[infinispan-dev] trace vs log.isTraceEnabled

[Emmanuel Bernard]

On 7 déc. 2011, at 17:37, Sanne Grinovero wrote:

> On 7 December 2011 16:15, Manik Surtani <manik at jboss.org> wrote:
>> Sanne,
>> 
>> +1 to consistency.  I suggest we
>> repurpose https://issues.jboss.org/browse/ISPN-1429 to a consistent logging
>> scheme.  My preference is for a static final boolean trace =
>> log.isTraceEnabled().
>> 
>> But I do see Erik's concerns about wanting to switch on tracing on the fly.
>> 
>> - Manik
> 
> Not sure if you can wait, but as soon as I have some free time (i.e.
> christmas?) I'll make a POC of a solution taking the best of two
> worlds. Need a POC before talking further about it :D

Just to put some perspectives on things.
Yes the code will waste some cycles checking for trace level but I'd rather have you guys invent a better distro or local / tx / lock algorithm that will speed Infinispan by 10x rather than chasing 1/10 of a %. Just saying :)

Profilers are fun but they tend to make you lose the big picture.