random oddities and questions

[David Relson]

At 01:52 AM 10/8/02, Eric Seppanen wrote:
>What is populate_stats() for?  It's not called anywhere.

It's the internal loop of select_indicators().  At one time 
select_indicators() was factored into a couple of component 
functions.  This has been undone - possibly it was undone in the switch 
from Judy arrays to word hashes.  Anyhow, I'll delete the function.

>Is there a reason we no longer print the top "interesting" words?  This is
>the single most useful debugging tool in there.  It used to appear at
>verbose >= 1 .  Presumably the correct action is to add
>     if (verbose) print_stats(stats);
>after select_indicators() is called.

There's a timing issue here, i.e. where in the header do you want to print 
these statistics.  In my personal copy of bogostats, I call print_stats() 
right after printing the "X-Bogosity" message.  Don't forget that it _does_ 
print 15 lines of statistics which significantly adds to the number of 
header lines and which can be distracting.  FWIW, my code calls it only for 
spam - probably ought to be an option.

>Are there any remaining objections to logging debug (verbose) data
>exclusively to stderr?  There's still a heap of stuff going to stdout,
>which makes a real mess.

Perhaps some of the stdout stuff should be removed, or turned off.

>Multiple wordlists are sort-of working now in my directory.  Just gotta
>figure out why the locking code doesn't lock all the lists.

To ask the simple question, are you setting the active flags and using the 
lock_list functions?  You might want to forward a code diff to Gyepi so he 
can look at how you're using locks.

>I notice that open_wordlist() is being called outside the locking.  Is
>that bad?  I don't know if db_open() and db_init() are doing anything that
>needs locking.  I almost think that locking should take place before
>touching the files at all; an open db handle is an invitation to touch the
>database.

Seems to me that one needs a handle before one can lock a database.  The 
call to db_open() gets the handle.  I suppose it would be possible for 
bogofilter to have a single "bogofilter-lock" rather than use database 
locking.  I don't think it'd be a good thing.