pre-init-spec for composite keys

[weavejester]

pre-init-spec for composite keys

[weavejester]

Composite keys are just keys that inherit from all of their keywords. If we wanted some way of combining specs, then we can't use multimethods, as inheritance doesn't combine.

I've actually written a library for combined inheritance, but whether this is the best way of handling things for Integrant I'm not certain, as it introduces a new dependency and new syntax.

[weavejester]

Your example doesn't appear to match up to the rest of the issue. If you have a configuration:

{[:group/a :adapter/jetty] {:port 8080, :handler #ig/ref [:group/a :handler/greet]}
 [:group/a :handler/greet] {:name "Alice"}
 [:group/b :adapter/jetty] {:port 8081, :handler #ig/ref [:group/b :handler/greet]}
 [:group/b :handler/greet] {:name "Bob"}}

And specs:

(defmethod ig/pre-init-spec :adapter/jetty [_]
  (s/keys :req-un [::port ::handler]))

(defmethod ig/pre-init-spec :handler/greet [_]
  (s/keys :req-un [::name]))

Then it should work fine; I just tried it. It's only when you have multiple specs for the same keyword that the inheritance is forced to pick one.

[julienfantin]

@weavejester I'm not sure I understand because the two defmethods you quoted above never get invoked for that config. pre-init-spec dispatches on the four composite keys.

[julienfantin]

Composite keys are just keys that inherit from all of their keywords. If we wanted some way of combining specs, then we can't use multimethods, as inheritance doesn't combine.

I think I understand that constraint better now. Even my cursory glance at the problem already hinted at some sort of hierarchy-driven spec composition mechanism, which seems like it will inevitably lead to some tradeoffs.

But while this behavior is self-explanatory when looking at the source code, I think it should be mentioned in the README and/or the pre-init-spec docstring.

On the other hand I'm now wondering what kind of hooks could be provided so that if I have N composite keys, that should all somehow use the same spec from one of their components, I could define a single method for that hook rather the N boilerplate methods.

[weavejester]

@weavejester I'm not sure I understand because the two defmethods you quoted above never get invoked for that config. pre-init-spec dispatches on the four composite keys.

Ah, you're right. There's a bug; pre-init-spec doesn't normalize the composite keys first.

[julienfantin]

@weavejester I actually noticed that discrepancy but ig/normalize-key returns something like :integrant.composite/group.a+handler.greet_35275 which seemed like something that would be used for an internal dispatch mechanism.

Could you describe what you think the behavior should be from the perspective of a user trying to write pre-init-spec methods on a composite key?

[weavejester]

The behaviour should be the same as all the other multimethods, in that composite keys should be treated as keywords that derive from all the keywords contained in the vector.

In other words, this:

[:group/a :handler/greet]

Should have the same semantic meaning as:

(derive :temp/a :group/a)
(derive :temp/a :handler/greet)
:temp/a

[julienfantin]

I'm sorry it's still not clear to me what method(s) would get dispatched for that composite key, and if there are more than one, how would the different specs get combined?

[weavejester]

It's the same rules as normal multimethods. The most specific key is chosen, and ties are broken by prefer-method.

So in the above case we have a key that's derived from :group/a and :handler/greet. Clojure will look for a multimethod that handles either of those. If there is a method for both, it uses prefer-method to determine which to use. If neither has a multimethod, it looks at their parents, walking down the ancestors until it finds one with a method.

[weavejester]

Commit e97c4b3 should fix the dispatch of pre-init-spec.

[julienfantin]

Thanks! Any chance you could cut a release for this fix?

[julienfantin]

Awesome, thank you @weavejester

[weavejester]

Released 0.6.2