On Thu 27 Nov 2003 at 23:37:52 +0100, Rhialto wrote:
> I am running it now under a line-by-line trace as suggested in the
> perlrun(1) page in the section on the -D option. It is still running (it
> is a lot slower this way).

The tail of the output is like this:

       1163:      foreach my $prob (@sorted) {
       1164:        $S *= 1.0 - $prob;
       1165:        $H *= $prob;
       1166:        if ($S < 1e-200) {
       1171:        if ($H < 1e-200) {

many repetitions of the above block, then

       1163:      foreach my $prob (@sorted) {
       1164:        $S *= 1.0 - $prob;
       1165:        $H *= $prob;
       1166:        if ($S < 1e-200) {
       1171:        if ($H < 1e-200) {
       1163:      foreach my $prob (@sorted) {
       1180:      $S = log($S) + $Sexp * LN2;
       entering CODE(0x12188bbb0)
        91:                 *$full_name = sub () { $scalar };
       exited CODE(0x12188bbb0)
       1181:      $H = log($H) + $Hexp * LN2;
       entering CODE(0x12188bbb0)
        91:                 *$full_name = sub () { $scalar };
       exited CODE(0x12188bbb0)
       1183:      my $result;
       1184:      if ($num_clues) {
       1185:        $S = 1.0 - chi2q(-2.0 * $S, 2 * $num_clues);
       entering Mail::SpamAssassin::Bayes::chi2q
        1136:     my ($x2, $v) = @_;
        1138:     die "v must be even in chi2q(x2, v)" if $v & 1;
        1139:     my $m = $x2 / 2.0;
        1140:     my ($sum, $term);
        1141:     $sum = $term = exp(0 - $m);

Floating point exception

$m is apparently 736.285624221884. And indeed, the following also crashes:

$ perl -e 'print exp(0 - 736.285624221884), "\n";'
Floating point exception (core dumped)

This is ok:

$ perl -e 'print exp(0 - 708), "\n";'
3.30755300363841e-308

So it must be underflow that is trapping...

Next question: how to fix that?

-Olaf.
-- 
___ Olaf 'Rhialto' Seibert
\X/ rhialto/at/xs4all.nl        -- Cetero censeo "authored" delendum esse.