Change rtc::TimeNanos and rtc::TimeMicros return value from uint64_t to int64_t.

webrtc/base/opensslstreamadapter.cc

 /*
  *  Copyright 2004 The WebRTC Project Authors. All rights reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 46 matching lines @@
 static SrtpCipherMapEntry SrtpCipherMap[] = {
     {"SRTP_AES128_CM_SHA1_80", SRTP_AES128_CM_SHA1_80},
     {"SRTP_AES128_CM_SHA1_32", SRTP_AES128_CM_SHA1_32},
     {"SRTP_AEAD_AES_128_GCM", SRTP_AEAD_AES_128_GCM},
     {"SRTP_AEAD_AES_256_GCM", SRTP_AEAD_AES_256_GCM},
     {nullptr, 0}};
 #endif
 
 #ifdef OPENSSL_IS_BORINGSSL
 static void TimeCallback(const SSL* ssl, struct timeval* out_clock) {
-  uint64_t time = TimeNanos();
+  int64_t time = TimeNanos();
   out_clock->tv_sec = time / kNumNanosecsPerSec;
+  // TODO(nisse): The below looks very wrong.

[pthatcher1, 2016/11/18 22:11:16]

Why?

[nisse-webrtc, 2016/11/21 07:45:07]

It's wrong in several ways. 

With struct timeval representation, we should have 

 nanosecs == tv_sec * kNumNanoSecsPerSec +
             tv_usec * kNumNanoSecsPerMicroSec

with tv_usec in the range [0, 999999].

The most severe error is that the assignment to tv_usec includes the seconds
already accounted for, and in addition, the result if the division most likely
doesn't fit in 32 bits, and truncation makes the value even worse.

It should be

  tv_usec = (time % kNumNanoSecsPerSec) /
             kNumNanoSecsPerMicroSec;

Second, with the above fix, we get bad values if time happens to be negative,
because integer division in C truncates to zero, rather than producing the
floor, resulting in tv_usec < 0. (That wasn't an issue before the change in this
cl).

Third, speaking of the sign of time, tv_sec is usually expected to be seconds
since the unix 1970 epoch, while the epoch used for TimeNanos is arbitrary,
e.g., uptime. If TimeCallback is only used for calculation of intervals,
timeouts and the like, that's no problem, and actually an improvement over
gettimeofday, since the latter jumps whenever the system time is changed. But
even so, using struct timeval seems confusing.

[kwiberg-webrtc, 2016/11/21 12:12:34]

Post a bug about this and point to it in the comment?

[nisse-webrtc, 2016/11/21 12:39:12]

Filed bug https://bugs.chromium.org/p/webrtc/issues/detail?id=6737

I'll consider just removing this TODO, since it's not really related to the rest
of the cl.

   out_clock->tv_usec = time / kNumNanosecsPerMicrosec;
 }
 #else  // #ifdef OPENSSL_IS_BORINGSSL
 
 // Cipher name table. Maps internal OpenSSL cipher ids to the RFC name.
 struct SslCipherMapEntry {
   uint32_t openssl_id;
   const char* rfc_name;
 };
 
@@ skipping 1180 matching lines @@
         return true;
     }
   }
 
   return false;
 }
 
 }  // namespace rtc
 
 #endif  // HAVE_OPENSSL_SSL_H