Moving to -hackers.
> Unfortunately, having all my users run contrib/adddepend isn't an
> option for me. However, that script does contain a good deal of
> information that I may be able to use for detecting old-style foreign
> key constraints in my own code.
I assume you're doing the database upgrade for them or providing
instructions? Could this be a mandatory portion of that process?
> Okay, more questions:
> I see that adddepend detects old-style foreign key constraints by
> looking for groups of 3 triggers having 6 or more identical function
> arguments. Is that the best way to do it? It occurs to me that an
> alternative might be to find triggers that call RI_FKey_check_ins()
> and have the tgisconstraint flag set. Will either approach be safe in
> postgres 7.4? Perhaps a combination of the two would be best?
Yes, a combination of the two would probably be better. You would need
to be careful about function call names for FKeys, there are a fair
number of them. Checking for 3 triggers with the function name starting
with RI_FKey* would probably be better.
That said, I've not heard of any issues with the current implementation
of adddepend, which is also used in a few other well used programs.
--
Rod Taylor <rbt@rbt.ca>
PGP Key: http://www.rbt.ca/rbtpub.asc