GLOG不初始化同樣會輸出到stderr以及LOG(INFO)解析

glog的LOG(INFO)<<"xx";到底是怎樣執行的？

1)源碼

#define LOG(severity) COMPACT_GOOGLE_LOG_ ## severity.stream()

可知
LOG(INFO) == COMPACT_GOOGLE_LOG_ INFO.stream()
2)源碼

#if GOOGLE_STRIP_LOG == 0
#define COMPACT_GOOGLE_LOG_INFO google::LogMessage( \
      __FILE__, __LINE__)
#define LOG_TO_STRING_INFO(message) google::LogMessage( \
      __FILE__, __LINE__, google::GLOG_INFO, message)
#else

可知
LOG(INFO) == google::LogMessage( __FILE__, __LINE__).stream()
3）源碼

ostream& LogMessage::stream() {
  return data_->stream_;
}

可知，LOG(INFO)<<"data..."將要寫得信息data…插入到data_->stream_流裏面。
4）源碼

LogMessage::~LogMessage() {
  Flush();
  delete allocated_;
}

每次寫日誌得時候，創建一個LogMessage類，什麼時候寫吶，就是當LogMessage析構得時候，調用Flush()寫。
5）源碼

void LogMessage::Flush() {
  if (data_->has_been_flushed_ || data_->severity_ < FLAGS_minloglevel)
    return;

  data_->num_chars_to_log_ = data_->stream_.pcount();
  data_->num_chars_to_syslog_ =
    data_->num_chars_to_log_ - data_->num_prefix_chars_;

  // Do we need to add a \n to the end of this message?
  bool append_newline =
      (data_->message_text_[data_->num_chars_to_log_-1] != '\n');
  char original_final_char = '\0';

  // If we do need to add a \n, we'll do it by violating the memory of the
  // ostrstream buffer.  This is quick, and we'll make sure to undo our
  // modification before anything else is done with the ostrstream.  It
  // would be preferable not to do things this way, but it seems to be
  // the best way to deal with this.
  if (append_newline) {
    original_final_char = data_->message_text_[data_->num_chars_to_log_];
    data_->message_text_[data_->num_chars_to_log_++] = '\n';
  }

  // Prevent any subtle race conditions by wrapping a mutex lock around
  // the actual logging action per se.
  {
    MutexLock l(&log_mutex);
    (this->*(data_->send_method_))();
    ++num_messages_[static_cast<int>(data_->severity_)];
  }
  LogDestination::WaitForSinks(data_);

  if (append_newline) {
    // Fix the ostrstream back how it was before we screwed with it.
    // It's 99.44% certain that we don't need to worry about doing this.
    data_->message_text_[data_->num_chars_to_log_-1] = original_final_char;
  }

  // If errno was already set before we enter the logging call, we'll
  // set it back to that value when we return from the logging call.
  // It happens often that we log an error message after a syscall
  // failure, which can potentially set the errno to some other
  // values.  We would like to preserve the original errno.
  if (data_->preserved_errno_ != 0) {
    errno = data_->preserved_errno_;
  }

  // Note that this message is now safely logged.  If we're asked to flush
  // again, as a result of destruction, say, we'll do nothing on future calls.
  data_->has_been_flushed_ = true;
}

實際得logging操作是以上代碼中

// Prevent any subtle race conditions by wrapping a mutex lock around
  // the actual logging action per se.
  {
    MutexLock l(&log_mutex);
    (this->*(data_->send_method_))();
    ++num_messages_[static_cast<int>(data_->severity_)];
  }

通過LogMessage類函數指針成員變量send_method_進行寫得，send_method_的賦值即以下函數，也就是說，實際寫操作是在以下函數中進行。

void LogMessage::SendToLog() EXCLUSIVE_LOCKS_REQUIRED(log_mutex) {
  static bool already_warned_before_initgoogle = false;

  log_mutex.AssertHeld();

  RAW_DCHECK(data_->num_chars_to_log_ > 0 &&
             data_->message_text_[data_->num_chars_to_log_-1] == '\n', "");

  // Messages of a given severity get logged to lower severity logs, too

  if (!already_warned_before_initgoogle && !IsGoogleLoggingInitialized()) {
    const char w[] = "WARNING: Logging before InitGoogleLogging() is "
                     "written to STDERR\n";
    WriteToStderr(w, strlen(w));
    already_warned_before_initgoogle = true;
  }

