On Sunday 12 February 2012 8:10:34 PM David Kuehling wrote:
> >>>>> "Charles" == Charles G Montgomery <[email protected]>
writes:
> > On Sunday 12 February 2012 6:22:20 PM Marcel Hendrix wrote:
> >> What has changed? It appears something is wrong with array
> >> indexing. However, the official Gforth files from the FSL
> >> (http://www.taygeta.com/fsl/scilib.html) are used?
> >> X-Online-Scanned: by Cloudmark authority (on smtp09.online.nl)
> >>
> >> I have append all three files under question.
> >>
> >> Sequences of pi xx{ 1 } f! xx{ 1 } f@ f. seem to work. Putting
> >> it inside a do loop in a definition fails as shown.
> >>
> >> -marcel ....
> >
> > This doesn't seem to happen with just gforth rather than
> > gforth-fast. (Debian Version: 0.7.0+ds1-6 ) I see David Kuehling
> > <[email protected]> finds this too.
>
> Do you say you can reproduce the bug report on debian i386 with
> gforth-fast? For me the bug doesn't happen at all, neither with
> gforth-fast nor gforth. But then I didn't try a i386 version of
> gforth (only amd64 and 32-bit mipsel).
>
> cheers,
>
> David
Stranger and stranger. When I use gforth-fast on amd64 I do not see
Marcel's problem; it runs correctly with either gforth or gforth-fast.
So I tried it on i386. There I get (apparently) correct results with
gforth. But with gforth-fast I get different results for "32 wtest-
init", The xx{ array starts with 195.09 11.8997 17.849 ... instead of
195.09 382.683 555.57 ...
If I continue by running "walsh-test", the transform and its inverse are
consistent.
(The Debian version is 0.7.0+dsl-6 but for the i386 architecture.)
So something is wrong somewhere, but I don't know where. Maybe I can
look further tomorrow.
regards cgm
