> I would too like to see the BerkeleyDB bayes backend to be sorted out > before the release. I made a couple of minor attempts to deal with the > more obvious issues, but it would be nice to get some help from > Michael, its author (or some other interested party), and some > experience from a production use. Bug 6046.
Sorry, I've been swamped by work for the last couple of months. I haven't had time to pull in your changes from a month or two ago, which I want to look at, but I have done some work independently. The solution to the consistency problems you were running into is, as far as I can discern, to use transactions---which, to my eye, would be perfectly fine: it slows things down a little bit, but it's still faster than the full-on SQL back-ends. Unfortunately, when I do that, I then have problems with BDB complaining about lack of available transaction slots...even when there's only ever been one or two transactions in play at one time. (Oh, and you can only adjust the number of transaction slots using a DB_CONFIG file, which is a whole other annoyance---that's just part of the BDB API that isn't exposed in the perl module.) Perhaps the combination of your changes and mine will make some forward progress. I'm going to be very busy for the next couple of weeks, but maybe I can carve out some time to reconcile our changes and start up some tests. Mike.
