Suppose I have a class A as follows:
class A
class B
end
class C < B
end
...
end
And I want to create a class D that has the same nested class structure as A
class D
Replicate nested class structure of A here
end
I want D to look like:
class D
class B
end
class C < B
end
...
end
So that I can do A::B and D::B with different results
How can I achieve this? Thanks in advance.
class Module
def replicate m
m.constants.each do |sym|
case mm = m.const_get(sym)
when Class then const_set(sym, Class.new(mm.superclass)).replicate(mm)
when Module then const_set(sym, Module.new(mm.superclass)).replicate(mm)
end
end
end
end
class D
replicate A
end
But the superclass part may not be correct with this code.
class A
end
class D
end
[A, D].each do |c|
c.class_eval %Q(
class B
def bar; puts "B#bar in #{c} is not shared" end # <--- new
end
class C < B
def foo; puts "C#foo in #{c}" end
end
)
end
p A.constants
p A::C.instance_methods(false)
p D.constants
p D::C.instance_methods(false)
A::C.new.foo
D::C.new.foo
New
A::B.new.bar
D::B.new.bar
=begin
class B # creates a new class B
def quux; puts "A::B#quux in #{self}" end
end
A::B.new.quux #=> undefined method `quux' for #<A::B:0x101358a98> (NoMethodError)
=end
class A::B # reopens B in A
def quux; puts "A::B#quux in #{self}" end
end
A::B.new.quux
Execution :
$ ruby -w t.rb
["B", "C"]
["foo"]
["B", "C"]
["foo"]
C#foo in A
C#foo in D
New
B#bar in A is not shared
B#bar in D is not shared
A::B#quux in #<A::B:0x10402da28>
It's more duplicating than replicating the whole internal structure, including the methods and possible variables. For this you need reflection, or maybe marshall out and in.
New : If you put something in the text inside %Q(), class_eval will evaluate it for each class, hence it is not shared. B is not independent, you have two different classes A::B and D::B.
If you want to add the same code to both classes, create a module and include it. Ruby creates a proxy which points to the module and inserts the proxy in the chain of pointers starting from the class of the object to its superclass, so that the search method mechanism will look for module's methods after methods of the class and before methods in the superclass.
class D
extend A
end
will define instance methods of A as class (singleton) methods of D. Sounds ugly. I think that you should experiment, display what happens with puts, p, instance_methods, singleton_methods and the like.
Related
I'm trying to do something like this in Ruby:
class A
def b(c)
Class.new do
def d
c
end
end
end
end
class B < A.b('whoa'); end
#
# What I want:
#
B.new.d # => 'whoa'
#
# What I get:
#
# NameError: undefined local variable or method `d' for B:0xfdfyadayada
Is there any way to do this?
The context is I'm trying to get some code reuse by constructing classes that are mostly similar except for some minor config.
I'm looking for an API similar to Sequel::Model where you can do something like:
class Order < Sequel::Model(:order)
The contents of a method definition using the def keyword are not lexically scoped the way blocks are. In other words you can’t do something like this:
foo = 7
def bar
# Code here inside the method definition can't
# "see" foo, so this doesn't work.
foo
end
However you can use the define_method method to dynamically define a method using a block, and the block will be able to refer to local variables in its outer scope. So in your case you could do this (I’ve also changed def b to def self.b, which I think is what you meant):
class A
def self.b(c)
Class.new do
define_method(:d) do # N.B. the d here is a symbol
c
end
end
end
end
class B < A.b('whoa'); end
B.new.d # => 'whoa'
This would also work if you created other classes:
class C < A.b('dude'); end
C.new.d # => 'dude'
# B still works
B.new.d # => 'whoa'
Assuming that b is a class method (i.e def self.b instead of def b), you could store c in a class variable.
class A
def self.b(c)
Class.new do
##c = c
def d
##c
end
end
end
end
class B < A.b('whoa'); end
puts B.new.d # => whoa
There could be a better solution, I haven't looked at the source code of Sequel. This is just the first thing that came to my mind.
