On Fri, May 02, 2008 at 12:27:38PM +0100, Tom Hughes wrote: > > I won't pretend that I know nearly as much about the rails code as you > > do, but it seems like some of these would be better abstracted out. If > > that were the case -- that is, that all the Rails code on the site used > > the same underlying methods to talk to the database, given a 'fixed' > > API, and amf_controller was just about encoding the returned data into > > AMF -- then I thiink it would be possible to change the underlying slow > > methods into SQL (after proper profiling), because the main reason not > > to is 'maintaining two different codebases sucks', rather than 'no one > > likes SQL over rails'. > > If I thought Steve would let me get away with doing raw SQL instead > of using the rails object model I might have done so long ago ;-) > > Doing so would bypass all the integrity checks though, which is > bad - that's a side effect of having the integrity checks in the > wrong place (the rails object model rather than the database).
Some things don't require referential integreity: selecting ways/nodes within a bounding box can't hurt the referential integrity of the database (so long as the code is well-maintained), so the harm in converting those methods (which are probably the single most performance important aspect of Potlatch?) to SQL is relatively low, so far as I can tell... Regards, -- Christopher Schmidt MetaCarta _______________________________________________ talk mailing list talk@openstreetmap.org http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/talk