Damon wrote: > > Thank you very much for the explanation. It is what I feared was the > case. > > One of the great features we love about SA is the mappers, allowing us > to define table relationships in such a way that we can decide what > table(s) around which to "pivot", giving us different ways of > returning data even when processed from the same query. It seemed to > us that if the mappers are able to traverse all the joins necessary to > render the mapped objects -- we greatly admire SA's ability to > construct all the outer joins required to do this in one fell swoop -- > that it should also be possible to have SA follow similar logic to > construct query objects as well -- in a completely analogous fasion -- > when supplied with filters. > > Alas that this is not the case. :(
you have to realize in all of those cases, the mappers have been given by you the exact paths which it is to join on - it's never just "looking" for something and picking the first match. The "outer join" formulation (i.e. via eager loading) is present since you've placed "lazy=False" on those relations. The key practice here is requiring explicit statement of all behaviors, and we try to stick to that pretty often unless a behavior has absolutely zero chance of being surprising or appearing inconsistent with the steps required in more complex scenarios. Specifically with query(A).join(B) working "automatically", if your code were greatly dependent on this, and you then someday added a second relation() between A and B, all your existing code would break. > On Sep 3, 12:29 pm, "Michael Bayer" <mike...@zzzcomputing.com> wrote: >> Damon wrote: >> >> >> > MUST we explicitly supply the join to such query objects? Or is >> there >> >> > some way that SA can figure out that tbl_people_documents is in >> >> > between tbl_people and tbl_documents on its own? Perhaps there is >> >> > something we can add to the tbl_people/tbl_documents object >> >> > definitions that clues SA in? >> >> >> join on the relation. >> >> >> query(A).join(A.relation_to_b).filter(B.foo == 'bar') >> >> > The problem with that, from what we're trying to build, is that we >> > have to explicitly know that relation object and supply it. >> >> > We want SA to *infer* the relationship between any two tables based on >> > the ORM relationships that we have already defined in our mapper >> > objects. >> >> but you're asking for it to infer the join between *three* tables - i.e. >> your association table. The current SQLA functionality is that >> ORM-level >> joins, that is joins which occur due to the presence of a relation(), >> must >> be expressed explicitly in terms of the relation between the two entity >> classes. Right now only a SQL level join, that is joins which occur >> due to the presence of a known foreign key between the two tables, is >> what >> happens if you don't specify the relation() you'd like to join on. >> >> The proposed enhancement would require that we change the method used >> when >> someone joins from A to B using query.join(), in that it would >> specifically search for ORM-level relations, instead of relying upon >> SQL-level joining which searches only for foreign keys between the two >> tables. It would also throw an error if there were any ambiguity >> involved. I'm not 100% sure but I think it's quite possible that we >> had >> such a "assume the only relation() in use" feature a long time ago when >> constructing joins, and it was removed in favor of explicitness, but I'd >> have to dig through 0.3 functionality to see if that was the case. >> >> My initial take on this feature is -1 on this since I don't think being >> explicit about an ORM relation is burdensome or a bad idea (plus we >> might >> have already made this decision a long time ago). We might just need >> some >> better error messages when a join can't be found between "A" and "B" to >> suggest that its only looking for immediate foreign keys in that case, >> not >> ORM relations. >> >> Alternatively, SQL-expression level join() would "search" for any number >> of paths from table A to table B between any other tables that may >> create >> a path between them. that would also find the association table between >> A >> and B and create a longer series of joins without ORM involvement. I'm >> strongly -1 on such a feature as the expression language shouldn't be >> tasked with performing expensive graph traversals just to formulate a >> SQL >> query, and table.join()'s contract is that it produces a JOIN between >> only >> two tables, not a string of joins. > > > --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "sqlalchemy" group. To post to this group, send email to sqlalchemy@googlegroups.com To unsubscribe from this group, send email to sqlalchemy+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sqlalchemy?hl=en -~----------~----~----~----~------~----~------~--~---
