Re: Patch for ARM java: gcjh

[Scott Bambrough <scottb at netwinder dot org>]

Tom Tromey wrote:
> 
> gcjh runs on the host, not the target.
> So we should be looking at HOST_FLOAT_WORDS_BIG_ENDIAN.
> Also I think we should use "#if" and not "#ifdef" (like real.h).

Ok.  I'll change that.  It works in my case as host = target =
armv4l-unknown-linux-gnu.

> Finally your patch uses "else" and not "#else".

Oops.  I used CVS and my gcc-2.96 tree to generate the patch.  I've been testing
with my 2.95.2 rpm.  I don't have a tree to compare to as readily available. 
Sorry.

> It would be cool if gjavah.c used union DWord instead of its own
> union.

I didn't make a change in this area.  I agree however, and can make the change
if you want.

>  It would also be nice to change the code part instead of
> changing the infinity define.

I don't know about that.  I considered that but I thought the code was more
readable if the define changes.  I also think it is clearer what is happening
and why if the define is changed.  Just my opinion, I'll change it if you want.

> Alex will have to look at the other part.  I imagine the same
> considerations apply.

Yes.  The WORDS_TO_DOUBLE fix will be required only if running on an ARM host. 
I'll fix my patch this weekend and wait for his comments before I resubmit it.

Scott

-- 
Scott Bambrough - Software Engineer
REBEL.COM    http://www.rebel.com
NetWinder    http://www.netwinder.org