My understanding of instance_eval was that if I have module M then the following were equivalent:
module M
def foo
:foo
end
end
class C
class << self
include M
end
end
puts C.foo
equivalent to:
module M
def foo
:foo
end
end
class C
end
C.instance_eval do
include M
end
puts C.foo
However, the first example prints :foo and the second throws a NoMethodError? (Ruby 2.3.0)
In both cases above, if I had replaced:
include M
with:
def foo
:foo
end
ie directly defining the method rather than including a module then both cases would have resulted in a C.foo method being defined. Should I be surprised at this difference between include and defining the method directly?
Or does it ever even make sense to call include within the context of instance_eval? Should it only ever be called within a class_eval?
In each of these cases, what object are you calling include on? In your first example, you're calling include on C's singleton class:
class C
class << self
p self == C.singleton_class
include M
end
end
# => true
p C.foo
# => :foo
...so your include line is equivalent to C.singleton_class.include(M).
In your second example, however, you're calling include on C itself:
class C
end
C.instance_eval do
p self == C
include M
end
# => true
p C.foo
# => NoMethodError: undefined method `foo' for C:Class
p C.new.foo
# => :foo
...so you're doing the equivalent of C.include(M), which is the same as:
class C
p self == C
include M
end
# => true
p C.new.foo
# => :foo
What would work like you want would be to call instance_eval on C's singleton class:
class D
end
D.singleton_class.instance_eval do
p self == D.singleton_class
include M
end
# => true
p D.foo
# => :foo
Module#class_eval() is very different from Object#instance_eval(). The instance_eval() only changes self, while class_eval() changes both self and the current class.
Unlike in your example, you can alter class_instance vars using instance_eval though, because they are in the object scope as MyClass is a singleton instance of class Class.
class MyClass
#class_instance_var = 100
##class_var = 100
def self.disp
#class_instance_var
end
def self.class_var
##class_var
end
def some_inst_method
12
end
end
MyClass.instance_eval do
#class_instance_var = 500
def self.cls_method
##class_var = 200
'Class method added'
end
def inst_method
:inst
end
end
MyClass.disp
#=> 500
MyClass.cls_method
#=> 'Class method added'
MyClass.class_var
#=> 100
MyClass.new.inst_method
# undefined method `inst_method' for #<MyClass:0x0055d8e4baf320>
In simple language.
If you have a look in the upper class defn code as an interpreter, you notice that there are two scopes class scope and object scope. class vars and instance methods are accessible from object scope and does not fall under jurisdiction of instance_eval() so it skips such codes.
Why? because, as the name suggests, its supposed to alter the Class's instance(MyClass)'s properties not other object's properties like MyClass's any object's properties. Also, class variables don’t really belong to classes—they belong to class hierarchies.
If you want to open an object that is not a class, then you can
safely use instance_eval(). But, if you want to open a class definition and define methods with def or include some module, then class_eval() should be your pick.
By changing the current class, class_eval() effectively reopens the class, just like the class keyword does. And, this is what you are trying to achieve in this question.
MyClass.class_eval do
def inst_method
:inst
end
end
MyClass.new.inst_method
#=> :inst
It is known that in Ruby, class methods get inherited:
class P
def self.mm; puts 'abc' end
end
class Q < P; end
Q.mm # works
However, it comes as a surprise to me that it does not work with mixins:
module M
def self.mm; puts 'mixin' end
end
class N; include M end
M.mm # works
N.mm # does not work!
I know that #extend method can do this:
module X; def mm; puts 'extender' end end
Y = Class.new.extend X
X.mm # works
But I am writing a mixin (or, rather, would like to write) containing both instance methods and class methods:
module Common
def self.class_method; puts "class method here" end
def instance_method; puts "instance method here" end
end
Now what I would like to do is this:
class A; include Common
# custom part for A
end
class B; include Common
# custom part for B
end
I want A, B inherit both instance and class methods from Common module. But, of course, that does not work. So, isn't there a secret way of making this inheritance work from a single module?
