[Spambayes] Does anyone care about this report?

[T. Alexander Popiel]

In message:  <3AFB3197-8646-11D7-B809-00039345DE8C at email.com>
             Abhay Saxena <ark3 at email.com> writes:
>
>Training into a temporary database and then replacing the live database 
>by renaming over it sounds like a good option. At least on Unix-ish 
>filesystems, aren't we guaranteed that the rename-replace operation is 
>atomic?

Yes, on unix-oids, the rename operation is filesystem-atomic.

>I suppose this line of thought still makes the assumption that the 
>underlying database implementation doesn't close and reopen the file or 
>try to open the file more than once...

Bingo.  Also, in multi-file db implementations, it gets hairier.

>On Wednesday, May 14, 2003, at 12:19  PM, T. Alexander Popiel wrote:
>> (Yes, I deliberately have my procmail set up so that if spambayes is 
>> barfing and/or not labeling messages, they get delivered to my 
>> standard inbox; fail to passthrough is much better in this case than 
>> fail to bitbucket.)
>
>Are you the Alex who wrote the stuff in spambayes/contrib (bulkgraph.py 
>et al)?

Somebody reads my stuff!  Yay!

>If so, does the sample procmailrc in that directory have your 
>trick for doing the right thing if hammiefilter.py barfs?

Yes.  The failsafe nature is not really obvious.  For the first
part, standard procmail holds that if a filter (and I'm using
spambayes as a filter) exits with a non-zero exit code, then the
original mail is retained instead of the filtered version.  For
the second part, I check for specific X-Spambayes-Classification
lines, and if there isn't one present, then the mail goes into
my inbox.

>Thanks for helping me get clued-in, folks.

Not a problem.

- Alex