Most of the API that you need for using g3log is described in this readme. For more API documentation and examples please continue to read the [API readme](API.markdown). Examples of what you will find here are:
Conditional logging is made with ```LOG_IF(INFO, <boolean-expression>) << " some text" ``` or ```LOGF_IF(WARNING, <boolean-expression>) << " some text".``` Only if the expressions evaluates to ```true``` will the logging take place.
*<aname="fatal_logging">A call using FATAL</a> logging level, such as the ```LOG_IF(FATAL,...)``` example above, will after logging the message at ```FATAL```level also kill the process. It is essentially the same as a ```CHECK(<boolea-expression>) <<...```withthedifferencethatthe```CHECK(<boolean-expression)```triggerswhentheexpressionevaluatesto```false```.*
The contract API follows closely the logging API with ```CHECK(<boolean-expression>) <<...```forstreamingor(*)```CHECKF(<boolean-expression>, ...);``` for printf-style.
If the ```<boolean-expression>``` evaluates to false then the the message for the failed contract will be logged in FIFO order with previously made messages. The process will then shut down after the message is sent to the sinks and the sinks have dealt with the fatal contract message.
(\* * ```CHECK_F(<boolean-expression>, ...);``` was the the previous API for printf-like CHECK. It is still kept for backwards compatability but is exactly the same as ```CHECKF``` *)
The default logging levels are ```DEBUG```, ```INFO```, ```WARNING``` and ```FATAL``` (see FATAL usage [above](#fatal_logging)). The logging levels are defined in [loglevels.hpp](src/g3log/loglevels.hpp).
For some windows framework there is a clash with the ```DEBUG``` logging level. One of the CMake [Build options](#build_options) can be used to then change offending default level from ```DEBUG``` TO ```DBUG```.
When the option is enabled there will be a slight runtime overhead for each ```LOG``` call when the enable/disable status is checked. For most intent and purposes this runtime overhead is negligable.
There is **no** runtime overhead for internally checking if a level is enabled//disabled if the cmake option is turned off. If the dynamic logging cmake option is turned off then all logging levels are enabled.
Custom logging levels can be created and used. When defining a custom logging level you set the value for it as well as the text for it. You can re-use values for other levels such as *INFO*, *WARNING* etc or have your own values.
Any value with equal or higher value than the *FATAL* value will be considered a *FATAL* logging level.
The default sink for g3log is the one as used in g2log. It is a simple file sink with a limited API. The details for the default file sink can be found in [filesink.hpp](src/g3log/filesink.hpp), [filesink.cpp](src/filesink.cpp), [filesinkhelper.ipp](src/filesinkhelper.ipp)
More sinks can be found at [g3sinks](http://www.github.com/KjellKod/g3sinks) (log rotate, log rotate with filtering on levels)
A logging sink is not required to be a subclass of a specific type. The only requirement of a logging sink is that it can receive a logging message of
### Using the default sink
Sink creation is defined in [logworker.hpp](src/g3log/logworker.hpp) and used in [logworker.cpp](src/logworker.cpp). For in-depth knowlege regarding sink implementation details you can look at [sinkhandle.hpp](src/g3log/sinkhandle.hpp) and [sinkwrapper.hpp](src/g3log/sinkwrapper.hpp)
### Override log formatting in default and custom sinks
The default log formatting look can be overriden by any sink.
If the sink receiving function calls `toString()` then the default log formatting will be used.
If the sink receiving function calls `toString(&XFunc)` then the `XFunc`will be used instead (see `LogMessage.h/cpp` for code details if it is not clear). (`XFunc` is a place holder for *your* formatting function of choice).
The API for the function-ptr to pass in is
```
std::string (*) (const LogMessage&)
```
or for short as defined in `LogMessage.h`
```
using LogDetailsFunc = std::string (*) (const LogMessage&);
```
### Override the log formatting in the default sink
The default file sink will flush each log entry as it comes in. For different flushing policies please take a look at g3sinks [logrotate and LogRotateWithFilters](http://www.github.com/KjellKod/g3sinks/logrotate).
