On Jan 31, 2014, at 8:51 PM, Josh Kuhn <[email protected]> wrote:

> This is pretty amazing. I get it to *almost* work. The issue seems to be that 
> I don't use the database column names as my model attribute names. The db 
> column names are really obfuscated, so it's more like:
> 
> class Version(Base):
>     id = Column('vrsn_nbr', Integer, primary_key=True)
>     date = Column('dt', DateTime)
> 
> etc..
> 
> It seems when this relationship is populated, it creates a Version object, 
> but it sets all of the database column names as attributes, and the defined 
> Column names are all None. Is there a way to get around that?

well if you map to a select() then it uses the column names, as in 
http://docs.sqlalchemy.org/en/rel_0_9/orm/mapper_config.html#naming-columns-distinctly-from-attribute-names
 you can rename with properties={“whatever_name”: stmt.c.xyz_column}, etc.  
same thing as saying “id = Column(‘vrsn_nbr’)” on your declarative mapping.




> 
> 
> On Fri, Jan 31, 2014 at 6:38 PM, Michael Bayer <[email protected]> 
> wrote:
> well I can get it to work for lazy loading like this:
> 
> expr = select([func.max(Version.id)]).\
>         where(Version.object_id == Object.id).\
>         correlate_except(Version).as_scalar()
> 
> Object.current_version = relationship(Version,
>                                    primaryjoin=and_(
>                                                     expr == Version.id,
>                                                     Version.object_id == 
> Object.id
>                                                 )
>                                    )
> 
> 
> but for joined load, the criteria needs to fit into a LEFT OUTER JOIN ON 
> clause.  Both SQLite and Postgresql reject an aggregate function in the ON 
> clause.    So I didn’t really know how to get that because you have to think 
> in terms of the SQL….but then the usual approach is that you need to JOIN to 
> a subquery that has the aggregate inside of it.  So I use instead the pattern 
> you see here: 
> http://docs.sqlalchemy.org/en/rel_0_9/orm/tutorial.html#using-subqueries  
> which is the subquery of “X.foo_id, func.AGGREGATE(X.id)” that then joins to 
> the parent table, and then I go with “non primary mapper”, a use case that I 
> recently added to the documentation at 
> http://docs.sqlalchemy.org/en/rel_0_9/orm/relationships.html#relationship-to-non-primary-mapper,
>  even though this has been around for years.
> 
> and it looks like……
> 
> expr = select([
>                 func.max(Version.id).label("max_id"),
>                 Version.object_id
>             ]).group_by(Version.object_id).alias()
> 
> stmt = select([Version]).\
>             select_from(join(Version, expr, Version.id == expr.c.max_id)).\
>             alias()
> 
> current_version = mapper(Version, stmt, non_primary=True)
> 
> Object.current_version = relationship(current_version)
> 
> I think I might have actually written a mapping like this as an example back 
> in SQLAlchemy 0.1 even, this was the goofy kind of thing I thought everyone 
> would be doing all the time.
> 
> works with joinedload.  Query is not too efficient, but is like:
> 
> SELECT objects.id AS objects_id, anon_1.id AS anon_1_id, anon_1.object_id AS 
> anon_1_object_id 
> FROM objects LEFT OUTER JOIN (SELECT versions.id AS id, versions.object_id AS 
> object_id 
> FROM versions JOIN (SELECT max(versions.id) AS max_id, versions.object_id AS 
> object_id 
> FROM versions GROUP BY versions.object_id) AS anon_2 ON versions.id = 
> anon_2.max_id) AS anon_1 ON objects.id = anon_1.object_id 
> WHERE objects.id = %(id_1)s
> 
> 
> 
> 
> On Jan 31, 2014, at 5:35 PM, Josh Kuhn <[email protected]> wrote:
> 
>> I've got a two tables I'd like to create a relationship for. One is the 
>> object, and another tracks versions.
>> 
>> Here's a gist with the setup:
>> https://gist.github.com/deontologician/8744532
>> 
>> Basically, the object doesn't have a direct reference to the current version 
>> stored in the table. Instead, the current version is defined as the maximum 
>> version that points to that object.
>> 
>> I'd like to have a one-to-one "current_version" relationship, but this has 
>> proven difficult (at least in 0.8.4). The primary goal is to allow using the 
>> joinedload options to control populating the current_version field, but that 
>> only works when a relationship is defined and is non-dynamic.
>> 
>> Any hints as to how to get this to work like I want?
>> 
>> -- 
>> You received this message because you are subscribed to the Google Groups 
>> "sqlalchemy" group.
>> To unsubscribe from this group and stop receiving emails from it, send an 
>> email to [email protected].
>> To post to this group, send email to [email protected].
>> Visit this group at http://groups.google.com/group/sqlalchemy.
>> For more options, visit https://groups.google.com/groups/opt_out.
> 
> 
> 
> -- 
> You received this message because you are subscribed to the Google Groups 
> "sqlalchemy" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email to [email protected].
> To post to this group, send email to [email protected].
> Visit this group at http://groups.google.com/group/sqlalchemy.
> For more options, visit https://groups.google.com/groups/opt_out.

Attachment: signature.asc
Description: Message signed with OpenPGP using GPGMail

Reply via email to