Dataset extensions not applied when methods are proxied through Sequel::Model

[eriklovmo]

We have some dataset extensions enabled, for example the excellent any_not_empty extension. Excerpt from our application initializer:

Sequel.extension :pg_hstore_ops
# ...and additional Sequel extensions

DB = Sequel.connect # (details omitted)

DB.extension :any_not_empty
# ...and additional DB extensions

Sequel::Model.plugin :auto_validations
# ...and additional model plugins

From experimenting, it seems that enabling a dataset extension does not make it available when a method is called on a Sequel::Model and sent to the associated dataset. This is somewhat surprising and could be a potential footgun. For example:

[1] pry(main)> Models::User.any?
D, [2025-10-23T09:40:43.407747 #84929] DEBUG -- : (0.018487s) SELECT * FROM "users"
=> true
[2] pry(main)> Models::User.dataset.any?
D, [2025-10-23T09:40:48.554919 #84929] DEBUG -- : (0.002106s) SELECT 1 AS "one" FROM "users"
=> true

Questions:

Is this behavior by design, or are we missing some configuration? If it is by design, is there a recommended way to make dataset extensions available when calling dataset methods from Sequel::Model? If this is intended behavior, could it be documented somewhere to prevent confusion? I could not find this documented, but I might very well have missed it.

Environment:

Sequel version: 5.97.0
Ruby version: 3.3.7
Database: PostgreSQL

[jeremyevans]

The behavior is currently expected. It is documented that Sequel::Model will only forward certain methods to the dataset, and any? is not one of the listed methods (https://sequel.jeremyevans.net/rdoc/classes/Sequel/Model/ClassMethods.html). The reason any? works anyway, is that Sequel::Model extends Enumerable.

It's simple to add any? to the list of methods that Model sends to the dataset, so I think that's the best fix currently.

We could extend this is to do the same for all Enumerable methods, but this is a double edged sword. I think it results in better behavior in the common case, but in the case where Model.each has been overridden, it changes behavior for the worse. Possibly something to be considered in Sequel 6, with a plugin to revert to the current behavior.

[eriklovmo]

OK, got it. Thanks for the quick reply. I just happened to use #any? as an example and actually first noticed this when using an extension that adds methods to Sequel::Dataset without corresponding proxies in Sequel::Model. I initially expected those methods to be dynamically forwarded to the associated dataset from a model class, but after thinking about it, I agree that the current, less magical behavior is much better.

[jeremyevans]

Change made in 43889b9