Re: Patch: java class registration via .jcr section

["Anthony Green" <green at redhat dot com>]

Tom wrote:
> I think we'd prefer to use _Jv_RegisterClasses here.  For one thing
> that saves one synchronization per class.  If this means changing
> _Jv_RegisterClasses to accept an `int length' argument, that is fine
> by me.

That won't work.  This function lives in crtend.o which only has access to
__JCR_END__.  Everything fails in strange ways if I make __JCR_LIST__
non-static and refer to it for a call to _Jv_RegisterClasses in crtend.o.
Maybe it has something to do with linking multiple shared libraries and the
main program together, all providing the global __JCR_LIST__.

Perhaps we could make a new function like _Jv_RegisterClasses in libgcj
which works backwards from __JCR_END__.

AG