Why calling LoggerFactory.getLogger(...) every time is not recommended?

Here is one obvious problem with this approach: the String messages will be constructed on each call to debug(), there is no obvious way to use a guard clause with your wrapper.

The standard idiom with log4j/commons-logging/slf4j is to use a guard clause such as:

if (log.isDebugEnabled()) log.debug("blah blah blah");

With the purpose being that if the DEBUG level is not enabled for the logger, the compiler can avoid concatenating together any longer strings you may send it:

if (log.isDebugEnabled()) log.debug("the result of method foo is " + bar 
     + ", and the length is " + blah.length());

See "What is the fastest way of (not) logging?" in the SLF4J or log4j FAQ.

I would recommend against the "wrapper" your boss suggests. A library like slf4j or commons-logging is already a facade around the actual underlying logging implementation used. In addition, each invocation of the logger becomes much lengthier - compare the above with

 MyLoggerWrapper.debug(Foo.class, "some message");

This is the type of trivial and unimportant "wrapping" and obfuscation that serves no real purpose other than adding layers of indirection and ugly-fying your code. I think your boss can find more important issues to obsess over.

The logger objects are surely reused, so no extra instantation is going to happen either way. The bigger problem I see is that your file/line number info will be useless, since the logger will always faithfully log that each message was issued from class LoggerWrapper, line 12 :-(

OTOH SLF4J is already a wrapper facade to hide the specific logging framework used, allowing you to freely change between different logging implementations. Therefore I see absolutely no point in hiding that behind yet another wrapper.

Repeated calls to LoggerFactory.getLogger(clazz) should not result in a new Logger object each time. But that does not mean that the calls are free. While the actual behaviour depends on the logging system behind the facade, it is highly likely that each getLogger entails a lookup in a concurrent or synchronized data structure¹ to look for a pre-existing instance.

If your application makes lots of calls to your MyLoggerWrapper.debug method, this can all add up to a significant performance hit. And in a multi-threaded application, it might be a concurrency bottleneck.

Other issues mentioned by other answers are also important:

• Your application can no longer use logger.isDebugEnabled() to minimize the overheads when debugging is disabled.
• The MyLoggerWrapper class obscures the class names and line numbers of your application's debug calls.
• The code using MyLoggerWrapper will probably be more verbose if you make multiple logger calls. And the verbosity will be in the area where it impacts readability most; i.e. in the methods that do things that need logging.

Finally, this is just "not the way that it is done".

^{1 - Apparently it is a Hashtable in Logback and Log4j, and that means that the potential for a concurrency bottleneck definitely exists. Note that this is not a criticism of those logging frameworks. Rather, the getLogger method was not designed/optimized to be used this way.}

To add to the reasons already mentioned, your boss's suggestion is bad because:

• It forces you to repeatedly type something that has nothing to with logging, every time you want to log something: this.getClass()
• Creates a non-uniform interface between static and non-static contexts (because there is no this in a static context)
• The additional unnecessary parameters creates room for error, makes it possible for statements in the same class to go to different loggers (think careless copy pasting)
• While it saves 74 characters of logger declaration, it adds 27 extra characters to each logging call. This means if a class uses the logger more than 2 times, you're increasing boilerplate code in terms of character count.

When using something like: MyLoggerWrapper.debug(this.getClass(), "blah") You will get wrong classnames when using AOP frameworks or code-injection tools. The classnames are not like the origin, but a generated classname. And another drawback of using the wrapper: For every log statement, you must include additional code "MyClass.class".

The 'caching' of the loggers depends on the used frameworks. But even when it does, it must still look up the desired logger for every log statement you make. So having 3 statements in a method, it must look it up 3 times. Using it as a static variable it must only look it up once!

And said before: you lose the ability to use if( log.isXXXEnabled() ){} for a set of statements.

What has your boss against the "community default accepted and recommended way"? Introducing the wrapper is not adding more efficiency. Instead you must use the classname for every log statement. After a while you want to "improve" that, so you add another variable, or another wrapper making it more difficult for yourself.