Bug #10191: Possible memory leak using dup and setting an unassigned instance variable (Windows)

• Status: Closed
• Assignee: -
• Target version: -
• ruby -v: 1.9.3p327 (2012-11-10) [i386-mingw32] (also present in 2.0.0p481 [i386-mingw32] from rubyinstaller.org)
• Backport: 2.0.0: DONE, 2.1: DONE

[benhell (Ben Hellerstein)]

Steps:

1. Run the following code (tested on Window 7):

class Leaky
  attr_accessor :v
  def leak; d = dup; d.v = nil; end
end

l = Leaky.new
while true; l.leak; end

2. Watch the memory usage in Task Manager or equivalent
   Expected result:
   Infinite loop, memory usage fairly stable (as GC cleans up)

Actual result:
Interpreter consumes more and more memory, and eventually throws the following exception: "in `dup': failed to allocate memory (NoMemoryError)." Presumably, this is a memory leak.

Workaround:
If you manually set the instance variable before dup-ing, memory usage is stable. Accordingly, this code does not appear to leak memory:

class NonLeaky
  attr_accessor :v
  def leak; @v ||= nil; d = dup; d.v = nil; end
end

l = NonLeaky.new
while true; l.leak; end

[nobu (Nobuyoshi Nakada)]

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

[nobu (Nobuyoshi Nakada)]

• Description updated (diff)

[nobu (Nobuyoshi Nakada)]

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

Applied in changeset r47372.

---

object.c: fix memory leak

• object.c (rb_obj_copy_ivar): allocate no memory for empty
  instance variables. [ruby-core:64700] [Bug #10191]

[nagachika (Tomoyuki Chikanaga)]

• Backport changed from 2.0.0: REQUIRED, 2.1: REQUIRED to 2.0.0: REQUIRED, 2.1: DONE

r46501, r47372 and r47460 were backported into ruby_2_1 branch at r47520.
I add limit: 2.5 option argment to assert_no_memory_leak in the testcase because it fails with default threshold (1.5) on 2.1.

[usa (Usaku NAKAMURA)]

• Backport changed from 2.0.0: REQUIRED, 2.1: DONE to 2.0.0: DONE, 2.1: DONE

backported into ruby_2_0_0 at r47548.

In my environment, I don't have to change the limit of assert_no_memory_leak.
I may change it if CIs report some problems.