Comment 6 for bug 1304261

I can't be absolutely sure why this is happening, because allocation crashes like this often show up somewhere after the point where things started to go wrong. However, the way man-db keeps track of its database handle is asking for trouble: it's all done by way of a big global variable, which might be OK if it were always zeroed after closing but it isn't, so there are probably paths where it double-closes the handle and sometimes crashes. I think I'll rearrange all that to use sensibly-scoped local variables instead, which should get rid of this kind of bug.