In general I agree with your points. However, I don't think we can get rid of all the reflection. ThrowableProxy specifically includes support for the enhancements introduced in Java 7. I'm not in favor of setting that to the minimum version as my employer hasn't moved to that yet and isn't even talking about it yet. However, supporting the new "suppressed" data was something I thought was important for users so I added via reflection. The difference is that I do periodically build and test with Java 7.
Ralph On Feb 26, 2013, at 1:17 PM, Jacob Kjome wrote: > > Ok, I now understand your point about using "-source 1.6" for documentation > purposes while keeping "-target 1.5" for Java5 compatibility. If Java5 is > what we decide upon to target, and we are intent upon choosing Java > version-specific implementations at runtime, I agree that we should > definitely set "-source 1.6" at a minimum. > > Nevertheless, using a later JDK to build than you are targeting opens up the > possibility of unintended Java6 API references leaking into the source. I > realize that's exactly what you intended in this case, and use reflection to > choose the appropriate implementation for the runtime Java version. But > someone else might check in code that doesn't have such a check. Or you > might check it in yourself on accident. And the release coordinator would > never detect it at build time. Only if we ran tests under Java5 might we > catch it at runtime. More likely, we'd have to wait for some user to report > it. > > Again, I'm not against targeting Java6 if there's a compelling reason to do > so. I'm against generating release artifacts using a later JDK than we are > targeting. > > Bottom line: get rid of the reflection. Either the performance improvement > of the Java6-specific getBytes(Charset) (along with other Java6-specific API) > is compelling enough to use it or we just deal with the Java5 > getBytes(String). The source/target values should be identical and release > Java build version should match the source/target or we risk a brittle build. > > > Jake > > On Tue, 26 Feb 2013 11:48:43 -0800 > Ralph Goers <ralph.go...@dslextreme.com> wrote: >> No - I am actually using the Java 6 compiler and it is allowing >> AbstractStringLayout to reference a method that is in Java 6 but not Java 5. >> The code is actually testing if the method exists so it won't get an error >> when running in Java 5 but will use a less efficient method instead. By >> specifying the source we are documenting that Java 6 is the minimum compiler >> version we support. >> If specifying Java 6 for both source and target is what the majority thinks >> is best I am OK with that. I actually have no pressing need to support Java >> 5 for my own purposes. >> Ralph >> On Feb 26, 2013, at 11:26 AM, Jacob Kjome wrote: >>> All "-source 1.6" does above and beyond 1.5 is allow for using "@Override" >>> on overridden interface methods, which is nice but not critical; certainly >>> no reason to switch to Java 6. >>> What you actually seem to be proposing to use "-target 1.6" because you >>> wish to bind to 1.6+ API. This will both prevent compilation with any JDK >>> less than 1.6 and allow for creating 1.6 compatible binaries under later >>> JDKs. >>> If there is API in 1.6 that we simply can't live without to produce a >>> quality logging implementation, then go for it, since 1.5 is pretty >>> outdated. But if it's just minor stuff that could be worked around using >>> existing 1.5 API, there are plenty of users stuck with JDK 1.5 that will be >>> left out of using Log4j2 for no good reason. >>> In any case, the bug is primarily in our release process rather than code. >>> It is due to the fact that we are generating official release artifacts >>> using a JDK (6 or 7) later than the specified target JDK (5). While javac >>> source/target allows for a flexible build system by not forcing everyday >>> users to install an older JDK in order to generate artifacts compatible >>> with documented target JDK, project release managers should >>> *****always***** use the documented target JDK to generate official release >>> artifacts to avoid accidentally introducing bindings to API that exists in >>> later JDKs, but not the documented target JDK that the project claims to be >>> compatible with. Had this been done, this issue would have been caught >>> long before any official release. >>> So, decide on a JDK to target and then actually use that JDK to generate >>> official release artifacts, rather than some arbitrary later one, and we'll >>> never run into this issue again. >>> Jake >>> On Mon, 25 Feb 2013 23:24:50 -0800 >>> Ralph Goers <ralph.go...@dslextreme.com> wrote: >>>> Remko found a bug in AbstractStringLayout where the code actually requires >>>> Java 6 to compile although it should run fine with Java 5. Does anyone >>>> have a problem with setting the compile source setting to 1.6 for Log4j 2? >>>> That is actually what I have been building all the releases with anyway. >>>> Ralph >>>> --------------------------------------------------------------------- >>>> To unsubscribe, e-mail: log4j-dev-unsubscr...@logging.apache.org >>>> For additional commands, e-mail: log4j-dev-h...@logging.apache.org >>> --------------------------------------------------------------------- >>> To unsubscribe, e-mail: log4j-dev-unsubscr...@logging.apache.org >>> For additional commands, e-mail: log4j-dev-h...@logging.apache.org >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: log4j-dev-unsubscr...@logging.apache.org >> For additional commands, e-mail: log4j-dev-h...@logging.apache.org > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: log4j-dev-unsubscr...@logging.apache.org > For additional commands, e-mail: log4j-dev-h...@logging.apache.org > --------------------------------------------------------------------- To unsubscribe, e-mail: log4j-dev-unsubscr...@logging.apache.org For additional commands, e-mail: log4j-dev-h...@logging.apache.org
