Re: [Maxima-discuss] Careful arithmetic ... was Re: +-Inf and NaN

On Tue, Mar 5, 2024 at 6:05 AM Camm Maguire <[email protected]> wrote:

> Greetings!
>
> Richard Fateman <[email protected]> writes:
>
> > 2. fix sbcl flags so that the handler still works after the first
> floating-point error..
>
> Not sure about sbcl, but this sounds like the standard IEEE SIGFPE
> handling in which it is impossible to continue or return from the
> instruction raising the signal without disabling the trap.  One can
> lonjmp somewhere else leaving the trap in place.  It is not clear if
> this explains your results as it does not appear you are trying to
> continue.  See the following in GCL:
>
>
> =============================================================================
>
> >(fpe::break-on-floating-point-exceptions :floating-point-overflow t)
>
> (:FLOATING-POINT-OVERFLOW)
>
> >(expt 2.0 1000000)
>
> Correctable error:
> Fast links are on: do (si::use-fast-links nil) for debugging
> Signalled by EXPT.
> If continued: Continue disabling floating point exception trapping
> INTERNAL-SIMPLE-FLOATING-POINT-OVERFLOW:
> Arithmetic error when performing (:INSN "mulsd  0x8(%rsp),%xmm0 ;" :OP
> ...) on (1.3407807929942598E154 1.3407807929942598E154):
>
> Broken at EXPT.  Type :H for Help.
>     1 (continue) Continue disabling floating point exception trapping
>     2  Return to top level.
> >>1
>
> #<inf>
>

Is this the current development version of gcl?  What's the difference
between continuing and aborting here?  Does this mean that if I had done (*
10 (expt 2d0 100000)) and selected the continue restart, I would have
gotten #<inf>?
If I had selected option 2, would the traps be preserved?


-- 
Ray