Re: Release Plans

[Philip Nienhuis]

John W. Eaton wrote:
> On 04/23/2013 09:14 AM, Philip Nienhuis wrote:
>
> > Yes you're right, I remember the FPU flags issue. But was that an Octave
> > bug (not properly preserving & resetting those flags) or a Java bug?
> > It's a bit hard to believe that other SW developers never hit this
> > issue, as Java is so immensely widely used.
>
> I wasn't aware that Octave is setting FPU flags to some non-default
> values except on BSD systems where it was apparently required to avoid
> trapping when generating Inf and NaN values.

Yeah, we're talking about FPU flags/CPU flags because I started 
mentioning that in my previous post. I was inspired by my memory of 
posts mentioning FPU/CPU flags, but I can't find them anymore.

But even if flags are involved, I think we'd rather say "Octave incl. 
the Java subsystem".
I don't know if Octave sets flags; perhaps dependency libs do; but as 
Mike mentioned there is obviously some interference, only noted after 
Java got moved to core Octave.

> So I think it would be unusual to have to save and restore FPU flags
> before and after calling some library routine. Maybe the library
> routines should be taking care of that detail?

From our POV that would be obvious. But maybe not from the POV of the 
developers of that library.

> I suspect this usually goes unnoticed because people don't care. As I
> recall, we only noticed because of small (on the order of the machine
> precision) changes in some test results. If we hadn't had the previous
> values, I don't think that we would have noticed either.

AFAIK Mike hints to http://savannah.gnu.org/bugs/?37863 (still open); 
however this seems to be related to readline (?)

http://savannah.gnu.org/bugs/?func=detailitem&item_id=37891 turned out 
to be a bug exposed due to a different memory layout caused by loading 
the JVM.

The third one (Mike mentioned three issues) I can't find.

So all in all, ignoring Mike's 3rd issue, I think it remains doubtful 
whether it really is the JVM that is the cause. Maybe there are 
underlying bugs (in Octave and/or the dependencies) that went unnoticed 
until recently and the JVM is just the last straw that breaks the 
camel's back.

Philip