I'm trying to alias a method that use's Ruby's special $& (returns last regex match). I can do this manually and it works:
original = String.instance_method(:sub)
String.send(:define_method, :sub) do |*args, &block|
puts "called"
original.bind(self).call(*args, &block)
end
"foo".sub(/f/) { $&.upcase }
called
# => "Foo"
However if I try to write a method that does this for me, it fails:
def programatic_alias(klass, method_name)
original = klass.instance_method(method_name)
klass.send(:define_method, method_name) do |*args, &block|
puts "called"
original.bind(self).call(*args, &block)
end
end
programatic_alias(String, :sub)
"foo".sub(/f/) { $&.upcase }
called
NoMethodError: undefined method `upcase' for nil:NilClass
called
called
called
from (irb):19:in `block in irb_binding'
It looks like the global state is being affected by the scope of the programatic_alias method, but I'm not sure if that's what's going on. The questions is this: how can I programmatically alias String#sub so that it still works with Ruby's special global variables?
As far as I know, you can't do this. The docs say
These global variables are thread-local and method-local variables.
If you dig into the ruby source, accessing $& calls last_match_getter which gets its data from rb_backref_get, which calls vm_svar_get which (skipping over a few more internal methods) gets the current control frame and reads the data from there. None of this data is exposed to the ruby api - there's no way to propagate this data from one frame to the one you want to access it in.
In your second example the call to the original method is happening inside your programatic_alias method, and so $& is being set in that scope. For the same reason
'foo'.try(:sub, /f/) {$&.upcase}
won't work either.
Your first example half works because the place where sub is called and the place where $& is referenced (inside the block) are in the same method scope (in this case the ruby top level). Change it to:
original = String.instance_method(:sub)
String.send(:define_method, :sub) do |*args, &block|
puts "called"
original.bind(self).call(*args, &block)
end
def x
"foo".sub(/f/) { $&.upcase }
end
x()
and $& is no longer defined in your block (if you catch the exception thrown by x you can see that $& is being set at the top level)
Related
I'm able to get access to a Ruby method's arguments using the TracePoint API:
def foo(foo_arg)
end
trace = TracePoint.trace(:call, :c_call) do |tp|
tp.disable
case tp.method_id
when :foo, :sub
method = eval("method(:#{tp.method_id})", tp.binding)
method.parameters.each do |p|
puts "#{p.last}: #{tp.binding.local_variable_get(p.last)}"
end
end
tp.enable
end
trace.enable
foo(10)
# => foo_arg: 10
However when I try this with a c method call, I get an error.
"foo".sub(/(f)/) { $1.upcase }
script.rb:20:in `method': undefined method `sub' for class `Object' (NameError)
from script.rb:20:in `<main>'
from script.rb:8:in `eval'
from script.rb:8:in `block in <main>'
from script.rb:20:in `<main>'
This looks like it happens because of a discrepancy between the binding returned when using a C method call and regular Ruby method call.
In the Ruby case tp.self is equal to tp.binding.eval("self") is main however in the C case tp.self is "foo" and tp.binding.eval("self") is main. Is there a way to get the arguments passed into a method using TracePoint for both Ruby and C defined methods?
As you point in your question and as it documented in ruby documentation, tp.self returns a traced object, which have a method method you are looking for.
I think you should use
method = tp.self.method(tp.method_id)
instead of
method = eval("method(:#{tp.method_id})", tp.binding)
UPDATE. Some explanation regarding your last paragraph in question. tp.self in first case (when you call foo) is point to main, because you define foo method in main context and it points to String object in second case because sub is defined there. But tp.binding.eval("self") returns main in both cases because it returns a calling context (not a 'define' context as you expect) and in both cases it is main.
UPDATE (in reply to comment) I think that the only way to do this is to monkey patch sub and all other methods that you are interesting for. Code example:
class String
alias_method :old_sub, :sub
def sub(*args, &block)
old_sub(*args, &block)
end
end
trace = TracePoint.trace(:call, :c_call) do |tp|
tp.disable
case tp.method_id
when :sub
method = tp.self.method(tp.method_id)
puts method.parameters.inspect
end
tp.enable
end
trace.enable
"foo".sub(/(f)/) { |s| s.upcase }
One big drawback is that you can't use $1, $2, ... vars in your original blocks. As pointed here where is no way to make it works. However you can still use block parameters (s in my example).
Is there a way to bind an existing method to an existing instance of an object if both the method and the instance are passed as symbols into a method that does that if the instance is not a symbol?
For example:
def some_method
#do something
end
some_instance = Klass.new(something)
def method_that_binds(:some_method, to: :some_instance)
#how do I do that?
end
Your requirements are a little unusual, but it is possible to do this mostly as you say:
class Person; end
harry = Person.new
barry = Person.new
def test
puts 'It works!'
end
define_method :method_that_binds do |a_method, to|
eval(to[:to].to_s).singleton_class.send(:define_method, a_method, &Object.new.method(a_method))
end
method_that_binds :test, to: :harry
harry.test
# It works! will be sent to STDOUT
barry.test
# undefined method 'test'
This doesn't actually use a named parameter, but accepts a hash with a to key, but you can see you can call it in the way you want. It also assumes that the methods you are defining are defined globally on Object.
