On Tue, Aug 12, 2008 at 12:26 AM, Andrus Adamchik <[EMAIL PROTECTED]> wrote: > > On Aug 9, 2008, at 12:18 PM, Øyvind Harboe wrote: > >> I guess the problem I see is that I test only against one or two databases >> so I'd rather see the *same* behaviour across databases regardless of what >> JDBC is defined to do... > > I understand. > >> We flipped scale="2" for all our NUMERIC types >> since that is what our application assumes. > > I think this is the "correct solution" as it explicitly tells the driver > what to do leaving no room for ambiguity. I am still unsure about what the > Cayenne default behavior should be (when there's no scale set).
Good question indeed. Perhaps this is just a bug in Derby? Shouldn't scale -1 mean infinite precision for BigDecimal? I *could* add a BigDecimal extended type to handle this in a way that follows the *applications* rules, right? (whatever those rules might be). Make scale=0 default? Forces user to choose? Perhaps add a file to cayenne which lists known ideosynchrazies across databases? Keep it under Subversion such that it is in sync w/source code from a historical point of view, a javadoc comment would do... That would only document it, not necessarily help the developers catch the problem up front(as that would require reading documentation and everybody knows how well that works...) I can't think of anything that is a 30% improvement over the current state of affairs, so perhaps doing nothing is just as well... -- Øyvind Harboe http://www.zylin.com/zy1000.html ARM7 ARM9 XScale Cortex JTAG debugger and flash programmer
