# Implementing a logging system in C++17

I've been programming for what is a probably a little bit and made a very simple logging system for my personal projects. While it works out fairly well so far, I would like advice on how to make it perform better and stuff I can do to make it more usable in general.

Main questions include:

1. Is my code thread safe? If not, how do I make it thread safe?

2. Is there a way to make the interface a bit cleaner for use? Right now #define feels a bit hacky.

3. Is there a way to avoid the use of a global unique_ptr?

4. Should I just make it a header only library? Is there any benefits in doing so?

.hpp:

#pragma once

/*
Use it by including this header in every file required, and in your main function start a new log.
Logger::startLog("Log.txt");
Use the various error levels by naming them and simply passing the info and what you want to output.
Logger::log(ERROR, "Something went wrong.");
*/

// For the unique pointers.
#include <memory>
// Filestream.
#include <fstream>
// String class for names and parameters passed around.
#include <string>

#define FATAL Logger::Level::Fatal
#define ERROR Logger::Level::Error
#define WARNING Logger::Level::Warning
#define INFO Logger::Level::Info
#define DEBUG Logger::Level::Debug

namespace Logger {
// Severity level enum.
enum class Level {
Fatal,
Error,
Warning,
Info,
Debug
};

// Initialize the log.
void startLog(const std::string& filepath);

// Log a message.
void log(Level s, const std::string& msg);

// Logging class.
class Log {
public:
Log(const std::string& filepath);
void addLog(Level s, const std::string& msg);
~Log();
private:
// File for logging.
std::ofstream m_logfile;
std::string levels[5] = {"Fatal", "Error", "Warning", "Info", "Debug"};
};
}


.cpp:

#include "Log.hpp"

namespace Logger {
// Global Logging Object.
std::unique_ptr<Log> g_log;

// Initalize our logging object.
void startLog(const std::string& filepath) {
g_log = std::make_unique<Log>(filepath);
Logger::log(Level::Info, "Started logging system.");
}

// Method which logs.
void log(Level s, const std::string& msg) {
}

// Create our global logging object.
Log::Log(const std::string& filepath) : m_logfile{} {
m_logfile.open(filepath);
}

// Add a message to our log.
void Log::addLog(Level s, const std::string& msg) {
if (m_logfile.is_open()) {
m_logfile << levels[static_cast<int>(s)] << ": " << msg << std::endl;
}
}

Log::~Log() {
m_logfile.close();
}
}


1. Is my code thread safe? If not, how do I make it thread safe?

No, of course it's not thread-safe. You don't do anything to make it thread-safe.

A more nuanced answer would be: It's thread-safe as long as you don't use it in an unsafe way. For example, calling Logger::log(FATAL, "hello world") from two different threads concurrently would of course be unsafe. But if your program has only one thread... :)

If you want to allow calling Logger::log from two threads concurrently, you'll have to do something to eliminate the data race on m_logfile which is caused by the two threads' both calling m_logfile << levels[static_cast<int>(s)] at the same time. For example, you could throw a mutex lock around addLog.

1. Is there a way to make the interface a bit cleaner for use? Right now #define feels a bit hacky.

The only place you use #define is in #define FATAL Logger::Level::Fatal and so on. (By the way, technically #define ERROR ... triggers undefined behavior, because all macros of the form EXXXX are reserved for use by POSIX error codes.) My question is, if you wanted these values to be referred to as FATAL, ERROR, etc., why didn't you just declare them that way?

inline constexpr int FATAL = 0;
inline constexpr int ERROR = 1;
inline constexpr int WARNING = 2;
// ...


Or, probably better:

namespace Logger {
enum Level {
FATAL, ERROR, WARNING, // ...
};
}


Making this an enum (rather than an enum class) allows your user to refer to the enumerators without needing to redundantly name the enum type: just Logger::FATAL, Logger::ERROR, et cetera.

Personally, I would consider writing convenience functions to eliminate the boilerplate:

namespace Logger {
void log_fatal(const std::string& msg) { log(FATAL, msg); }
void log_error(const std::string& msg) { log(ERROR, msg); }
// ...
}


By the way, I think your numbering scheme is backwards. "Level" represents the severity of the message, right? Some messages have higher severity than others? So how would I test whether one message had a higher severity than another? Well, I think I'd write:

if (one_severity > another_severity) ...


But with the values you gave your enumeration, this is actually going to be completely backwards! And so my code for testing severity levels is going to have a bug (or else I'll catch the bug, but then have to write unintuitive code that uses < to mean "greater than"). So, I recommend switching the values around.

1. Is there a way to avoid the use of a global unique_ptr?

Sure; declare it static! And to make it really non-global, stick it in a function. It'll still have static lifetime, though. There's not much getting around that.

inline Log *get_glog() {
static std::unique_ptr<Log> glog = std::make_unique<Log>();
return glog.get();
}

void startLog(const std::string& filepath) {
Log *glog = get_glog();
glog->set_filepath_and_open(filepath);
}


1. Should I just make it a header only library? Is there any benefits in doing so?

The big benefit of a header-only library is that it's super easy to incorporate into another project — the user just drops the header file into his include/ directory and he's good to go.