The API you want doesn't easily work, because you have to know from which scope you want to access the local variable. It's not quite clear to me why you want to pass the name of the local variable instead of passing the content of the local variable … after all, the local variable is present at the call site.
Anyway, if you pass in the scope in addition to the name, this can be accomplished rather easily:
def some_method(*args)
puts args
puts "I can access some_instance's ivar: ##private_instance_var"
end
class Foo; def initialize; #private_instance_var = :foo end end
some_instance = Foo.new
def method_that_binds(meth, to:, within:, with: [])
self.class.instance_method(meth).bind(within.local_variable_get(to)).(*with)
end
method_that_binds(:some_method, to: :some_instance, within: binding, with: ['arg1', 'arg2'])
# arg1
# arg2
# I can access some_instance's ivar: foo
As you can see, I also added a way to pass arguments to the method. Without that extension, it becomes even simpler:
def method_that_binds(meth, to:, within:)
self.class.instance_method(meth).bind(within.local_variable_get(to)).()
end
But you have to pass the scope (Binding) into the method.
If you'd like to add a method just to some_instance i.e. it's not available on other instances of Klass then this can be done using define_singleton_method (documentation here.)
some_instance.define_singleton_method(:some_method, method(:some_method))
Here the first use of the symbol :some_method is the name you'd like the method to have on some_instance and the second use as a parameter to method is creating a Method object from your existing method.
If you'd like to use the same name as the existing method you could wrap this in your own method like:
def add_method(obj, name)
obj.define_singleton_method(name, method(name))
end
Let's say we have a class A with a method a and a local variable c.
class A
def a; 10 end
end
c = '5'
And we want to add the method A#a to c.
This is how it can be done
c.singleton_class.send :define_method, :b, &A.new.method(:a)
p c.b # => 10
Explanations.
One way to add a method to an object instance and not to its class is to define it in its singleton class (which every ruby object has).
We can get the c's singleton class by calling the corresponding method c.signleton_class.
Next we need to dynamically define a method in its class and this can usually be accomplished by using the define_method which takes a method name as its first argument (in our case :b) and a block. Now, converting the method into a block might look a bit tricky but the idea is relatively simple: we first transform the method into a Method instance by calling the Object#method and then by putting the & before A.new.method(:a) we tell the interpreter to call the to_proc method on our object (as our returned object is an instance of the Method, the Method#to_proc will be called) and after that the returned proc will be translated into a block that the define_method expects as its second argument.
I am trying to make a simplistic implementation of AOP in ruby. I was able to implement before and after advices, I got stuck with around advice.
This is the target class that is going to be advised:
class MyClass
def method
puts "running method"
end
end
This is the Aspect class to instantiate objects capable of making advices:
class Aspect
def advise(class_name, method, type, &block)
class_name.send(:alias_method, :proceed, :method)
class_name.send(:define_method, :method) do
case type
when :before
yield
proceed
when :after
proceed
yield
when :around
yield(proceed) # * proceed is the old version of the method
end
end
end
end
(*) Yield should execute the block around MyClass#proceed on the current object when method is invoked.
Creating the target and the aspect:
mc = MyClass.new
a = Aspect.new()
Invoking the method without advising it:
puts mc.method
Advising MyClass#method with around:
a.advise(MyClass, :method, :around) do |proceed|
puts "First"
proceed # this is not working *
puts "Last"
end
puts mc.method
(*) I am not being able to pass something to identify the call of proceed, that is the invocation of the old method without the advice.
The output should be:
First
running method
Last
In Ruby, a method call looks like this:
receiver.method(arguments)
Or, you can leave off the receiver if the receiver is self.
So, to call a method named proceed on some receiver, you would write
receiver.proceed
However, in your implementation, you don't keep track of what the receiver should be, so since you don't know the receiver, you simply cannot call the method.
Note that there are lots of other problems with your approach as well. For example, if you advise multiple methods, you will alias them all to the same method, overwriting each other.
I believe there are two things going wrong here.
This section of code
when :around
yield(proceed) # * proceed is the old version of the method
end
Calls the block given to advise providing the output of the proceed method as an argument.
So your output probably looks something like:
running method
First
Last
This block
a.advise(MyClass, :method, :around) do |proceed|
puts "First"
proceed # this is not working *
puts "Last"
end
Just evaluates the argument given as proceed. If a method is given it does not call it. So taking problem 1 into consideration in your case the original definition of method (aliased to proceed) returns nil (output of return) which will be passed as the value to the proceed argument in the block when yielded. the block ends up evaluating to something like
puts "First"
nil
puts "Last"
mc.method is called.
