At 07:11 PM 10/19/2002 +0400, Dmitry Yemanov wrote:

>
>I understand your scepticism, but I was giving you facts, not fantasies.
>IMO, if a few people share with you the same negative experience, probably
>something is really broken here. There was one GC related problem fixed
>recently in the FB2 tree and probably there are others. Unfortunately, one
>little bug may completely brake any clear and reliable design.
>

I agree completely. An unavoidable characteristics of designing
careful write, deadlock free, fault tolerant code is that it is
very unforgiving. Any screwup results in a corrupted database.
The technology is sound, but requires total attention to details,
rigorous testing, and tracking every apparent bug to death. Many
algorithms, particularly update backout (Ann's very favorite
VIO_backout) is on the surface awkward and inefficient (cache
design makes it less so). All it takes a for one bozo to "optimize"
it and the database will never work again.

Jim Starkey