On Jan 25, 2007, at 4:50 PM, Jos B. wrote:
While I’m grateful for Pit’s suggestion it’s a rather suboptimal
solution to
the problem imho, and I wish this were possible in Ruby in a
straightforward
manner somehow.
ruby-core: this was originally just a response to Jos, but it has some
thoughts related to the public/protected/private discussion on ruby-core
The problem with the import-module solution is that it is ultimately
manipulating a global data structure (the class/module hierarchy) so
that the visibility of the changes (even if temporary) is defined
temporally and globally rather then lexically localized to particular
modules or classes.
One of the interesting aspects of Plan 9 is the idea of per-process
namespaces. In Plan 9, the file system is not a global data structure
that is visible to and shared by all processes. Instead, each
process constructs its own private view of the world. There are various
mechanisms to allow processes to share views (default) or to create
and manipulate their own private hierarchy (think of per-process
symlinks).
Switching back to Ruby, I wonder if the Plan 9 concepts of private
file hierarchies could be adapted to permit per-module/class views of
the
class hierarchy?
class A
enhance String, MyModule
def foo
“bar”.my_method # my_method is resolved to MyModule#mymethod
# because the enhancement is visible in
A#foo
B.new.baz
end
end
class B
def baz
“bar”.my_method # NameError because the enhancement is not
# visible in B.
end
end
The idea is that when methods defined in A are executing, they
see the String class enhanced (i.e. included) via MyModule. Methods
that are not explicitly defined within A don’t see the enhancements.
I feel like this sort of thing is related to the ongoing discussion of
private over on ruby-core where the goal is to figure out how to keep
private methods visible only in a particular lexical scope (at least I
think that is the/a goal).
How about:
class A
enhance self do
def private_method_1
end
end
def public_method
private_method_1
end
end
A.new.private_method_1 # NameError
My main point is that as long as the class-hierarchy is considered a
global
data structure you are going to have a hard time dealing with name
clashes.
Having a way to construct lexically scoped class-hierarchies may be a
way around
this problem.
Gary W.