(In reply to comment #29)
> > FWIW the fix in fixSubstitutedLoggers() method isn't really thread-safe either.
> 
> As per current code flow it is thread safe. 'fixSubstitutedLoggers' is
> invoked *after* the INITIALIZATION_STATE is set to
> SUCCESSFUL_INITIALIZATION. So by the time it is invoked TEMP_FACTORY would
> not be in use and no further SubstituteLogegrs would be created.
> 
> So the only downside is that only during the initialization phase some log
> calls would be NOOP and log output would be lost. Similar situtation occurs
> in Logback if the Logback config is reset and any log message arrives during
> that phase. 
> 
> If that also needs to be addressed then we can move just the SimpleLogger to
> slf4j-api module and use that in place of NOOP logger for the initial
> delegate in SubstituteLogger

Sorry for the late reply. Yeah, sorry I missed that. 

That said I still don't understand why initialization just can't be
thread-safe. Using SubstitueLoggers and what not just hides the issue. The
bottom line is initialization of a binder is not thread-safe and I see no
reason why it shouldn't be. If it's just made thread-safe there is no need for
all this substitute logger code.