Re: [cp-patches] RFC: replacing jni_md-x86-linux.h by a generic jni_md.h

[Dalibor Topic]

Tom Tromey wrote:
> > > > > > "Dalibor" == Dalibor Topic <address@hidden> writes:
>
>
> Dalibor> the attached patch replaces the os/cpu/toolchain specific
> Dalibor> jni-md.h by a generic version that uses fixed size C99's
> Dalibor> types to typedef java types. It's been successfully used in
> Dalibor> Kaffe for a while.
>
> I think there are 2 problems with this patch.
>
> Dalibor> +#include "config-int.h"
>
> First, we need jni_md.h to be installed.  This means that config-int.h
> must be installed ... but it isn't.  

Thanks for spotting that bug, Tom, I'll post an updated version of the 
patch.

Kaffe indeed installs config-int.h along with jni_md.h. I think the file 
should be renamed then to something like classpath_stdint_wrapper.h to 
make sure it differentiates itself from other headers.

> And, simply installing it may be
> wrong as it could redefine things like int8_t or whatever.

If the macro works as it should (delegating to stdint.h/inttypes.h when
those are available), I don't think that is going to be an issue. Could 
you elaborate?

> Dalibor> +/* Define some defaults */
> Dalibor> +#define JNICALL
> Dalibor> +#define JNIEXPORT
> Dalibor> +#define JNIIMPORT
>
> These aren't correct for Windows.  The old approach offered the
> possibility of defining these correctly, at least in theory :-).

That's right. I've got the old version of them in Kaffe's version the 
file which defined them for Windows. ;)

I am not quite sure what the best approach is here. Would #including 
"jni_md_vm.h", which would be empty per default, and having all JNI* 
macros be defined conditionally upon them being #if ! defined be a 
workeable solution, as it would give those runtimes that need to 
redefine JNI* an entry point?

cheers,
dalibor topic