On Fri, Aug 12, 2016 at 01:32:33PM +0200, Chris Travers wrote:
>>> My preference is stored procedures plus service locators >> >> Would you care to elaborate a little on the latter (service locators) ? >> > > Sure. What I prefer to do is to allow for a (cacheable) lookup on the > basis of some criteria, either: > 1. Function name or > 2. Function name and first argument type > > This assumes that whichever discovery criteria you are using leads to > uniquely identifying a function. > > Then from the argument list, I know the names and types of the arguments, > and the service locator can map them in. This means: > > 1. You can expose an API which calls arguments by name rather than just > position, and > 2. You can add arguments of different types without breaking things as > long as it is agreed that unknown arguments are passed in as NULL.
Maybe I am a bit dense. Can you please give an example ?
Ok. Two ways of doing this based on different discovery criteria.. The first would be:
CREATE FUNCTION person_save(in_id int, in_first_name text, in_last_name text, in_date_of_birth date)
RETURNS person LANGUAGE ... as $$ ... $$;
Then you have a service locator that says "I have a person object and want to call person_save." It then looks up the function argument names and calls it something like this:
The second approach is to tie to the first argument type (think 'self' in Python).
In this case, we'd have a function defined like this:
CREATE FUNCTION save(person) RETURNS person LANGUAGE ... AS $$ ...$$;
Then we have a different service locator that maps this to the safe function as:
SELECT * FROM save(?::person);
with a argument that is basically:
serialize_to_record_form($object)
Of course that's just the start. To make this really usable you have to add some additional functionality but that should be enough to describe the process.