nan [was: grftest failure]

[David Relson]

At 01:21 PM 1/5/03, Clint Adams wrote:

> > Using -O0, t.grftest passes when not run from make, but fails in make
> > check.  Now the only differences from CVS are configure.in and the sed
> > expression in t.frame.
>
>Same deal for -O1.

Clint,

Now that I've started to think about compaq and risc, I'm remembering 
Thanksgiving.  The regression tests had just been released and Allyn was 
having trouble with HP-UX on risc.  The problem was floating point roundoff 
(register vs. memory) in graham.c with the following fix:

-           if (dev>slotdev && hitdev>slotdev)
+           if (dev-slotdev>EPS && hitdev-slotdev>EPS)

Given that history, here's a shot in the dark for you to try:

Index: robinson.c
===================================================================
RCS file: /cvsroot/bogofilter/bogofilter/robinson.c,v
retrieving revision 1.22.2.3
diff -r1.22.2.3 robinson.c
220c220
<         if (fabs(EVEN_ODDS - prob) >= min_dev) {
---
 >         if (fabs(EVEN_ODDS - prob) - min_dev >= EPS) {
-------------- next part --------------
A non-text attachment was scrubbed...
Name: robinson.c.patch
Type: application/octet-stream
Size: 316 bytes
Desc: not available
URL: <https://www.bogofilter.org/pipermail/bogofilter-dev/attachments/20030105/7f4f01e8/attachment.obj>