Looks like we're in a tricky spot... I count four people in favor of keeping the current levels who don't want to add levels (Paul, Christian, Matt and myself), two people who really want to add levels (Gary and Nick) and two people who are not pushing for new levels but don't mind the change (Scott and Ralph).
I don't think of adding new levels as a compromise, as it would only satisfy a few of us. Can we try finding a way that would satisfy all parties? Why don't we think a little more about finding some mechanism that allows users to add custom levels that are *as easy or nearly as easy to use as the pre-defined levels*? That mechanism can be anything (markers, extensible enum, an interface, something else?), I am open to ideas. This is an engineering problem, let's use our engineering skills (instead of our debate skills :-).) Let's think about this more, with open minds, before rushing into a solution that many have reservations about. On Thu, Jan 23, 2014 at 6:07 AM, Paul Benedict <[email protected]> wrote: > Yes, I know. It is a big change but it is also one that's necessary to > support custom logging levels that are enums. > > > On Wed, Jan 22, 2014 at 3:04 PM, Ralph Goers > <[email protected]>wrote: > >> Paul, >> >> Take a look at the Logger, LoggerConfig and Lo4jLogEvent classes and >> LogEvent interface in log4j-core. Then look at the Filter interface and >> the ThresholdFilter implementation. Looking at those classes you will see >> that the change you are proposing has a much larger impact than what you >> are thinking. Since the custom levels would not be part of the enum all >> the code would have to be changed to use the new Interface you are >> proposing, not the Level enum. >> >> Ralph >> >> >> On Jan 22, 2014, at 12:37 PM, Paul Benedict <[email protected]> wrote: >> >> Ralph, >> >> Perhaps you're misunderstanding or I was unclear (let's say it's the >> latter). >> >> The interface is only so you can allow custom log levels as an enum. The >> client code could still use the enums you provided today. All that's >> changing is the API signatures to accept the interface -- which >> conveniently all the enums would implement. As I said, >> FATAL/ERROR/INFO/WARN/DEBUG/TRACE would implement the interface. >> >> It's worth considering this because I think we're about to pollute log4j >> with logging levels that really don't belong in the public api. These are >> definitely custom things people should implement themselves. >> >> Paul >> >> >> On Wed, Jan 22, 2014 at 12:33 AM, Ralph Goers <[email protected] >> > wrote: >> >>> First, I assume you meant to code “implements LogLevelStrength” instead >>> of “extends LogLevelStrength” since an enum already implicitly extends Enum >>> and a Class (or Enum) can’t extend an Interface. >>> >>> Second, doing this would mean that the Log4j 2 core would have to be >>> modified to never use the Level enum and only use the Interface, except >>> perhaps in ThresholdFilter which can really only be configured with one of >>> the Level enum values. Not being able to use Level as a method parameter >>> and field in the LogEvent makes its value as an enum minimal. Only being >>> able to use Level values in the ThresholdFilter means anyone with a custom >>> Level has to write their own custom Level Filter. >>> >>> I think providing the extra levels is a fair compromise. >>> >>> Ralph >>> >>> >>> On Jan 21, 2014, at 8:50 AM, Paul Benedict <[email protected]> wrote: >>> >>> Or if you really want to get fancy (!!!), don't make the log4j API >>> accept an Level, but an interface that each logging level Enum object >>> implements. Then programmers can use enums. Example: >>> >>> >>> public interface LogLevelStrength { >>> int getStrength(); >>> } >>> >>> public enum Level extends LogLevelStrength { >>> FATAL() { >>> public int getStrength() { return 600; } >>> } >>> ... >>> } >>> >>> public enum MyCustomLevel extends LogLevelStrength { >>> DIAGNOSTIC() { >>> public int getStrength() { return 250; } >>> } >>> ... >>> } >>> >>> >>> >>> On Tue, Jan 21, 2014 at 10:45 AM, Paul Benedict <[email protected]>wrote: >>> >>>> It won't be possible with an enum, yes, but we should have a way to >>>> allow extensions. For example, if we publically document the integer level >>>> of the enums (separated by 100), then we can provide an overload that >>>> accepts an integer. That's how you can allow people to slide in their >>>> extensions. Philospohy: enums for the standard, ints for the custom >>>> programmer. >>>> >>>> Paul >>>> >>>> >>>> On Tue, Jan 21, 2014 at 10:42 AM, Gary Gregory >>>> <[email protected]>wrote: >>>> >>>>> On Tue, Jan 21, 2014 at 11:33 AM, Paul Benedict >>>>> <[email protected]>wrote: >>>>> >>>>>> On Tue, Jan 21, 2014 at 10:25 AM, Ralph Goers < >>>>>> [email protected]> wrote: >>>>>> >>>>>>> >>>>>>> I tend to agree that there is ambiguity between TRACE and VERBOSE, >>>>>>> but I have no problem adding it if it means end users will have more >>>>>>> flexibility with little cost. >>>>>>> >>>>>>> >>>>>> I think this is meaningless flexibility. It smells of adding a >>>>>> feature without a good reason. Imagine the conversations people will have >>>>>> to explain the difference between TRACE and VERBOSE. I can't think of any >>>>>> good universal justification for its use that demands an addition to >>>>>> log4j. >>>>>> >>>>> >>>>> If you do not like it, do not use it ;) >>>>> >>>>> This is best reserved for a personal extension. >>>>>> >>>>> >>>>> Which is not possible since Level is an enum, hence this discussion >>>>> before the API freezes. >>>>> >>>>> Gary >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> -- >>>>> E-Mail: [email protected] | [email protected] >>>>> Java Persistence with Hibernate, Second >>>>> Edition<http://www.manning.com/bauer3/> >>>>> JUnit in Action, Second Edition <http://www.manning.com/tahchiev/> >>>>> Spring Batch in Action <http://www.manning.com/templier/> >>>>> Blog: http://garygregory.wordpress.com >>>>> Home: http://garygregory.com/ >>>>> Tweet! http://twitter.com/GaryGregory >>>>> >>>> >>>> >>>> >>>> -- >>>> Cheers, >>>> Paul >>>> >>> >>> >>> >>> -- >>> Cheers, >>> Paul >>> >>> >>> >> >> >> -- >> Cheers, >> Paul >> >> >> > > > -- > Cheers, > Paul >