It seems inelegant to me to split this into two different modules, one to include, the other to extend. Another possible solution would be to use a class Common instead of a module. But this is just a workaround. (What if there are two sets of common functionalities Common1 and Common2 and we really need to have mixins?) Is there any deep reason why class method inheritance does not work from mixins?
A common idiom is to use included hook and inject class methods from there.
module Foo
def self.included base
base.send :include, InstanceMethods
base.extend ClassMethods
end
module InstanceMethods
def bar1
'bar1'
end
end
module ClassMethods
def bar2
'bar2'
end
end
end
class Test
include Foo
end
Test.new.bar1 # => "bar1"
Test.bar2 # => "bar2"
Here is the full story, explaining the necessary metaprogramming concepts needed to understand why module inclusion works the way it does in Ruby.
What happens when a module is included?
Including a module into a class adds the module to the ancestors of the class. You can look at the ancestors of any class or module by calling its ancestors method:
module M
def foo; "foo"; end
end
class C
include M
def bar; "bar"; end
end
C.ancestors
#=> [C, M, Object, Kernel, BasicObject]
# ^ look, it's right here!
When you call a method on an instance of C, Ruby will look at every item of this ancestor list in order to find an instance method with the provided name. Since we included M into C, M is now an ancestor of C, so when we call foo on an instance of C, Ruby will find that method in M:
C.new.foo
#=> "foo"
Note that the inclusion does not copy any instance or class methods to the class – it merely adds a "note" to the class that it should also look for instance methods in the included module.
What about the "class" methods in our module?
Because inclusion only changes the way instance methods are dispatched, including a module into a class only makes its instance methods available on that class. The "class" methods and other declarations in the module are not automatically copied to the class:
module M
def instance_method
"foo"
end
def self.class_method
"bar"
end
end
class C
include M
end
M.class_method
#=> "bar"
C.new.instance_method
#=> "foo"
C.class_method
#=> NoMethodError: undefined method `class_method' for C:Class
How does Ruby implement class methods?
In Ruby, classes and modules are plain objects – they are instances of the class Class and Module. This means that you can dynamically create new classes, assign them to variables, etc.:
klass = Class.new do
def foo
"foo"
end
end
#=> #<Class:0x2b613d0>
klass.new.foo
#=> "foo"
Also in Ruby, you have the possibility of defining so-called singleton methods on objects. These methods get added as new instance methods to the special, hidden singleton class of the object:
obj = Object.new
# define singleton method
def obj.foo
"foo"
end
# here is our singleton method, on the singleton class of `obj`:
obj.singleton_class.instance_methods(false)
#=> [:foo]
But aren't classes and modules just plain objects as well? In fact they are! Does that mean that they can have singleton methods too? Yes, it does! And this is how class methods are born:
class Abc
end
# define singleton method
def Abc.foo
"foo"
end
Abc.singleton_class.instance_methods(false)
#=> [:foo]
Or, the more common way of defining a class method is to use self within the class definition block, which refers to the class object being created:
class Abc
def self.foo
"foo"
end
end
Abc.singleton_class.instance_methods(false)
#=> [:foo]
How do I include the class methods in a module?
As we just established, class methods are really just instance methods on the singleton class of the class object. Does this mean that we can just include a module into the singleton class to add a bunch of class methods? Yes, it does!
module M
def new_instance_method; "hi"; end
module ClassMethods
def new_class_method; "hello"; end
end
end
class HostKlass
include M
self.singleton_class.include M::ClassMethods
end
HostKlass.new_class_method
#=> "hello"
This self.singleton_class.include M::ClassMethods line does not look very nice, so Ruby added Object#extend, which does the same – i.e. includes a module into the singleton class of the object:
class HostKlass
include M
extend M::ClassMethods
end
HostKlass.singleton_class.included_modules
#=> [M::ClassMethods, Kernel]
# ^ there it is!
Moving the extend call into the module
This previous example is not well-structured code, for two reasons:
We now have to call both include and extend in the HostClass definition to get our module included properly. This can get very cumbersome if you have to include lots of similar modules.
