Add unit tests for NSS-backed SSLPrivateKeys.

net/ssl/ssl_platform_key_chromecast.cc

Index: net/ssl/ssl_platform_key_chromecast.cc
diff --git a/net/ssl/ssl_platform_key_chromecast.cc b/net/ssl/ssl_platform_key_chromecast.cc
index b913e77f828c16fed95e330481d9d25f8ca5844a..5993a3bbe3e3727484567ee3c2ee492543c459ce 100644
--- a/net/ssl/ssl_platform_key_chromecast.cc
+++ b/net/ssl/ssl_platform_key_chromecast.cc
@@ -41,11 +41,8 @@ class SSLPlatformKeyChromecast : public ThreadedSSLPrivateKey::Delegate {
   SSLPrivateKey::Type GetType() override { return SSLPrivateKey::Type::RSA; }
 
   std::vector<SSLPrivateKey::Hash> GetDigestPreferences() override {
-    static const SSLPrivateKey::Hash kHashes[] = {
-        SSLPrivateKey::Hash::SHA256, SSLPrivateKey::Hash::SHA1,
-        SSLPrivateKey::Hash::MD5_SHA1};

[Ryan Sleevi, 2016/12/14 18:42:39]

This change isn't obvious - was removing MD5_SHA1 intentional (because of the
RSA special logic)? Or is it because the TPM doesn't support it?

[davidben, 2016/12/14 18:49:18]

Oh, I mentioned this on the previous patchset, but I forgot to transplant it on
re-upload. It's the first one:

MD5-SHA1 is weird and isn't advertised. (Doesn't make a difference whether we
advertise here it or not since SSLClientSocketImpl will filter it out, but other
SSLPrivateKey implementations don't list it.) The prefs are used for TLS 1.1 and
below where you always sign MD5-SHA1 for RSA. [Amended: The prefs are use for
TLS 1.2 and up. MD5-SHA1 is just a magic value for TLS 1.1 and below. Since it's
always used, we can just ask the caller to sign it and try it.]

-    return std::vector<SSLPrivateKey::Hash>(kHashes,
-                                            kHashes + arraysize(kHashes));
+    return std::vector<SSLPrivateKey::Hash>{SSLPrivateKey::Hash::SHA256,
+                                            SSLPrivateKey::Hash::SHA1};
   }
 
   size_t GetMaxSignatureLengthInBytes() override {