Mixing in modules?


#1

hello,

given:

  1. class C that implements some arbitrary methods
  2. a database with code fragments stored as text

i create a variable ‘var’ of class C and then read some code fragments
from
the database. each fragment i read is evaluated in a dynamic anonymous
module, and this module is used to extend ‘var’:

var = C.new
… . .
m = Module.new
m.module_eval(code_fragment)
… . .
var.extend(m)

so if a fragment implements some methods that are already defined in C
mixing in effectively replaces the original methods with the ones
defined in
the fragment.

the problem:

class C defines an instance variable @x:

class C
def initialize; @x = ‘C’; end
def to_s; @x; end
end

i want to override ‘to_s’ method. but the new implementation that i have
in
my fragment still needs to refer to @x. so i do this:

fragment = “def to_s; foo(@x); end”

this works fine. but this does not seem like a good programming style. a
module refers to a variable that is defined in the object that this
module
happens to be mixed in to. can anyone comment on this? basically the
problem
is to be able to override the behaviour of ‘var’ at run time as code
fragments are read.

to give a concrete example of what i am trying to do, suppose you have a
database with text fields. each field is represented at run time by
class C
that can be converted to a string. the basic to_s implementation just
outputs the value of the text field. but i want to also store text
fields
that contain textile markup, or eruby markup or whatever, in other words
i
want to associate a type with each text field and i want the to_s method
to
behave differently depending on the type. but for to_s to be able to run
it
needs to get the value of the text field from C.

thanks
konstantin


#2

hello,

let me rephrase the question, may be this will get more responses:

is it a bad programming style in ruby to use ‘super’ in a module
definition?
that is, is it bad to invoke methods that will be defined in the object
that
will mix in this module?

thanks
konstantin


#3

konsu wrote:

class C defines an instance variable @x:

class C
def initialize; @x = ‘C’; end
def to_s; @x; end
end

i want to override ‘to_s’ method. but the new implementation that i have
in my fragment still needs to refer to @x.

I don’t see a (stylistic or whatever) problem in directly referring to
@x
from the module. Since it works, it must be supposed to be done like
this,
I guess.

Malte


#4

On 2/8/06, konsu removed_email_address@domain.invalid wrote:

is it a bad programming style in ruby to use ‘super’ in a module definition?

I don’t think so. I myself haven’t yet run into the need to call super
in a module method, but I don’t see why you shouldn’t be able to, as
long as you make that expectation clear in the documentation for
anyone that might try and mix it in.

that is, is it bad to invoke methods that will be defined in the object that
will mix in this module?

Definitely not. The implementation of the methods provided by
enumerable are based on the assumption that the class that is mixing
in Enumerable will provide a few bootstrap methods, such as each. If
you try and mix in Enumerable without defining each, things just won’t
work. The important thing is that this dependency/expectation is
provided up front in the documentation.

Regarding your original question on using @instance_vars in module
methods, I skipped over it at first since I wasn’t fully sure about
which part of your question was about modules in general and which was
about eval’ing code from the DB. I’ll try to answer the general
portion now…

There is nothing wrong with using @instance_vars in module methods
that are intended to be mixed in. You need to be careful though. I
consider two types of instance vars that a module method might use:

  • variables used exclusively by the module
  • variables the module expects the class to provide

In the first case, the module is keeping its own state for its
operations. The class that’s mixing in the module should not care
about these instance variables and should leave them alone. As such,
you need to be very careful that you don’t stomp on any of the class’
instance variables, and that it won’t stomp on yours. This requires
very careful naming.

In the second case, I’d consider refactoring the module method to use
accessor methods on the class and thus abstract away the
implementation. This makes it easier to then adapt other classes which
may have different implementations to use the same mixin. Then, of
course, document the required interface (as opposed to required
instance variables) as mentioned above.

Jacob F.


#5

DÅ?a Streda 08 Február 2006 19:38 konsu napísal:

is it a bad programming style in ruby to use ‘super’ in a module
definition? that is, is it bad to invoke methods that will be defined in
the object that will mix in this module?

Actually, I’ll go a bit further than Jacob and say that is -exactly-
what
mixins are supposed to do - enhance the functionality by providing some
extra
methods that build on existing class functionality. Mixins that are
completely independent from the class would best be separated into
stand-alone classes in the most cases.

As far as instance variables go, I tend to avoid making stateful mixins
myself, and prefer using accessor methods instead of directly accessing
the
including class. I could go on about decoupling logic in here, but it
pretty
much boils down to a matter of taste / religion.

David V.