On 16 May 2000 22:58:45 CDT, the world broke into rejoicing as
John Hasler <[EMAIL PROTECTED]> said:
> Rob Browning writes:
> > Anyway, this is the motivation for the current discussion. Hope it makes
> > my motivation at least somewhat clearer, even if you're not convinced.
>
> I'm convinced. I'd prefer to have the journal entries in a fixed record
> length append-only text file, but I know I'll never talk anyone into that.
That is something that CBB did, at my behest, back in the mid'90s,
with regard to logging changes.
(I actually once went through and replayed the log through the engine
to reconstruct a data file; it most definitely worked.)
Feel free to argue for there to be a "monotonically increasing"
text file that contains _logs_ of the changes. I'd argue the same;
it can go a _LONG_ way towards establishing peoples' confidence in
the system.
Scenario:
"Ah. We can watch exactly what's going on, just by running
tail -f ~/finances/gnucash.log
And if something goes bad, I can make a copy of that file, take out
the offending lines that made GnuCash crash, and load it back in,
thus meaning we lost _no_ work..."
Note that by doing this, it no longer matters if the "database" data
gets stored in some bizarro database format only readable by GnuCash;
all we need do is play with the transaction logs, and we can fix the
contents as needed.
--
`I am convinced that interactive systems will never displace batch
systems for many applications.' - Brooks, _The Mythical Man-Month_
(And this does indeed seem true. MVS/CICS systems have *NOT* gone
away...)
[EMAIL PROTECTED] - <http://www.ntlug.org/~cbbrowne/lsf.html>
