Project

General

Profile

Actions

Feature #11911

closed

Immutable method definitions and/or static dispatch

Added by mlarraz (Matt Larraz) about 8 years ago. Updated about 8 years ago.

Status:
Feedback
Assignee:
-
Target version:
-
[ruby-core:72563]

Description

One of Ruby's biggest strengths is the ability for anyone, at any time, to redefine (almost) any behavior. But this is also one of its biggest weaknesses.
Ruby has a very liberal dynamic dispatch, so any method can be redefined anywhere in the code, meaning we can never have any guarantees about behavior.

Other languages with dynamic dispatch (like C++ or Java) also allow for static dispatch. In particular, Java has dynamic dispatch by default, with the final keyword marking a method as immutable.

In Ruby, this might look something like:

def foo
 'foo'
end

final :foo

# Raises an exception
def foo
 'bar'
end

I see this as analogous to freezing a string.

Note that if somebody really needs to overwrite an immutable method, they can still do so, just in a more explicit way:

undef_method :foo

# Works as expected
def foo
 'bar'
end

This eliminates some ambiguity.

I'm not sure how feasible this is (or whether this is the ideal syntax), but I'd like to hear what the community thinks of such a concept in general.

Updated by duerst (Martin Dürst) about 8 years ago

  • Status changed from Open to Feedback

Ruby's ability to change any method anytime, and C++/Java's ability to overwrite some methods in subclasses, are conceptionally quite different. Which one are you interested in, and why? What is your use case?

Updated by mlarraz (Matt Larraz) about 8 years ago

I suppose I'm talking specifically about the first, that is, the ability to change any method at any time.

The most obvious use case I can imagine is an application that wants to guarantee that it's running the stock stdlib, with no monkey patches. Given a large enough number of gems (or even files in the codebase itself), auditing all of them for monkey patches becomes expensive. Consistently auditing all of them to ensure no monkey patches get introduced becomes cost-prohibitive. In such a case, it might also be convenient to have a command-line flag that disables any modifications to the stdlib.

As a highly contrived example, a malicious gem author could hide a monkey patch in the middle of his codebase, overwriting Kernel#puts to spy on all of the application's output. There is presumably a non-negligible number of Ruby developers who would like to easily guard against something like this.

Actions

Also available in: Atom PDF

Like0
Like0Like0