Floating point errors?

[Matthias Andree]

Ingomar Wesp schrieb:

> This is exactly what I was to propose. Interestingly, the current set of 
> actions resulted from a debate [1] about a change in bogofilter's options 
> back in 2004. The people, who were involved back then, realised the fact that 
> using the "unregister"-options together with unclassified messages might be 
> problematic, but did so nonetheless. Maybe they were not entirely sure about 
> the severity of the possible consequences.

Too bad they were on the right track and drew the wrong conclusions WRT the
-u option. It's very costly, for questionable gains.

>> It would also speed things up a bit as the "check" no longer involves
>> costly synchronous writes and keep the database size smaller.
> 
> I realized that. It makes a quite a big difference on my system.

Yes, kernels have finally started to get more picky about using hardware
and caches on disk drives properly so as be able to provide the "D" in
"ACID" features in databases. (That's nothing about corrosive materials,
but atomicity, consistency, isolation and durability guarantees. DB lingo.)

Best regards,
Matthias