locking [was: change in registration options]

[Matthias Andree]

On Mon, 03 Mar 2003, David Relson wrote:

> I'll check the NEWS and CHANGES files to see if they need updating (unless 
> you'd rather do that)

Go ahead.

One thing that came to me while falling asleep two days ago (I've had a
busy week-end, it was my Mom's birthday and my girl-friend is preparing
to move), is: we can lose virtually any amount of data by a premature
exit, crash, abort, whatever: if DB can't update its structures, and a
premature exit prevents it from flushing its cache (i. e. it crashes
before completion of its sync operation), we're heading for corruption.

We should see after 0.11 if using some other means (transactional store
or PostgreSQL) is acceptable performance-wise, installation effort wise
and if it is more "survivable". Ideally, our users wouldn't have to
rebuild the data base.

I wonder if we should provide a framework to backup textual
representations of the data base and write a journal since the last
checkpoint.

-- 
Matthias Andree