Jennifer Spencer <jennifer.spencer at stanford.edu> writes:
> Hi, and thanks for the response.  I sent this months ago so it's nice to see it!
>
> We don't use pg_dump to back up.  We use SELECT
> pg_start_backup('label'); to track the WAL files, and we tar up all
> our data files and put those onto tape (using a logical volume
> snapshot to keep the backup consistent).
>
> If I needed an exact table output as a backup instead of files, I
> would use "copy out" in binary mode (which it has) for my tables.  I
> don't know that that is fundamentally better than text, when I really
> get to thinking about it, past a certain decimal place.  My grasp of
> binary versus text here is not what I'd like, but maybe you guys can
> shed more light on this?  How much accuracy are we really getting with
> text copying?

I hate to be evasive, but I'll have to evade this, as I don't know the
answer ;-).

My usual practices involve using NUMERIC, as I'm usually working with
financial figures, where I don't want any floating point rounding
errors, so I wouldn't see this scenario.

As I mentioned, the issue is really the same as would be the case with
pg_dump; I'll forward the question over to pgsql-general, where others
may hve the relevant involvement with FP math.
-- 
output = reverse("gro.mca" "@" "enworbbc")
http://www3.sympatico.ca/cbbrowne/x.html
"There are no threads in a.b.p.erotica, so there's no  gain in using a
threaded news reader."  -- (Unknown source)