2015-06-12 9:43 GMT+02:00 Cédric Krier <[email protected]>: > On 2015-06-11 21:55, Cédric Krier wrote: > > On 2015-06-11 21:37, Jordi Esteve wrote: > > > On 11/06/15 21:10, Cédric Krier wrote: > > > >>By the way, the default option in the new parameter 'target_search' > for m2o > > > >>fields is 'subquery'. Maybe is it better to set 'join' by default? > If not, > > > >>then the m2o relations suspected to have a lot of records like move > lines, > > > >>invoice lines, sale lines, stock moves, ... should have the 'join' > parameter > > > >>set in the tryton modules? > > > >I will explain why I kept 'subquery' as default. On small database, > > > >there will be almost no difference with the 'join'. On larger > database, > > > >in many cases the query planner could infer the 'join' if it worths it > > > >base on its statistics. But the 'in' join could be worse on specific > > > >case. > > > > > > If on small database there is almost no difference between the two > methods > > > and on larger databases the 'in' joint is worse, I don't see the > advantage > > > to keep the 'subquery' method as the default one. > > > > > > I make the question in another way: Which are the advantages of using > > > 'subquery' method instead of 'join' method? > > > > They are faster if the target table is small. > > And I bet that we have more small table than big one because small table > > are the referential while big table are the operational and I think we > > have much more referential Model than operational (we try to reuse the > > operational). > > I made some tests and indeed joins even on large table are not so > expensive if there is a where clause that reduces a lot the rows to join > on because pg can filter before doing the join. > So I made 'join' the default target_search method.
good job -- Guillem Barba http://www.guillem.alcarrer.net