Single-header libraries are particularly nice because they can easily be dropped into Godbolt or Wandbox.

In your case, the tradeoff is that presumably this header is going to get included all over the place (because logging is ubiquitous), and so the bigger you make it, the more work you're forcing the compiler to do in every translation unit.

Since you said "C++17", consider rewriting all your void foo(const std::string& s) signatures into void foo(std::string_view sv) signatures. (Yes, pass string_view by value.)

• Thank you for the insight. I think I managed to resolve most of my concerns and update the code. – Rietty Jan 22 at 16:47

A couple of things. You can avoid the #defines with a using directive . You don't need to give the caller any knowledge of the Log class. It does not need to be in the header at all. What happens if I call start log more than once? There's no protection against that. Also, I know it probably works as is, but the static cast in add log scares me because you never defined an integer value for the enum class. It seems to be begging for a segfault.

Beyond that, how to improve the interface depends in what you want. Currently the code is not thread safe. You could use lock guards in your methods to accomplish this.

Do you want to have one central log? Or do you want each instantiation of the log to be isolated?

If you want one central, consider the singleton pattern and give out shared or weak pointers via a static factory, and make the ctor private. Then instead of having the function interface, you can keep a shared pointer to the logging object and give it references to it, people can then call the functions on the class.

If you want each installation of the log to be separate, just have the user construct a log class.

• Welcome to Code Review. You didn't use any <code>yourcode</code> tags (with Markdown: yourcode), so you might be interested in the formatting section of our help center. – Zeta Jan 22 at 6:36

I know that this already has an excepted answer but I will demonstrate the current Logger that I am using and what it looks like. Mind you though; it inherits from a Singleton class so that only one Logger can be constructed per a single application instance. It also relies on a few other classes in which I can not show here.

There was some functionality to make it thread safe in the original design, however I am currently in the process of updating this class from using Windows headers for CRITICAL_SECTION and replacing them with mutex and lock_guard.

There is also an ExceptionHandler class where the two work together in an integrated manner, however, no exceptions can be thrown from within the Logger since the ExceptionHandler itself uses the Logger. This way if an exception is thrown and caught; it will write that exception both to the console and the log file.

The declaration of the Logger class looks like this:

-Logger.h-

#pragma once

#include "Singleton.h"

#include <array>
#include <string>
#include <sstream>

namespace util {

class Logger final : public Singleton {
public:
enum Type {
INFO = 0,
WARNING,
ERROR
};

private:
std::string filename_;
unsigned maxCharLength_;

std::array<std::string, 3> logTypes_;
const std::string unknownLogType_;

// CRICTICAL_SECTION  // replace with mutex and lockguard...

public:
explicit Logger(const std::string& filename);
virtual ~Logger();

Logger(const Logger& c) = delete;
Logger& operator=(const Logger& c) = delete;

static void log(const std::string& text, Type type = INFO);
static void log(const std::ostringstream& stream, Type type = INFO);
static void log(const char* text, Type type = INFO);
};

} // namespace util


And it's implementation looks like this:

-Logger.cpp-

#include "Logger.h"

#include "TextFileWriter.h"

#include <conio.h>
#include <iomanip>
#include <iostream>

namespace util {
static Logger* spLogger_ = nullptr;
} // namespace util

using namespace util;

Logger::Logger(const std::string & filename) :
Singleton( LOGGER ),
filename_( filename ),
maxCharLength_( 0 ),
unknownLogType_( "UNKNOWN" ) {
// Order MUST MATCH Types in Logger::Type
logTypes_[0] = "Info";
logTypes_[1] = "Warning";
logTypes_[2] = "Error";

// Find widest log type string
maxCharLength_ = static_cast<unsigned int>( unknownLogType_.size() );
for( const std::string& logType : logTypes_) {
if (maxCharLength_ < logType.size()) {
maxCharLength_ = static_cast<unsigned int>( logType.size() );
}
}

// critical section - mutex - thread lock

// Start Log File
TextFileWriter file(filename, false, false);

spLogger_ = this;
}

Logger::~Logger() {
spLogger_ = nullptr;

// remove critical section or destroy mutex - lockguard, thread etc.
}

void Logger::log(const std::string & text, Type type) {
log(text.c_str(), type);
}

void Logger::log(const std::ostringstream & stream, Type type) {
log(stream.str().c_str(), type);
}

#include <Windows.h>

void Logger::log(const char * text, Type type) {
if (nullptr == spLogger_) {
std::cout << "Logger has not been initialized, can not log " << text << std::endl;
}

// Choose Log Type text string, display "UNKNOWN" if Type is out of range.
std::ostringstream stream;
stream << std::setfill(' ') << std::setw(spLogger_->maxCharLength_);

try {
stream << spLogger_->logTypes_.at(type);
} catch (...) {
stream << spLogger_->unknownLogType_;
}

// I am currently in the process of removing Windows specific code:
// I am trying to do something similar to the date & time code below
// but in a more generic, portable and cross - platform way using
// only stand library code.