HostClass directly references M::ClassMethods, which is an implementation detail of the module M that HostClass should not need to know or care about.
So how about this: when we call include on the first line, we somehow notify the module that it has been included, and also give it our class object, so that it can call extend itself. This way, it's the module's job to add the class methods if it wants to.
This is exactly what the special self.included method is for. Ruby automatically calls this method whenever the module is included into another class (or module), and passes in the host class object as the first argument:
module M
def new_instance_method; "hi"; end
def self.included(base) # `base` is `HostClass` in our case
base.extend ClassMethods
end
module ClassMethods
def new_class_method; "hello"; end
end
end
class HostKlass
include M
def self.existing_class_method; "cool"; end
end
HostKlass.singleton_class.included_modules
#=> [M::ClassMethods, Kernel]
# ^ still there!
Of course, adding class methods is not the only thing we can do in self.included. We have the class object, so we can call any other (class) method on it:
def self.included(base) # `base` is `HostClass` in our case
base.existing_class_method
#=> "cool"
end
As Sergio mentioned in comments, for guys who are already in Rails (or don’t mind depending on Active Support), Concern is helpful here:
require 'active_support/concern'
module Common
extend ActiveSupport::Concern
def instance_method
puts "instance method here"
end
class_methods do
def class_method
puts "class method here"
end
end
end
class A
include Common
end
You can have your cake and eat it too by doing this:
module M
def self.included(base)
base.class_eval do # do anything you would do at class level
def self.doit #class method
##fred = "Flintstone"
"class method doit called"
end # class method define
def doit(str) #instance method
##common_var = "all instances"
#instance_var = str
"instance method doit called"
end
def get_them
[##common_var,#instance_var,##fred]
end
end # class_eval
end # included
end # module
class F; end
F.include M
F.doit # >> "class method doit called"
a = F.new
b = F.new
a.doit("Yo") # "instance method doit called"
b.doit("Ho") # "instance method doit called"
a.get_them # >> ["all instances", "Yo", "Flintstone"]
b.get_them # >> ["all instances", "Ho", "Flintstone"]
If you intend to add instance, and class variables, you will end up pulling out your hair as you will run into a bunch of broken code unless you do it this way.
I want the following module to be included in a class I have:
module InheritanceEnumerator
def self.included(klass)
klass.instance_eval do
instance_variable_set('#subclasses',[])
def self.subclasses
#subclasses
end
original_method = self.respond_to?(:inherited) ? self.public_method(:inherited) : nil
instance_variable_set('#original_inherited_method', original_method)
def self.inherited(subclass)
#original_inherited_method.call(subclass) if #original_inherited_method
#subclasses<<subclass
end
end
end
end
What I'm trying to achieve is that I want my parent class to have references to direct children. I also need any other previous "inherited" methods set on my class by other stuff to stay in place. What am I doing wrong?
Your code works in this situation (for me):
class C; include InheritanceEnumerator; end
C.subclasses #=> []
class C1 < C; end
class C2 < C; end
C.subclasses #=> [C1, C2]
But fails in the following situation:
class C11 < C1; end
C1.subclasses => NoMethodError: undefined method `<<' for nil:NilClass
This is because you are only initializing #subclasses when the module is included; but you are forgetting that subclasses of C also have access to the modules methods but do not explictly include it.
You fix this by doing the following:
def self.subclasses
#subclasses ||= []
#subclasses
end
def self.inherited(subclass)
#original_inherited_method.call(subclass) if #original_inherited_method
#subclasses ||= []
#subclasses << subclass
end
EDIT:
Okay, in future, please state what your problem is more fully and provide the test code you are using; as this was an exercise in frustration.
The following works fine with your code:
class C
def self.inherited(s)
puts "inherited by #{s}!"
end
include InheritanceEnumerator
end
class D < C; end #=> "inherited by D!"
C.subclasses #=> [D]
Perhaps the reason it wasn't working for you is that you included InheritanceEnumerator before you had defined the inherited method?