  // global flag: never log to file if set.  Also -- don't log to a
  // file if we haven't parsed the command line flags to get the
  // program name.
  if (FLAGS_logtostderr || !IsGoogleLoggingInitialized()) {
    ColoredWriteToStderr(data_->severity_,
                         data_->message_text_, data_->num_chars_to_log_);

    // this could be protected by a flag if necessary.
    LogDestination::LogToSinks(data_->severity_,
                               data_->fullname_, data_->basename_,
                               data_->line_, &data_->tm_time_,
                               data_->message_text_ + data_->num_prefix_chars_,
                               (data_->num_chars_to_log_ -
                                data_->num_prefix_chars_ - 1));
  } else {

    // log this message to all log files of severity <= severity_
    LogDestination::LogToAllLogfiles(data_->severity_, data_->timestamp_,
                                     data_->message_text_,
                                     data_->num_chars_to_log_);

    LogDestination::MaybeLogToStderr(data_->severity_, data_->message_text_,
                                     data_->num_chars_to_log_);
    LogDestination::MaybeLogToEmail(data_->severity_, data_->message_text_,
                                    data_->num_chars_to_log_);
    LogDestination::LogToSinks(data_->severity_,
                               data_->fullname_, data_->basename_,
                               data_->line_, &data_->tm_time_,
                               data_->message_text_ + data_->num_prefix_chars_,
                               (data_->num_chars_to_log_
                                - data_->num_prefix_chars_ - 1));
    // NOTE: -1 removes trailing \n
  }

以上函數有幾點值得注意：

• 如果實際使用中沒有調用InitGoogleLogging()進行初始化，那麼一切的LOG只會輸出到stderr中，即：

if (!already_warned_before_initgoogle && !IsGoogleLoggingInitialized()) {
    const char w[] = "WARNING: Logging before InitGoogleLogging() is "
                     "written to STDERR\n";
    WriteToStderr(w, strlen(w));
    already_warned_before_initgoogle = true;
  }

• 正如上面所說，也只有調用了initGoogleLogging()進行過初始化，所有設置的全局變量，例如FLAG_xxx,等等纔會生效。因爲寫日誌的函數裏會檢查是否初始化，如果未初始化，一切全局變量都沒用上，直接輸出到stderr.
• 但實際上InitGoogleLogging()函數本身並未做什麼初始化的工作

void InitGoogleLogging(const char* argv0) {
  glog_internal_namespace_::InitGoogleLoggingUtilities(argv0);
}
void InitGoogleLoggingUtilities(const char* argv0) {
  CHECK(!IsGoogleLoggingInitialized())
      << "You called InitGoogleLogging() twice!";
  const char* slash = strrchr(argv0, '/');
#ifdef OS_WINDOWS
  if (!slash)  slash = strrchr(argv0, '\\');
#endif
  g_program_invocation_short_name = slash ? slash + 1 : argv0;
  g_main_thread_id = pthread_self();
 
#ifdef HAVE_STACKTRACE
  InstallFailureFunction(&DumpStackTraceAndExit);
#endif
}

    g_program_invocation_short_name，g_main_thread_id這兩個變量只是獲取一下信息爲了防止重複初始化和打印時輸出線程ID，根本不控制是否可以輸出。

caffe源代碼中glog的應用框架

    glog不允許重複初始化，在caffe中的初始化都是在具體的exe中主線程進行初始化。
    比如在convert_imageset項目中的convert_imageset.cpp中main函數第一句進行進行初始化

int main(int argc, char** argv) {
#ifdef USE_OPENCV
  ::google::InitGoogleLogging(argv[0]);
  // Print output to stderr (while still logging)
  FLAGS_alsologtostderr = 1;
  ...
}

    再比如再caffe.exe裏面,通過caffe::GlobalInit函數裏面調用InitGoogleLogging()進行初始化

int main(int argc, char** argv) {
  // Print output to stderr (while still logging).
  FLAGS_alsologtostderr = 1;
  // Set version
  gflags::SetVersionString(AS_STRING(CAFFE_VERSION));
  // Usage message.
  gflags::SetUsageMessage("command line brew\n"
      "usage: caffe <command> <args>\n\n"
      "commands:\n"
      "  train           train or finetune a model\n"
      "  test            score a model\n"
      "  device_query    show GPU diagnostic information\n"
      "  time            benchmark model execution time");
  // Run tool or show usage.
  caffe::GlobalInit(&argc, &argv);
  ...
  }

在前面已經提到，如果這些exe中沒有進行初始化，則log只會輸出到stderr中。