Yes in theory commons-logging would take care of the rest. If we went
with that approach I would suggest going with slf4j. However, I dont
think a test framework should require users to have to provide a logging
jar. I think it makes mrunit harder to use than necessary.
On 09/09/2012 04:25 PM, Dave Beech wrote:
OK, you're right. Fair point about not using the commons-logging brought in
transitively. It is a bad habit of mine, I guess. I think the comment still
stands for log4j though - as you say, no class from log4j is used in mrunit
and as we don't have a log4j.properties file, it is never invoked. I think
this one could be removed.
If the end user wanted to use log4j (or a different logging framework),
shouldn't they include it themselves? Our commons-logging code would take
care of the rest.
On 9 September 2012 20:45, Jim Donofrio <[email protected]> wrote:
I mean complaints on the internet about commons-logging in general.
We cannot remove commons-logging and log4j from our pom and rely on hadoop
because that is not good use of Maven. If anything log4j should be changed
to runtime scope since it is not referenced statically anywhere in an
import. I will make a JIRA to fix that.
On 09/09/2012 03:32 PM, Dave Beech wrote:
I think I'd prefer to stick with commons-logging to be consistent with
Hadoop's code. That's a good point about the dependencies though - we
could
simply remove commons-logging and log4j from our pom right now and the
build would still succeed. As Jim mentioned, commons-logging is brought in
by Hadoop transitively and log4j isn't required at all by our code.
I'm aware that slf4j is probably regarded as being better than
commons-logging these days, but I don't see much point in switching.
Where have the complaints come from? I haven't seen anything on the JIRA
or
mailing list about logging. Just curious.
Cheers,
Dave
On 9 September 2012 19:48, Jim Donofrio <[email protected]> wrote:
Since we only use logging for the most part to print out basic
information
for comparing expected inputs and outputs, what are the thoughts on
switching to the jdk logger. I admit to not knowing much about logging
but
there seems to be tons of complaints out there about commons-logging
which
we dont really need since our logging needs are simple. We could also
remove our dependencies on log4j and commons-logging even though hadoop
brings these in transitively anyway.
slf4j would be another option but then users would have to no to include
a
slf4j binding, I would rather avoid that.
If you seem to agree, I will make this change.
