Bug #10314: Default argument lookup fails in Ruby 2.2 for circular shadowed variable names

• Status: Closed
• Assignee: nobu (Nobuyoshi Nakada)
• Target version: -
• ruby -v: 2.2.0-preview1
• Backport: 2.0.0: DONTNEED, 2.1: DONTNEED

[lsegal (Loren Segal)]

The following code prints nil in Ruby 2.2.0-preview1 but worked in all previous version of Ruby back to 1.8.7:

class Foo
  def foo; "abc" end
 
  # this default param should resolve at runtime to the #foo method call
  def run(foo = foo)
    p foo # print shadowed local var defaulting to attr value
  end
end
 
puts "Testing #{RUBY_VERSION}:"
Foo.new.run

# Ruby 2.2.0-preview1
# => nil

# Ruby 1.x 2.x etc
# => "abc"

My guess is this is happening because "foo" in "foo = foo" is resolving to the argument variable "foo", which currently has the value of nil. It would be equivalent to setting "qux = qux" in a method body, which has been the expected behavior for a long time.

I understand that shadowing variables is something you should probably never do, but unfortunately this code was already written and working for quite a while, so I figured it would be wise to file a bug report for the following reasons:

1. This seems like a breaking change in Ruby 2.2.0-preview1 that was not announced in the changelog. My guess is this change may have been unintentional, but if it was we need a changelog entry at the very least.
2. If this is newly expected behavior, I wanted to chime in that I don't think it makes much sense. I can't think of any time when a user would expect the default value of a "foo = foo" argument to be the same foo argument itself. That would be tautologically nil. Arguably, it doesn't make much sense inside of a method body either when there is a shadowed method that could be called instead.

[ko1 (Koichi Sasada)]

Loren Segal wrote:

1. This seems like a breaking change in Ruby 2.2.0-preview1 that was not announced in the changelog. My guess is this change may have been unintentional, but if it was we need a changelog entry at the very least.

I agree. How about to show a warning?

2. If this is newly expected behavior, I wanted to chime in that I don't think it makes much sense. I can't think of any time when a user would expect the default value of a "foo = foo" argument to be the same foo argument itself. That would be tautologically nil. Arguably, it doesn't make much sense inside of a method body either when there is a shadowed method that could be called instead.

I think this change is to make same as method body ("foo = foo" in method body).

I'm not sure your proposal is natural or not (matz issue?).
I like current behavior because it is simple (define a local variable foo' after the sentence foo=').

[hsbt (Hiroshi SHIBATA)]

• Related to Bug #9593: Keyword arguments default argument assignment behaviour not consistent with optional argument added

[hsbt (Hiroshi SHIBATA)]

• Related to Bug #10280: Regression while evaluating default argments of a method added

[matz (Yukihiro Matsumoto)]

• Status changed from Open to Rejected

In other part of Ruby, foo = foo makes foo to nil.
To get old behavior, try foo = foo().

Matz.

[ko1 (Koichi Sasada)]

• Status changed from Rejected to Assigned
• Assignee set to nobu (Nobuyoshi Nakada)

Maybe we need warning() strongly (without verbose).

[nobu (Nobuyoshi Nakada)]

• Status changed from Assigned to Closed
• % Done changed from 0 to 100

Applied in changeset r48188.

---

parse.y: warn circular argument reference

• parse.y (gettable_gen): warn circular argument reference, for
  transition from 2.1 and earlier. [ruby-core:65990] [Bug #10314]

[tmm1 (Aman Karmani)]

This change in behavior breaks rails 3.2, for example here: https://github.com/github/rails/blob/3-2-github/activerecord/lib/active_record/associations/has_many_association.rb#L53-L55

In this case, I think the intended behavior is def has_cached_counter?(reflection = self.reflection) to access the reflection attributed defined in the superclass: https://github.com/github/rails/blob/3-2-github/activerecord/lib/active_record/associations/association.rb#L21

Is adding an explicit self. prefix necessary to preserve compatibility in 2.2 and forward?

[tmm1 (Aman Karmani)]

I see, @matz (Yukihiro Matsumoto) recommended foo = foo(). I guess this is a breaking change in ruby 2.2.

[hsbt (Hiroshi SHIBATA)]

tmm1

Please request to backport has_many_association patch into Rails 3.2.

[tmm1 (Aman Karmani)]

I opened a pull request to add ruby 2.2 compatibility to rails 3-2-stable: https://github.com/rails/rails/pull/18160.

In our large rails app, I only had to fix 5 function definitions, by adding parenthesis to 4 of them. The warnings made it very easy to find and fix the code.

[usa (Usaku NAKAMURA)]

• Backport changed from 2.0.0: UNKNOWN, 2.1: UNKNOWN to 2.0.0: DONTNEED, 2.1: DONTNEED

[funny_falcon (Yura Sokolov)]

It looks like it is better to make it an error than a warning.

[ko1 (Koichi Sasada)]

• Status changed from Closed to Open

Nobu, what do you think about funny falcon's idea?

[nobu (Nobuyoshi Nakada)]

I'm in favor of that error.

[olivierlacan (Olivier Lacan)]

nobu (Nobuyoshi Nakada) wrote:

I'm in favor of that error.

Do you still want to implement this error?

Also, wasn't def run(foo = self.foo) a viable alternative to def run(foo = foo()) to avoid circular reference warnings?

Empty parens tend to trigger errors from tools like Rubocop.

[nobu (Nobuyoshi Nakada)]

olivierlacan (Olivier Lacan) wrote:

Also, wasn't def run(foo = self.foo) a viable alternative to def run(foo = foo()) to avoid circular reference warnings?

Unless #foo is private.

Empty parens tend to trigger errors from tools like Rubocop.

Then you should report it to Rubocop, not here.

[naruse (Yui NARUSE)]

• Target version deleted (2.2.0)

[jeremyevans0 (Jeremy Evans)]

• File circular-arg-ref-syntax-error-10314.patch circular-arg-ref-syntax-error-10314.patch added

Attached is a patch to turn circular argument reference from a warning to a SyntaxError. Do we still want to do that?

[matz (Yukihiro Matsumoto)]

I think it's OK now to make this error.

Matz.

• Status changed from Open to Closed

Applied in changeset git|0162e7e6471b639dfeeded29943e9e27c9519826.

---

Make circular argument reference a SyntaxError instead of a warning

Fixes [Bug #10314]