Implement TestRootCerts for Android

net/base/test_root_certs.h

Index: net/base/test_root_certs.h
diff --git a/net/base/test_root_certs.h b/net/base/test_root_certs.h
index 4d52dbcda5c088fd520785659413b42dfad1bbb3..a63fdb6e758d13dd1a5630f6e35139244da3b98c 100644
--- a/net/base/test_root_certs.h
+++ b/net/base/test_root_certs.h
@@ -91,7 +91,7 @@ class NET_EXPORT_PRIVATE TestRootCerts {
   base::mac::ScopedCFTypeRef<CFMutableArrayRef> temporary_roots_;
 #endif
 
-#if defined(OS_WIN) || defined(USE_OPENSSL)
+#if defined(OS_WIN) || defined(USE_OPENSSL) || defined(OS_ANDROID)

[Ryan Sleevi, 2012/12/07 03:25:26]

Isn't this redundant here - USE_OPENSSL is always true for OS_ANDROID

[ppi, 2012/12/11 04:24:21]

David already pointed this out earlier (so that's 2:1 in favor of removal).

My rationale for introducing the condition is the following:

"I wonder if relying on the USE_OPENSSL flag being defined on Android in this
case is not unnecessary coupling (what if we drop OpenSSL on Android in the
future?).

Also, shouldn't it be possible to read the code of test_root_certs.h in the
context of test_root_certs_android.cc without the knowledge about the
USE_OPENSSL flag being defined on Android?"

Please let me know if that seems convincing to you.

[Ryan Sleevi, 2012/12/11 17:40:10]

Whenever such a comment is coupled with "more code", I tend to be resistant.
"What ifs" can easily be used to justify a lot.
To date, we've hidden the fact that Android is using OpenSSL (except where it
isn't) by simply using USE_OPENSSL as the #define, and using code substitution
(via .gyp).

In general, I'm in favour of exposing as few details as possible about the
implementation within the code itself, and letting the .gyp rules hold as much
as possible.

So I'm not sure this #ifdef really adds much.

[ppi, 2012/12/11 19:12:50]

Removed the condition in patch set 11.

   // True if there are no temporarily trusted root certificates.
   bool empty_;
 #endif