I should have rephrased this. We cannot assume that the caller of the Log4j API method is the caller we need. It might be another Log4j API method or another logging Adapter. Since walking the stack trace is relatively slow we have to defer doing it until we absolutely need to. This generally means we are not walking only a few frames. We have run benchmarks that show using SecurityManager to find the caller class is faster than using StackWalker.
https://lists.apache.org/thread.html/29ca67b98c8a508f2836301ac33747b72ca0ce86f69f466073412bfc@%3Cdev.logging.apache.org%3E <https://lists.apache.org/thread.html/29ca67b98c8a508f2836301ac33747b72ca0ce86f69f466073412bfc@%3Cdev.logging.apache.org%3E> Ralph > On Oct 4, 2018, at 9:25 PM, Ralph Goers <rgo...@apache.org> wrote: > > Unfortunately, we can’t use getCallerClass. Some API methods can be called by > users of the API or by other methods in the API class. > > Ralph > >> On Oct 4, 2018, at 2:50 PM, Mandy Chung <mandy.ch...@oracle.com >> <mailto:mandy.ch...@oracle.com>> wrote: >> >> If you are looking for the immediate caller, you can try >> StackWalker::getCallerClass which only walks the top few >> frames and intends to be lower cost. >> >> Mandy >> >> On 10/4/18 2:04 PM, Ralph Goers wrote: >>> Hmm, it would probably be a safe assumption that a Logger will never be >>> used outside of the module. That isn’t true of the class name, method name >>> and line number though. I’m not sure how much extra overhead there is in >>> collecting the module name when you have to collect those to. >>> >>> I should add that when printing exceptions we do cache the file >>> name/location as we are processing the classes for an extended stack trace. >>> We probably will want to add the module name to the extended stack trace as >>> well. >>> >>> Ralph >>> >>>> On Oct 4, 2018, at 10:26 AM, fo...@univ-mlv.fr <mailto:fo...@univ-mlv.fr> >>>> wrote: >>>> >>>> I was thinking about capturing the call stack when you create the logger >>>> (to get the module), not when you call the logger. >>>> >>>> cheers, >>>> Rémi >>>> >>>> ----- Mail original ----- >>>>> De: "Ralph Goers" <ralph.go...@dslextreme.com> >>>>> <mailto:ralph.go...@dslextreme.com> >>>>> À: "Alex Sviridov" <ooo_satu...@mail.ru> <mailto:ooo_satu...@mail.ru> >>>>> Cc: "Remi Forax" <fo...@univ-mlv.fr> <mailto:fo...@univ-mlv.fr>, >>>>> "jigsaw-dev" <jigsaw-dev@openjdk.java.net> >>>>> <mailto:jigsaw-dev@openjdk.java.net> >>>>> Envoyé: Mercredi 3 Octobre 2018 05:08:27 >>>>> Objet: Re: Separate logging for JPMS module/layer >>>>> Log4j handles this by capturing the fully qualified class name of the >>>>> logging >>>>> adapter. Obviously, this doesn’t work if the adapter doesn’t pass Log4j >>>>> the >>>>> FQCN, but it does work for the adapters we support. That said, it is >>>>> very slow >>>>> to capture this and is probably the biggest pain point. Log4j recommends >>>>> not >>>>> capturing this information in production environments because it is so >>>>> slow. >>>>> Unfortunately, it seems to have gotten even slower in Java 9+. In an ideal >>>>> world we would be able to capture the caller information at compile time >>>>> but >>>>> Java provides no good way to do this. Wouldn’t it be great if I could >>>>> just code >>>>> something like logger.error(_CallerInfo_, “hello”) and the compiler would >>>>> provide the caller info data structure that was generated by the compiler? >>>>> >>>>> FWIW, I do plan to add the module information to the caller information >>>>> provided >>>>> with Log4j but just haven’t gotten to it. You are more than welcome to >>>>> provide >>>>> a patch. >>>>> >>>>> Ralph >>>>> >>>>>> On Oct 2, 2018, at 3:20 PM, Alex Sviridov <ooo_satu...@mail.ru> >>>>>> <mailto:ooo_satu...@mail.ru> wrote: >>>>>> >>>>>> Thank you for you suggestion. But can this be used when some library >>>>>> uses one logging system and for another uses some bridge. Because of this >>>>>> bridging >>>>>> LoggerFactory.getLogger is called somewhere in bridge, as I understand, >>>>>> >>>>>> >>>>>>> Среда, 3 октября 2018, 1:12 +03:00 от Remi Forax <fo...@univ-mlv.fr> >>>>>>> <mailto:fo...@univ-mlv.fr>: >>>>>>> >>>>>>> You can use the StackWalker >>>>>>> https://docs.oracle.com/en/java/javase/11/docs/api/java.base/java/lang/StackWalker.html >>>>>>> >>>>>>> <https://docs.oracle.com/en/java/javase/11/docs/api/java.base/java/lang/StackWalker.html> >>>>>>> >>>>>>> regards, >>>>>>> Rémi >>>>>>> >>>>>>> ----- Mail original ----- >>>>>>>> De: "Alex Sviridov" < ooo_satu...@mail.ru <mailto:ooo_satu...@mail.ru> >>>>>>>> > >>>>>>>> À: "jigsaw-dev" < jigsaw-dev@openjdk.java.net >>>>>>>> <mailto:jigsaw-dev@openjdk.java.net> > >>>>>>>> Envoyé: Mardi 2 Octobre 2018 23:54:48 >>>>>>>> Objet: Separate logging for JPMS module/layer >>>>>>>> Hi all, >>>>>>>> >>>>>>>> Could anyone say how the following problem can be solved. I want to >>>>>>>> create >>>>>>>> separate >>>>>>>> log file for every JPMS module/layer. The problem is that many >>>>>>>> libraries/programs >>>>>>>> use LoggerFactory.getLogger(String className) so in getLogger I have >>>>>>>> only >>>>>>>> the name of the class as String, so I can't get module and layer. >>>>>>>> >>>>>>>> If I had not String className, but Class klass then the problem would >>>>>>>> be easily >>>>>>>> solved. >>>>>>>> As I understand I can't load class by name because it would require >>>>>>>> all modules >>>>>>>> export >>>>>>>> their packages to logging framework that has no sense. >>>>>>>> >>>>>>>> Are there any solutions for such problem? >>>>>>>> >>>>>>>> >>>>>>>> -- >>>>>>>> Alex Sviridov >>>>>> -- >>>>>> Alex Sviridov >> >