// Date & Time
SYSTEMTIME time;
GetLocalTime(&time);

stream << " [" << time.wYear << '.'
<< std::setfill('0') << std::setw(2) << time.wMonth << '.'
<< std::setfill('0') << std::setw(2) << time.wDay << ' '
<< std::setfill(' ') << std::setw(2) << time.wHour << ':'
<< std::setfill('0') << std::setw(2) << time.wMinute << ':'
<< std::setfill('0') << std::setw(2) << time.wSecond << '.'
<< std::setfill('0') << std::setw(3) << time.wMilliseconds << "] ";

stream << text << std::endl;

std::cout << stream.str();

// Save message to log file
try {
TextFileWriter file(spLogger_->filename_, true, false);
file.write(stream.str());
} catch (...) {
// Not saved in log file, write message to console
std::cout << __FUNCTION__ << " failed to write to file: " << stream.str() << std::endl;
}
}


A few things to be aware of with this class is that this depends on a TextFileWriter class which I can not disclose.

I am also in the process of replacing the Date & Time section to be something more portable and standard instead of using Windows headers.

If you look closely at the design above, You can see that the Logger class contains an enum for basic types of errors. Their severity is already in order: {INFO, WARNING, ERROR }. I can easily expand this class to have different types of errors. The class stores a string for its filename, simple enough. There is an array of strings where the size of the array matches the types of errors.

The constructor simply takes a string as a filename. It initializes the names of the error types with a string value. The maxCharLength is currently used for formatting the output to the console and written text file. I use the TextFileWriter class to create a file instead of having this responsibility within the Logger. This way I can have many different things read and write to files through the FileHandler classes. There is also a static pointer to this class that gets set as the last operation being done within the constructor.

The destructor set's the pointer back to null, and will eventually remove or destroy any mutexes or locks.

There are 3 overloaded log functions that all get passed to the one that takes a const char*. They have a second parameter which allows the user or caller to specify the type of error and it has a default value of INFO just incase you want to quickly write something to the console that isn't an error. The log function will also write the same message with the same format to a log file.

Also Logger must be the first thing created because there is a stipulation in its parent class Singleton that Logger must be created before any other Singleton type. It was originally designed this way because there are several Manager type classes where you only ever want a single Manager class. So the Logger, Settings and various Manager - classes not shown here other than the Logger all inherit from Singleton. Once they are constructed an error will be thrown if you try to construct another per run of application.

Here is what a sample application would look like while using the Logger class:

-main.cpp-

#include "Logger.h"
#include "ExceptionHandler.h"

int main() {
try {
Logger log( "log.txt" );

Logger::log( "Hello World!" ); // Default Info Type
Logger::log( "Hello World!, Logger::INFO ); // Explicitly Stating Info Type
Logger::log( "Hello World!, Logger::WARNING ); // Warning Type
Logger::log( "Hello World!, Logger::ERROR ); // Error Type

// And since my ExceptionHandler also uses the Logger

ExceptionHandler( "Good Bye Cruel World!" );

} catch( ExceptionHandler& e ) {
std::cout << "Exception Thrown: " << e.getMessage() << std::endl;
return EXIT_FAILURE;
} catch( ... ) {
// Mind you I believe __FUNCTION__ is Windows or Visual Studio specific.
// I've tried using __PRETTY_FUNCTION__ but for my purposes the formatting
// wasn't quite right. You can use your desired choice.
std::cout << __FUNCTION__ << " Caught Unknown Exception" << std::endl;
}
}


My output to the console and text file would look like this from the above main.cpp.

-Output-

   Info [2019.02.20  6:54:17.282] Hello World!
Info [2019.02.20  6:54:17.283] Hello World!
Warning [2019.02.20  6:54:17.284] Hello World!
Error [2019.02.20  6:54:17.285] Hello World!
Error [2019.02.20  6:54:17.285] Good Bye Cruel World!


Another thing that was also incorporated but is Window's specific is that the code works with the standard console and it's property settings. When the logger writes to the console depending on what kind of text is being printed, basic cout will always white text against a black background, but as for the Logger types, each different type of error has a different set of parameters for the console's text color and background color. This way while debugging an application and some error is generated and or thrown; I can easily tell what type of error I'm reading just by the color of the text and background.

-Disclosure-

I hope that this gives you some insight towards a design of a versatile Logger. Now this code is not 100% mine as I did not fully write this. The overall structure and implementation has come from Marek A. Krzeminski as the original code is (c). You can find the full source material here.

He has given me permission to share parts of his code in situations like this provided that: I do not show fully compliable source code and that I give him credit when I do so. I am free to use his original code for my own personal uses since I have subscribed to his content in that past as long as I don't use any of it for commercial purposes. I do however, like to use parts of his code when applicable when someone needs some advice towards specific algorithms, functionality or just overall code design. I like to take what I have learned from him and to share that learning experience. I am also in the process of redesigning his original class to be my own that is fully generic, portable, cross-platform capable while using modern C++ standards and techniques.