kawa numerics unstable

[alex mitchell]

Thanks for the heads-up on this. Yes, the latest version does indeed break
some of my code...

Would the best strategy be to stick with an earlier rev (6365 seems to be ok
for most of my code) and wait for things to settle down?

By the way, I already noticed some performance problems as a result of
recent changes (something between 6330 and 6365), but interestingly only on
MacOS, and only in some 2D graphics code. Any suggestions for tools that I
could use to profile my code? It seems that tools such as Shark don't
support 64-bit profiling, and Java 2 SE is only available as 64-bit on MacOS
10.5.x. I'm not sure if Instruments supports Java profiling in the same way
that Shark does, and I'm not sure if upgrading to Snow Leopard, where
everything is 64-bit, will give me the tools that I need.

Any Mac users out there with any suggestions? :)

thanks!
Alex


On 9/15/09 4:20 PM, "Per Bothner" <per@bothner.com> wrote:

> I'm making some changes to Kawa numerics that make
> break some code and cause other code to be slower.
> One goal is to generalize Kawa arithmetic so it
> handles the standard java.lang.Number classes in
> addition gnu.math.*.  Another is to optimize and
> inline more functions.  On change I just checked in
> is that when ant 'int' is convert to an Object, we
> get a java.lang.Integer, not a gnu.math.IntNum
> This should make Kawa a little more consistent with
> Java in general, including reflection and varargs.
> 
> This is related to my attempt at writing Kawa ports of
> the benchmark programs at http://shootout.alioth.debian.org/
> with the goal of showing off Kawa's performance more widely.