Fixing undefined ia64_double_zero

[Matthias Klose]

On 02.12.2010 17:54, Bob Vandette wrote:
>
> I re-introduced some old IA64 workaround logic with this changeset.
>
> changeset:   1593:126ea7725993
> user:        bobv
> date:        Tue Aug 03 08:13:38 2010 -0400
> summary:     6953477: Increase portability and flexibility of building Hotspot
>
> As I understand, the work-around below is no longer needed.  I'll put together a webrev
> and push a fix but since I have no way of building for IA64 targets, I want to make sure
> that this is the only change causing the IA64 build to fail.  I intend on removing
> everything except line 1292.
>
> - 1288 #ifdef IA64
> - 1289           // IA64 gcc bug
> - 1290           r = ( f == 0.0f ) ? (jdouble) f : (jdouble) f + ia64_double_zero;
> - 1291 #else
> 1292           r = (jdouble) f;
> - 1293 #endif

Afaik, IcedTea did always build without this workaround. I didn't see any 
definition of ia64_double_zero yet.  I checked that it works with a manual 
build. Test results should appear shortly here:
https://buildd.debian.org/status/package.php?p=openjdk-6&suite=experimental

> Gary, have you successfully built all of the Zero targets/architectures since this changeset?
>
> changeset:   1732:3dc12ef8735e
> parent:      1728:dfb38ea7da17
> user:        bobv
> date:        Thu Oct 07 15:12:57 2010 -0400
> summary:     6989297: Integrate additional portability improvements
>
>
> I've tried to make sure that I don't break the Zero builds with my changes but we don't
> validate these targets here at Oracle.

All Zero archs should build, except for alpha and s390.  Hotspot and JDK do 
assume that size_t is unsigned int, while on s390 it is unsigned long.  There 
are patches in IcedTea, which could be forwarded, but usually these need updates 
with new pulls from OpenJDK.

Would it be possible to setup automated cross-build for such targets which break 
with every update?

Thanks, Matthias