[ruby-core:105738] [Ruby master Bug#18021] Mixins in Refinements: possibly multiple bugs, workarounds are awkward

Issue #18021 has been updated by jeremyevans0 (Jeremy Evans).

Status changed from Open to Closed

Refinement#include is now deprecated and will be removed in Ruby 3.2.

----------------------------------------
Bug #18021: Mixins in Refinements: possibly multiple bugs, workarounds are awkward
https://bugs.ruby-lang.org/issues/18021#change-94237

* Author: josh.cheek (Josh Cheek)
* Status: Closed
* Priority: Normal
* ruby -v: ruby 3.0.1p64 (2021-04-05 revision 0fb782ee38) [arm64-darwin20]
* Backport: 2.6: UNKNOWN, 2.7: UNKNOWN, 3.0: UNKNOWN
----------------------------------------
## Maybe bug 1

Refinements must be created after methods are defined.

```ruby
# This one seems buggy
module M1
  refine(Object) { include M1 }
  def m() = M1
  using M1
  m rescue $! # => #<NameError: undefined local variable or method `m' for M1:Module>
end

# It works if you refine it after the method is defined
module M2
  def m() = M2
  refine(Object) { include M2 }
  using M2
  m # => M2
end

# If it wasn't in a refinement, it would work
module M3
  Object.class_eval { include M3 }
  def m() = M3
  m # => M3
end
```


## Maybe bug 2

Here, a module can't delegate to another method it defines, while using a refinement. Presumably this is because the refinement is lexically scoped and where the method is defined, it is not within that scope.

```ruby
module M
  def a() = 1
  def b() = a # ~> NameError: undefined local variable or method `a' for main:Object
  refine(Object) { include M }
end

using M
a # => 1
b # => 
```

## Maybe bug 3

Refinements declared after methods are defined are not available to those methods, even when they use the refinement. We try to use the module to address #2, but because of #1, the `refine` must go after the definitions.

```ruby
module M
  using M # <-- this line is added to the example from #2
  def a() = 1
  def b() = a # ~> NameError: undefined local variable or method `a' for main:Object
  refine(Object) { include M }
end

using M
a # => 1
b # => 
```

## Awkward workarounds:

So, `refine` depends on `def`, which depends on `using`, which depends on `refine`, which is where we started.


Here are several ways I found to break the cycle of dependance:

#### 1. "declare" methods by predefining them

```ruby
module M
  # Predefine method stubs
  def a() = nil
  def b() = nil

  # Now the refinement can see them, define the refinement
  refine(Object) { include M }

  # Now that the refinement exists, `using` will make it available
  using M

  # Now method definitions can see each other,
  # so overwrite the stubs with the real implementations
  def a() = 1
  def b() = a
end

# And now things work as expected
using M
a # => 1
b # => 1
```

#### 2. Put the module body in a loop

First iteration defines them without visibility to each other, because the refinement hasn't been made yet. On the second iteration, the refinement has been made, so the second time they're defined, they can see each other.

```ruby
module M
  2.times do
    using M # this can go either here or after `refine` below
    def a() = 1
    def b() = a
    refine(Object) { include M }
  end
end

# And now things work as expected
using M
a # => 1
b # => 1
```

#### 3. Define the methods in a block and call it before and after refining

```ruby
module M
  methods = lambda do
    using M
    def a() 1 end
    def b() a end
  end

  methods.call
  refine(Object) { include M }
  methods.call
end

using M
b # => 1
```

#### 4. (this does not work) Include the refinement into the module

Including this one, b/c it feels like maybe some sort of combination refinement/mixin might be what is needed. Like generally, methods defined in a refinement can see other methods, even across other blocks, if they're all on the same module. The problem is there isn't an obvious way to get them back out so that they're visible in the module outside of the refinement.

```ruby
module M
  include refine(M) { # ~> ArgumentError: refinement module is not allowed
    def a() = 1
    def b() = a
  }
  refine(Object) { include M }
end
```

## The problem with the obvious workaround

There is an obvious workaround: don't put them in a module, just refine directly:

```ruby
module M
  refine Object do
    def a() = 1
    def b() = a
  end
end
using M
b # => 1
```

However, then that behaviour isn't available for mixing into other classes and objects. So if I want both (which I often find I do), then there don't seem to be any good options available.

For example, can anyone come up with a better way to write the example below? (without moving `to_type` to `Object`, as that can't be expected to work for the next helper method).

```ruby
module Types
  2.times do
    using Types
    private def to_type(o) = (Symbol === o ? RespondTo.new(o) : o)
    def |(type) = Or.new(to_type(self), to_type(type))
    refine(Module) { include Types }
    refine(Symbol) { include Types }
  end

  RespondTo = Struct.new(:name) {
    def ===(o) o.respond_to? name end
    def inspect() = ".respond_to?(#{name.inspect})"
  }.include(Types)

  Or = Struct.new(:left, :right) {
    def ===(o) left === o || right === o end
    def inspect() = "(#{left.inspect} | #{right.inspect})"
  }.include(Types)
end

using Types
Integer | String | (:to_int | :to_str) 
# => ((Integer | String) | (.respond_to?(:to_int) | .respond_to?(:to_str)))
```



-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>