Thank you very much for your help. The post_update=True addition is what I
was missing.
The test case I was using was c1.created_by = c1 where created_by was
previously null.
(c1 having previously been saved and flushed) This seems to be the one
corner case where
post_update=True was necessary.
I have attached a script to demonstrate the problem I encountered. I don't
know if this is
expected or known behavior or not, but it was certainly confusing to me. I
would assume that saving
against a null vs non-null value should not result in different behavior.
Thank you again.
-brad
On May 28, 8:47 pm, Michael Bayer <mike...@zzzcomputing.com> wrote:
Here's a script which exercises your mapping against 0.4 - the
remote_side should be on the "many-to-one" side of "updated_by".
Addtionally, depending on what kind of combinations of Contact objects
you want to store, you might run into the scenario where ContactA
references ContactB, and ContactB references ContactA. Theres no way
to INSERT data like that without an UPDATE (assuming FK integrity and
no sequences), so SQLA wants you to use a flag called "post_update" if
that occurs. It needs the flag if the co-dependent items are being
UPDATEd too since it works out dependencies for UPDATE and INSERTs in
the same way...this is something that could perhaps be improved upon.
So the second half of the script illustrates that mapping as an
alternative.
cyclical_contacts.py
3KDownload
6 messages in com.googlegroups.sqlalchemy[sqlalchemy] Re: Column name override...
.py