>>>> (2) does 6.4 offer stability improvements over 6.3.2?
> Yes.
FWIW, 6.4 is noticeably more stable than 6.3.2 in my company's
application involving concurrent users of a shared database.
We have not seen a backend crash or data corruption since installing
a pre-alpha-6.4 server in mid-September. We had several such problems
in the preceding couple of months with 6.3.2.
>>>> (3) does 6.4 support query lengths > 8192, or data blocks > 8192
>>>> (other than large objects)?
> Sometime in the past Darren K. worked to parameterize this limit.
There has been some discussion of allowing tuples to span multiple
disk blocks, which would remove the problem entirely, but it hasn't
happened yet. Maybe for 6.5?
The limit on the textual length of a query is an unrelated quantity
that by coincidence has the same value. (Well, maybe not total
coincidence... probably someone wanted to be sure they could INSERT
an 8K text string... but the code doesn't know there's a connection.)
I am planning to modify libpq and the backend to eliminate fixed-size
query text buffers, so this limit should go away for 6.5.
regards, tom lane