Re: RFC: gnu.classpath.SystemProperties

[Michael Koch]

Am Donnerstag, 2. Dezember 2004 16:57 schrieb Jeroen Frijters:
> Hi,
>
> I finally got around to it and implemented my idea to move the
> system properties out of java.lang.System into a public class in
> the gnu.* package.
>
> I'm running IKVM with the attached code with no Mauve regressions
> and it solves *all* known cyclic initialization issues. I built an
> IKVM specific test that basically loops thru all public Classpath
> classes and initializes them (in a new VM for each class) to make
> sure that at least the static initializer for the class doesn't
> cause problems.
>
> Before this patch there were a whole bunch of classes that failed
> this test, but with this patch they all went away.
>
> Note that this doesn't mean there aren't any more dependencies, but
> if they do exist, they will be easier to fix.
>
> Note also that this patch doesn't yet remove all cases were
> System.getProperty() or GetPropertAction() is used. I focused on
> the problem areas first, but I plan to remove the rest as well.
>
> Final note, this a draft version, I'm not looking for nitpicking
> comments, just feedback on whether anyone disagrees with the
> overall approach or has questions or suggestion.
>
> I removed loadLibrary("javalang") from System.java because System
> doesn't have any native methods anymore. If a VM depended on it, it
> should probably be moved elsewhere (one of the VM* classes). It may
> also be necessary to replace some System.loadLibrary() calls with
> Runtime.getRuntime().loadLibrary() to prevent System from being
> initialized too early (and that would also require fixing
> Runtime.loadLibrary() to not call System.mapLibraryName()).
>
> Comments?

Good idea but wouldnt it be better to move all stuff from  
SystemProperties to VMSystemProperties to make the system properties 
fully customizable ? Perhaps some VM implements want to customize the 
properties more, e.g. add additional properties.


Michael
-- 
Homepage: http://www.worldforge.org/