At shutdown all enqueued logs will be flushed to the sink.
At a discovered fatal event (SIGSEGV et.al) all enqueued logs will be flushed to the sink.
A programmatically triggered abrupt process exit such as a call to ```exit(0)``` will of course not get the enqueued log entries flushed. Similary a bug that does not trigger a fatal signal but a process exit will also not get the enqueued log entries flushed. G3log can catch several fatal crashes and it deals well with RAII exits but magic is so far out of its' reach.
Example usage where a [logrotate sink (g3sinks)](https://github.com/KjellKod/g3sinks) is added. In the example it is shown how the logrotate API is called. The logrotate limit is changed from the default to instead be 10MB. The limit is changed by calling the sink handler which passes the function call through to the actual logrotate sink object.
The default build uses a fixed size buffer for formatting messages. The size of this buffer is 2048 bytes. If an incoming message results in a formatted message that is greater than 2048 bytes, it will be bound to 2048 bytes and will have the string ```[...truncated...]``` appended to the end of the bound message. There are cases where one would like to dynamically change the size at runtime. For example, when debugging payloads for a server, it may be desirable to handle larger message sizes in order to examine the whole payload. Rather than forcing the developer to rebuild the server, dynamic message sizing could be used along with a config file which defines the message size at runtime.
The default behaviour for G3log is to catch several fatal events before they force the process to exit. After <i>catching</i> a fatal event a stack dump is generated and all log entries, up to the point of the stack dump are together with the dump flushed to the sink(s).
The default fatal handling on Linux deals with fatal signals. At the time of writing these signals were ```SIGABRT, SIGFPE, SIGILL, SIGILL, SIGSEGV, SIGSEGV, SIGTERM```. The Linux fatal handling is handled in [crashhandler.hpp](src/g3log/crashhandler.hpp) and [crashhandler_unix.cpp](src/crashhandler_unix.cpp)
A signal that commonly is associated with voluntarily process exit is ```SIGINT``` (ctrl + c) G3log does not deal with it.
By default the fatal signals are defined in [src/g3log.cpp](src/g3log.cpp) as
```
SIGABRT
SIGFPE
SIGILL
SIGSEGV
SIGTERM
```
If you want to define your own set of fatal signals, override the default ones, then this can be done as shown in [src/g3log/crashhandler.hpp](src/g3log/crashhandler.hpp)
```
// Example when SIGTERM is skipped due to ZMQ usage
Fatal signal handling can be disabled with a CMake option: `ENABLE_FATAL_SIGNALHANDLING`. See [Options.cmake](https://github.com/KjellKod/g3log/blob/master/Options.cmake) for more details
### <aname="PID1">PID1 Fatal Signal Recommendations</a>
If you are using g3log on a PID1 process then you absolutely should provide your own signal handling (ref: [issue 269](https://github.com/KjellKod/g3log/issues/269)) as g3log re-emits the fatal signal after it has restored the previous signal handler for that signal. PID1 processed do *not* shutdown the process for a normal fatal signal so the choice to exit the PID1 process after such a signal must be taken by the coder - not by g3log.
Windows fatal handling also deals with fatal signals just like Linux. In addition to fatal signals it also deals with unhandled exceptions, vectored exceptions. Windows fatal handling is handled in [crashhandler.hpp](src/g3log/crashhandler.hpp), [crashhandler_windows.cpp](src/crashhandler_windows.cpp), [stacktrace_windows.hpp](src/g3log/stacktrace_windows.hpp), [stacktrace_windows.cpp](src/stacktrace_windows.cpp)
This logger is available for free and all of its source code is public domain.
Writing API documentation is probably the most boring task for a developer.
Did it help you? Could it be better? Please suggest changes, send me feedback or even better: open a pull request.
You could also contribute by saying thanks with a a donation. It would go a long way not only to show your support but also to boost continued development of this logger and its API documentation
