Christophe Lizzi wrote:
> > I'm thinking of getting rid of stubs. if someone can't convince me not
> > to, they'll probably die in the next few days.
>
> If you remove stubs, japhar will be restricted to the targets supported
> by libffi.
yeah. but I think libffi already supports the most prevalent architectures
in use... we need hppa, but that's about the last really big architecture
that needs supporting.
> Anyway, it is possible to remove stubs, and to use libffi even if static
> linking is required (i.e. requested by the user, or if shared libs are not
> supported by the OS), but you have to retain the symbol mappings, and
> make them point directly to the native function, and not to the
> (no longer existing) stub.
Yeah, that's true. We could also use the JNI RegisterNatives call for
things, but it amounts to the same amount of work.
> I think the main issues are not induced by the stubs, as they can be
> generated, but by mappings, as they have to be manually kept up to date
> when new code is added to libnative.
true.
Chris
