[wildfly-dev] Heads up on JBoss Logger 3.2 : more changed than what it looks like

[James R. Perkins]

We faced some odd compile errors with JDK 7 when we switched WildFly 
Core to 3.2.1.Final. Really the only way to get around it would be to 
cast int's or long's to their object types. Though that's not really 
ideal either I realize.

That said it should be fine in WildFly 9 as we're using 3.2.1.Final now.

On 02/27/2015 12:08 PM, Sanne Grinovero wrote:
> Hi all,
> today I've upgraded jboss-logging from version 3.1.4.GA to 3.2.1.Final
> in an Hibernate project, and got some integration test failures.
>
> The surprising aspect is that everything seemed fine at compile level:
> just switch the version in the pom, and without needing any further
> changes it compiles fine and runs all unit tests successfully.. but
> fails when using the freshly built libraries in WildFly 8.2.
>
> It's not a regression of jboss-logging, but one of its improvements
> require a bit of attention.
>
> This is what happened to us:
> we have some log statements which look like this in source code (after
> simplifying):
>
>      int i = ...
>      log.debugf( "Number: %d", i );
>
> This does of course compile fine in both old an new versions of JBoss
> Logger. And it all works as expected in unit tests.
> But when deploying the modified version of this Hibernate library in
> WildFly 8.2, you'd get some of these:
>   - java.lang.NoSuchMethodError:
> org.hibernate.search.util.logging.impl.Log.debugf(Ljava/lang/String;J)V"}}
>
> When using the older version of JBoss Logger (at compile time), the
> above line of code is compiled as an invocation to:
>
>      void debugf(String format, Object param1);
>
> (which is a method present in both versions)
>
> When using the new version of JBoss Logger (at compile time), the same
> line of code is interpreted as the (better) invocation to:
>
>      void debugf(String format, int arg);
>
> So that's what the library is going to invoke at runtime - and that
> method doesn't exist in WildFly 8.2.
>
> Be aware of this when upgrading as it might look like a trivial
> version bump but it makes it quite hard to guarantee backwards
> compatibility with older versions of the logger.
>
> Personally since I want to upgrade the logger but don't want to drop
> compatibility with existing WildFly releases, I'm trying to figure how
> to reliably validate that no logging statement is going to invoke one
> of the new ones.. for now.
>
> I guess this also means that users won't actually benefit from the
> better performance of the new logging methods until we recompile all
> of its client libraries using the new version ;-)
> Auto-boxing is evil..
>
> Sanne
> _______________________________________________
> wildfly-dev mailing list
> wildfly-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/wildfly-dev

-- 
James R. Perkins
JBoss by Red Hat