I'm looking for a way of making a method "personal" - note NOT PRIVATE to a class
here is an example - by "personal" I mean the behaviour of method "foo"
class A
def foo
"foo"
end
end
class B < A
def foo
"bar"
end
end
class C < B
end
a=A.new; b=B.new;c=C.new
I'm looking for a way of producing the following behaviour
a.foo #=> "foo"
b.foo #=> "bar"
c.foo #=> "foo" (ultimate base class method called)
Instead of creating 'personal' methods, change your inheritance structure.
It appears that you want the C class to have only some of the same functionality of the B class while not making changes to the A class.
class A
def foo
"foo"
end
end
class BnC < A
end
class B < BnC
def foo
"bar"
end
end
class C < BnC
end
a=A.new; b=B.new;c=C.new
There's no standard way of doing this. It circumvents how inheritance works. You could implement B's method to do the logic like this:
def foo
instance_of?(B) ? "bar" : super
end
And you could of course define a method on Class that would do this for you similar to public and private.
class Class
def personal(*syms)
special_class = self
syms.each do |sym|
orig = instance_method(sym)
define_method(sym) {|*args| instance_of?(special_class) ? orig.bind(self).call(*args) : super}
end
end
end
Then you can personal :foo in B just like you'd private :foo.
(This isn't at all optimized and I didn't implement the zero-argument behavior that public and private have because frankly it's a huge PITA to do right and even then it's a hack.)
Seems like it could be confusing, but here's one option:
class A
def foo
"foo"
end
end
class B < A
def initialize #when constructing, add the new foo method to each instance
def self.foo
"bar"
end
end
end
class C < B
def initialize #when constructing, do nothing
end
end
More generally, using a similar approach, you can always add a method to a given instance, which of course has no effect on inherited classes or indeed on other instances of the same class.
If you give us specifics of what you're ultimately trying to accomplish we can probably be more helpful.
Answering this is a bit tricky since I don't really see what you want to accomplish in practice, but you could try something like
class C < B
def foo
self.class.ancestors[-3].instance_method(:foo).bind(self).call
end
end
(The ancestors[-3] assumes that A inherits from Object and Kernel and your intent was to access the method from the topmost non-builtin class. Of course you could substitute self.class.ancestors[-3] with just A, or figure out the class from the Array ancestors yourself, etc.)
In practice it would be simpler to alias the original in class B if you can modify it (i.e. alias :foo_from_A :foo in class B < A before the new def foo, then you can call foo_from_A in C). Or just redefine what you want in C. Or design the whole class hierarchy differently.
You can write a shortcut function to handle personalizing methods.
def personalize(*methodNames)
old_init = instance_method(:initialize)
klass = self
modul = Module.new {
methodNames.each { |m|
define_method(m, klass.instance_method(m)) if klass.method_defined?(m)
}
}
methodNames.each { |m|
remove_method(m) if method_defined?(m)
}
define_method(:initialize) { |*args|
# I don't like having to check instance_of?, but this is the only way I
# could thing of preventing the extension of child classes. At least it only
# has to happen once, during initialization.
extend modul if instance_of?(klass)
old_init.bind(self).call(*args)
}
self
end
class A
def foo
"foo"
end
end
class B < A
def foo
"bar"
end
def bam
'bug-AWWK!'
end
personalize :foo, :bam, :nometh
end
class C < B
end
a=A.new; b=B.new; c=C.new
a.foo #=> "foo"
b.foo #=> "bar"
b.bam #=> "bug-AWWK!"
c.foo #=> "foo"
C.instance_method(:foo) # => #<UnboundMethod: C(A)#foo>
c.bam #throws NoMethodError
Sometimes you don't really want an "is a" (inheritance) relationship. Sometimes what you want is "quacks like a." Sharing code among "quacks like a" classes is easily done by using modules to "mix in" methods:
#!/usr/bin/ruby1.8
module BasicFoo
def foo
"foo"
end
end
class A
include BasicFoo
end
class B
def foo
"bar"
end
end
class C
include BasicFoo
end
p A.new.foo # => "foo"
p B.new.foo # => "bar"
p C.new.foo # => "foo"