I wrote:
> Here's a patch that may fix the bug, but I haven't yet been able to test
> it.
I've now tested this with Adam Majer's test case. I compiled with a
patched jikes and unpatched jikes-gij on ARM, then ran the class file on
an i386. The output is as expected, i.e.:
$ java test
Boltzmann:
I'll test it this weekend.
Thank you!
Adam
Ben Hutchings wrote:
> Here's a patch that may fix the bug, but I haven't yet been able to test
> it.
>
> Ben.
>
> --- jikes-1.22/src/double.h.orig 2004-06-02 12:26:22.0 +0100
> +++ jikes-1.22/src/double.h 2006-10-19 11:38:23.0
Here's a patch that may fix the bug, but I haven't yet been able to test
it.
Ben.
--- jikes-1.22/src/double.h.orig2004-06-02 12:26:22.0 +0100
+++ jikes-1.22/src/double.h 2006-10-19 11:38:23.0 +0100
@@ -63,6 +63,11 @@
#include "platform.h"
#include "long.h"
+// We s
This problem is almost certanly due to the unusual double format used
on arm (which is old ABI, little endian on debian currently).
The two 32-bit words are arranged big-endian, whilst the bytes withint
each word are arranged little-endian. This (odd, but IEEE-754
compliant) format is unique to ar
4 matches
Mail list logo
