Access Modifiers (Internal Interfaces)
I’m would like to discuss with you the concept of Internal Interfaces.
Currently ruby offers three access modifiers choices to set methods visibility:
Public methods define what we may call the Public Interface.
Private methods are private to the class and its subclasses.
Protected methods are public for the subclasses.
I would like to work with a new access modifier that would allow the creation of Internal methods.
Internal methods would be object methods that could only be called within the namespace.
module Foo; end class Foo::Bar def baz puts ‘baz’ end internal :baz end class Foo::Qux def baz ::Foo::Bar.new.baz end end
Is this something that we can think about in a future implementation of ruby?
An extra feature that would not break backward compatibility.
#2 [ruby-core:63394] Updated by Daniel Ferreira almost 2 years ago
The notion of Internal Interface is about being able to use the methods identified as internal only in the context of a given root module and all its subclasses and/or submodules. Lets call it: namespace full tree.
The namespace full tree in this context would integrate all constants of the tree defined by the namespace root and all its subnodes.
That way if we develop a new gem lets call it
We would be able to define an Internal Interface that would only be used internally by
gem foo classes and modules
and by any other gem foo extension like for instance:
We have this notion of Public and Internal interface when looking at an API (Application Public Interface) in Web Services.
Web Services expose to the users a given set of methods/actions making them Public.
The remaining of the architecture is internal to the infrastructure.
In ruby libraries/packages/gems every class is available to the user.
That makes development of libraries hard because a change to the public interface of a given class may have side effects.
We may need to create a major release just because we need to change some method in one of our classes that should not be used outside of the library context.
What I aim as an architect is to allow a better defined contract between developers and users.
Having the definition of an Internal Interface the architect will have greater freedom to develop the required functionalities.
The development process may be like this:
- Define Public Interface
- Release version 1.0 once that Public Interface is well defined and will hardly change.
- Reengineer the Internal Interface and add functionality to the Public Interface without breaking backwards compatibility.
This way dependencies between libraries will be much better understood.
Versioning of releases with this feature will be able to communicate in a greater detail to the users what nature of changes were made in each release.
We may then identify:
- Major release
Third party libraries and/or applications might be broken.
Library extensions might be broken.
- Public Interface changes with broken backward compatibility
Public Interface changes with backward compatibility
- Minor release
Library extensions using internal interface features might be broken.
- Internal Interface changes with broken backward compatibility
Internal Interface changes with impact on overall behaviour
- Performance improvements
- New core engine
- Patch (Tiny) release
No side effects on extensions or third party libraries and/or applications.
- Internal Interface changes with backward compatibility (new features for extensions)
- Changes to private interface (private and protected methods)
- Unit test updates
With this feature the library development cycle will be much more linear.
Major releases will be in a much smaller number then nowadays.
We will get much more stable ecosystems.
My idea is to leverage ruby further in order to better fit the enterprise environment.
Happy to reply to any doubts and discuss with you this thoughts and what other alternatives may we have.
#3 [ruby-core:63463] Updated by Zachary Scott almost 2 years ago
I would suggest opening a feature request if you feel strongly about adding
internal to Ruby.
#5 [ruby-core:72985] Updated by Jörg W Mittag 3 months ago
I'm having trouble understanding what you mean by namespace. Classes and modules don't belong to namespaces. Constants belong to modules, but the fact that two classes which may or may not be referenced by two constants which are namespaced in the same module does not imply any sort of relationship whatsoever between either the two classes or one of the classes and the module.
Expanding your example a bit:
module Foo; end Baz = Bar = Foo class Foo::Bar def baz puts ‘baz’ end internal :baz end class Bar::Qux def baz ::Baz::Bar.new.baz end end
Should it work? If not, why? The modules and classes involved are the exact same ones as in your example.
I mean, what about this:
foo = Class.new bar = Class.new baz = Module.new baz.const_set(:Foo, foo) baz.const_set(:Bar, bar) quux = Module.new quux.const_set(:Goobledigook, foo) quux.const_set(:Blahdiblah, bar) baz.send(:remove_const, :Bar) module One; module Two; end end One::Two::Three = quux::Goobledigook
Okay, now what namespace relations are there?
If I now do this:
class One::Two::Three internal def blub; end end
Am I allowed to call
foo.new.blub from a method in