Re: [cp-patches] Logger initialization regression fix and a little story on security contexts during class initialization

[Mark Wielaard <mark at klomp dot org>]

Hi Archie,

On Thu, 2006-05-11 at 08:51 -0500, Archie Cobbs wrote:
> Mark Wielaard wrote:
> > We had an interesting regression with the logging code in GNU Classpath.
> > It wasn't immediately apparent because it was only caused with certain
> > compilers. The class initialization order was subtly different between
> > byte code compiled with gcj -C and jikes. So it didn't show up with gcj
> > -C which I used when committing this code.
> 
> Since class initialization is strictly defined by the JVM spec,
> doesn't this necessarily imply a bug in either gcj or jikes?
> 
> The only other alternative is that they both compile the same code
> correctly, but do so differenly enough to change the class initialization
> order, which to me seems even more surprising (but possible I guess.. ?)

Yeah, I didn't really investigate further since the issue was real and
not just a contrived case in Mauve. We need to make sure that if a class
is initialized from a non-trusted environment the static constructor
runs correctly.

Cheers,

Mark

Attachment: signature.asc
Description: This is a digitally signed message part