Bug #15877: Incorrect constant lookup result in method on cloned class

• Status: Closed
• Assignee: ko1 (Koichi Sasada)
• Target version: -
• ruby -v: 2.6.3
• Backport: 2.4: UNKNOWN, 2.5: UNKNOWN, 2.6: REQUIRED

[danielwaterworth (Daniel Waterworth)]

This behavior seems wrong to me:

class Foo
  def test
    TEST
  end
end

Bar1 = Foo.clone
Bar2 = Foo.clone

class Bar1
  TEST = 'bar-1'
end

class Bar2
  TEST = 'bar-2'
end

# If these two lines are reordered, 'bar-2' is produced each time
p [:bar1_method, Bar1.new.test] # outputs 'bar-1' (correct)
p [:bar2_method, Bar2.new.test] # outputs 'bar-1' (incorrect)

p [:bar1_const, Bar1::TEST] # outputs 'bar-1' (correct)
p [:bar2_const, Bar2::TEST] # outputs 'bar-2' (correct)

Possibly related to #9603, #7107

[danielwaterworth (Daniel Waterworth)]

Running with RubyVM::InstructionSequence.compile_option = {inline_const_cache: false} produces the correct result.

[danielwaterworth (Daniel Waterworth)]

• Description updated (diff)

[sag (Stephen Gregory)]

I tested this on 2.5, 2.4 and the behavior is the same there. Also this applies to subclasses as well. The first class READ determines the values for all of them.

class Unrelated
    TEST='UNRELATED'
    def test
      TEST
    end
end

class Foo
  def test
     TEST
   end
end
  
Bar2 = Foo.clone
Bar1 = Foo.clone
  

class Bar1
  TEST = 'bar-1'
end
class Bar2
  TEST = 'bar-2'
end
class Bar3 < Foo
  TEST = 'bar-3'
end

puts "Bar2 #{Bar2.new.test}"  # Bar2 bar-2
puts "Bar1 #{Bar1.new.test}"   # Bar1 bar-2
puts "Bar3 #{Bar3.new.test}"  # Bar3 bar-2
puts "Foo #{Foo.new.test}"     # Foo bar-2
puts "Unrelated #{Unrelated.new.test}"  #Unrelated UNRELATED

Interestingly if Bar3 is read first then you get:

sample/array_second.rb:12:in `test': uninitialized constant Foo::TEST (NameError)
	from sample/array_second.rb:30:in `<main>'

[ko1 (Koichi Sasada)]

• Status changed from Open to Assigned
• Assignee set to ko1 (Koichi Sasada)

oops!

[ko1 (Koichi Sasada)]

• Status changed from Assigned to Closed

Applied in changeset git|b004d3e8300ba803d4a499148fa4fc6a690149e6.

---

solve "duplicate :raise event" [Bug #15877]

Without this patch, "raise" event invoked twice when raise an
exception in "load"ed script.
This patch by danielwaterworth (Daniel Waterworth).
[Bug #15877]

[nagachika (Tomoyuki Chikanaga)]

• Status changed from Closed to Assigned
• Backport changed from 2.4: UNKNOWN, 2.5: UNKNOWN, 2.6: UNKNOWN to 2.4: UNKNOWN, 2.5: UNKNOWN, 2.6: REQUIRED

Re-open. This ticket has been closed accidentally by b004d3e8.
I'll correct commit link.

[nobu (Nobuyoshi Nakada)]

• Status changed from Assigned to Closed

Applied in changeset git|76bd0714cf1140ffd64bf564446c76c54f2c4870.

---

solve "duplicate :raise event" in require too [Bug #15877]

[nobu (Nobuyoshi Nakada)]

• Status changed from Closed to Open

I copied the wrong commit log, sorry.

[ko1 (Koichi Sasada)]

• Status changed from Open to Closed

Applied in changeset git|71efad1ed391ee0c5398a76306fdbaaadd4dc52e.

---

introduce RCLASS_CLONED flag for inline cache.

Methods on duplicated class/module refer same constant inline
cache (IC). Constant access lookup should be done for cloned
class/modules but inline cache doesn't check it.
To check it, this patch introduce new RCLASS_CLONED flag which
are set when if class/module is cloned (both orig and dst).
[Bug #15877]

[ko1 (Koichi Sasada)]

solve "duplicate :raise event" [Bug #15877]

sorry my mistake :p