Update some comments and code to reflect the fact that HMAC() is no longer hostile to NULL key data.

crypto/hmac_openssl.cc

Index: crypto/hmac_openssl.cc
diff --git a/crypto/hmac_openssl.cc b/crypto/hmac_openssl.cc
index 92eea19d780270cde6fa593986ed2015e9687f85..6cbd1b2b2ff1a8fe0c356f61fc98cbf7879a5f76 100644
--- a/crypto/hmac_openssl.cc
+++ b/crypto/hmac_openssl.cc
@@ -32,11 +32,8 @@ bool HMAC::Init(const unsigned char* key, size_t key_length) {
 
   plat_->key.assign(key, key + key_length);
   if (key_length == 0) {

[davidben, 2015/03/04 19:40:59]

Hrm. This seems to conflict with the documentation. Given we have passing tests
for it, I suspect this comment is just a lie.

https://code.google.com/p/chromium/codesearch#chromium/src/crypto/hmac.h&q=HM...

-    // Special-case: if the key is empty, use a key with one zero
-    // byte. OpenSSL's HMAC function breaks when passed a NULL key. (It calls
-    // HMAC_Init_ex which treats a NULL key as having already been initialized
-    // with a key previously.) HMAC pads keys with zeros, so this key is
-    // equivalent.
+    // HMAC::Sign() assumes that plat_->key is not empty.
+    // HMAC pads keys with zeros, so this key is equivalent.

[davidben, 2015/03/04 19:40:59]

Hrm. This seems to only be relevant for the DCHECK. What if we instead defer
initializing plat_ until HMAC::Init? Then we can DCHECK on plat_ existing
without extra state.

     plat_->key.push_back(0);
   }
   return true;