To address the second part, you may want to consider using send. Because the inner workings of your aspect may not be known to your code that calls it. It may change over time, so what ever calls Aspect.advise shouldn't make assumptions that the original method will still be accessible. Instead, it should take an argument (the new method name) and send it to the object. Making the block passed to advise:
a.advise(MyClass, :method, :around) do |aliased_method_name|
puts "First"
send(aliased_method_name)
puts "Last"
end
And adjusting the around item added to your class when advise is called to the following:
when :around
yield(:proceed) # * proceed is the old version of the method
end
If you do both of these things, your around section will calls the provided block, using the symbol for the new alias for the overridden method.
N.B.: This approach won't work for methods that require any arguments.
This is what I did. In the definition of Aspect#advise now I use a Proc, like this:
when :around
yield Proc.new { proceed }
end
And when calling the method to advise MyClass#method with :around parameter I use this:
a.advise(MyClass, :method, :around) do |original|
puts "First"
original.call
puts "Last"
end
I got:
First
running method
Last
Here's the fixed version that will work for arguments, and avoid clobbering.
class Aspect
##count = 0
def self.advise(class_name, method, type=nil, &block)
old_method = :"__aspect_#{method}_#{##count += 1}"
class_name.send(:alias_method, old_method, method)
class_name.send(:define_method, method) do |*args, &callblock|
case type
when :before
yield
send(old_method, *args, &callblock)
when :after
send(old_method, *args, &callblock)
yield
when :around, nil
yield lambda {
send(old_method, *args, &callblock)
}
end
end
end
end
class Foo
def foo(what)
puts "Hello, #{what}!"
end
end
Aspect.advise(Foo, :foo) do |y|
puts "before around"
y.yield
puts "after around"
end
Aspect.advise(Foo, :foo, :before) do
puts "before"
end
Aspect.advise(Foo, :foo, :after) do
puts "after"
end
Foo.new.foo("world")
# before
# before around
# Hello, world!
# after around
# after
Inside the body of a class, I'd like to pass a block to a method called with. For the lifetime of the block, I would like a with_value method to be available.
Otherwise, everything inside the block should behave as if it were outside the block.
Here's an example:
class C
extend M
with "some value" do
do_something_complicated
do_something_complicated
do_something_complicated
end
end
We can almost get this with:
module M
def with(str, &block)
Object.new.tap do |wrapper|
wrapper.define_singleton_method :with_value do # Here's our with_value
str # method.
end
end.instance_eval &block
end
def do_something_complicated # Push a value onto an
(#foo ||= []).push with_value # array.
end
end
but there's a problem: since we're evaluating the block passed to with inside the context of a different object, do_something_complicated isn't available.
What's the right way to pull this off?
This will make with_value available only within the block. However, _with_value will be defined within or outside of the block.
module M
def _with_value
...
end
def with(str, &block)
alias with_value _with_value
block.call
undef with_value
end
...
end
I cannot tell from the question whether this is a problem. If it is a problem, you need to further describe what you are trying to do.
Basically, the idea is to use method_missing to forward method calls from the dummy class to the calling class. If you also need to access instance variables, you can copy them from the calling class to your dummy class, and then back again after the block returns.
The Ruby gem docile is a very simple implementation of such a system. I suggest you read the source code in that repository (don't worry, it's a very small codebase) for a good example of how DSL methods like the one in your example work.
Here is a way that is closer to your attempt:
module M
def with(str, &block)
dup.tap do |wrapper|
wrapper.define_singleton_method :with_value do
...
end
end.instance_eval &block
end
...
end
dup will duplicate the class from where with is called as a class method.
I've been writing a DSL, and I'm trying to get a dynamically defined method to be accessible from a lambda. This works fine unless you try to do a setter something= in which case the lambda invocation just sets a local variable instead.
A simplified example:
class Caller
attr_accessor :cmd
def callme
self.class.send(:define_method, "something") { puts "Retrieve Something" }
self.class.send(:define_method, "something=") {|val| puts "Set Something = #{val}" }
instance_exec &cmd
end
end
c = Caller.new
c.cmd = lambda { something = 1 }
c.callme
This also works fine if I use self.something=. However that's less than ideal in the case of a DSL.
Is it possible to get this to work without self. in front of the method?
This has absolutely nothing to do with blocks or dynamically defined methods. It's just simple basic Ruby syntax:
foo = bar
is local variable assignment. Always.
self.foo = bar
is a method call.
Is it possible to get this to work without self. in front of the method?
No.
This is just basic Ruby syntax. The define_method metaprogramming, the instance_exec, the blocks in your code sample are just a red herring, the problem can be demonstrated with a much simpler example:
def foo=(*)
puts 'I was called!'
end
foo=('bar') # even removing spaces and adding parentheses won't help!
self.foo = 'bar'
# I was called!
Note also that foo= is private but was actually called with an explicit receiver (which is illegal for private methods). That's a special exception for setter methods in the rule for private methods, because they otherwise couldn't be called at all, precisely because they would always be interpreted as a local variable assignment.