Ralph, I agree with you entirely. My intent for these new log methods is that they only be called from compile-time generated code.
On Sat, Dec 9, 2017 at 4:30 PM, Ralph Goers <[email protected]> wrote: > I don’t understand how this is a good idea. To use this you would need to > do something like: > > Message msg = new StringMessage(getCaller(), “My Message”); > logger.debug(msg); > > Unfortunately the line number would point to the line where getCaller() is > called not to the logger statement. > > I had thought about modifying AbstractLogger to do > public void debug(final Marker marker, final String message) { > logIfEnabled(getCaller(), Level.DEBUG, marker, message, (Throwable) > null); > } > instead of the current > public void debug(final Marker marker, final String message) { > logIfEnabled(FQCN, Level.DEBUG, marker, message, (Throwable) null); > } > But the amount of changes required to get it into the LogEvent was large. > OTOH, if we create a CallerLocationMessage that contains the > StackTraceElement and then have existing Messages extend that then we could > store the location in the Message if it is a CallerLocationMessage. Calling > getCaller() in this way would be much better since it is at a fixed depth > from the caller. > > With Java 9 this could become: > public void debug(final Marker marker, final String message) { > logIfEnabled(stackWalker.walk(s->s.skip(1).findFirst(), Level.DEBUG, > marker, message, (Throwable) null); > } > although that would pass a StackFrame instead of a StackTraceElement. The > only problems with this is that there is still some overhead in calling > StackWalker like this. Also, if this is called from a facade, such as > log4j-slf4j-impl then the number of levels that have to be skipped would be > different. > > I would really prefer if there was some way to capture the line number > information for the various loggers when the annotation processor runs at > compile time. > > Ralph > > > On Dec 9, 2017, at 1:22 PM, Jeffrey Shaw <[email protected]> wrote: > > > > Thanks for the link, Mikael. I'll take a look at it. > > > > I added some plumbing to core to allow clients to pass a > StackTraceElement > > to loggers. I'd like a code review. I'm happy to try other methods. See > the > > following commit. > > https://github.com/shawjef3/logging-log4j2/commit/ > 9c42073e9ca4f25a2f4075b1791eee2893534c54 > > > > On Sat, Dec 9, 2017 at 10:09 AM, Mikael Ståldal <[email protected]> > wrote: > > > >> Have you tried the Log4j Scala API? > >> > >> http://logging.apache.org/log4j/2.x/manual/scala-api.html > >> > >> It does currently not support this, but it uses Scala macros, and this > >> could be added there. But log4j-api and/or log4j-core probably needs to > >> adapted as well. > >> > >> > >> > >> On 2017-12-09 07:30, Jeffrey Shaw wrote: > >> > >>> Hello, > >>> I've found that I am able to use Scala macros to provide compile-time > >>> source information for log messages. However, I don't see a way to > inject > >>> this into log4j's logging mechanism. > >>> > >>> I'm wondering if there is something I'm missing, or if LogEvent's > >>> getSource > >>> method could be duplicated in Message. > >>> > >>> We could then have zero-overhead location information in logs. I'm > >>> thinking > >>> that tools other than Scala could also take advantage of this. > >>> > >> > >
