[Box Backup-dev] Logging

[Martin Ebourne]

There seems to have been a huge amount of discussion on here about 
logging. I'm not replying to any specific message because they all have 
points.

I'm pretty sure we can come up with something that satisfies everyone's 
requirements, although it will include iostream using <<. ;)

Here's an example line from one of my projects. It's using log4cxx, but 
we can do something just the same, and since we only want a small 
subset of the functionality, it shouldn't be hard.

  LOG4CXX_DEBUG(s_log, "Incoming event: " << time << ", " << sourceType
                << ", " << sourceName << ", " << payload);

This is a macro, so it can be defined to NULL and therefore have zero 
impact in release builds.

It could expand to many things, but one possibility would be:

#define BOX_DEBUG(logger, stuff) \
do { \
  std::ostringstream line; \
  line << stuff; \
  logger.write(BOX_DEBUG_LEVEL, __FUNCTION__, __LINE__, line.str());
} while(0)

Another would be:

#define BOX_DEBUG(logger, stuff) \
do { \
  logger.stream(BOX_DEBUG_LEVEL) << stuff; \
} while(0)

For maximum performance while still being able to disable certain 
levels at run time this would do:

#define BOX_DEBUG(logger, stuff) \
do { \
  if(logger.active(BOX_DEBUG_LEVEL)) \
  { \
    std::ostringstream line; \
    line << stuff; \
    logger.write(BOX_DEBUG_LEVEL, __FUNCTION__, __LINE__, line.str());
  } \
} while(0)

Then separate macros for each of BOX_INFO, BOX_ERROR, BOX_FATAL.

Only the debug one has to compile out, and even that can be made 
conditional with configure.ac. We loose the TRACEn where the n really 
isn't nice at all.

As to the objects and over engineering. Really we only need a few small 
classes with a couple of virtual methods in each. There's no need to 
make this hugely complicated.

Cheers,

Martin.