015/05/25 05:54:01]

Sorry, by <iostream> I meant <ostream>.

+
+}  // namespace base
 
 // Macros for string literal initialization of FilePath::CharType[], and for
 // using a FilePath::CharType[] in a printf-style format string.