Bug #13305: Occasional segfaults after defining methods while running coverage

• Status: Closed
• Assignee: mame (Yusuke Endoh)
• Target version: -
• ruby -v: ruby 2.4.0p0 (2016-12-24 revision 57164) [x86_64-darwin16], ruby 2.5.0dev (2017-03-13 trunk 57944) [x86_64-darwin16]
• Backport: 2.2: DONTNEED, 2.3: DONTNEED, 2.4: DONE

[decuplet (Nikita Shilnikov)]

The problem occurs on MRI 2.4.0 and the latest trunk only (2017-03-13 trunk 57944). After running Coverage.start code defines the same method few times but every time the size of the method increases. Occasionally MRI can fail on attempt to get the coverage result. The bug can be reproduced quite easily with the following two files

a.rb

require 'coverage'
Coverage.start

require_relative './b'

add_method(Class.new)

Coverage.result

b.rb

REPEATS = 400

def add_method(target)
  REPEATS.times do
    target.class_eval(<<~RUBY, __FILE__, __LINE__ + 1)
      def foo
        #{"\n" * rand(REPEATS)}
      end
      1
    RUBY
  end
end

Run ruby a.rb to reproduce the bug. Several runs may be required before MRI fails, but for me it fails on almost every run.

Note that passing filename and line number to class_eval is important, so as adding 1 after the method definition.

The crash report attached to the issue.

[decuplet (Nikita Shilnikov)]

• Description updated (diff)

[shyouhei (Shyouhei Urabe)]

• Status changed from Open to Assigned
• Assignee set to mame (Yusuke Endoh)

[mame (Yusuke Endoh)]

Thank you for the report. I could reproduce the issue. I'll take a look later.

[mame (Yusuke Endoh)]

• Status changed from Assigned to Closed

Applied in changeset r57971.

---

Fix a consistency bug of ISEQ_COVERAGE [Bug #13305]

There is an invariant that ISEQ_COVERAGE(iseq) must be Qnil if and only
if option->coverage_enabled is false. This invariant was broken by
NODE_PRELUDE which updates option->coverage_enabled but not
ISEQ_COVERAGE(iseq).

[mame (Yusuke Endoh)]

I think I could fix the issue. Nikita, thank you for the very useful report!

[decuplet (Nikita Shilnikov)]

mame (Yusuke Endoh) wrote:

I think I could fix the issue. Nikita, thank you for the very useful report!

Awesome! Thank you, can confirm, I tested the latest trunk in the environment where I had found the issue and the build passed https://travis-ci.org/rom-rb/rom-sql/builds/211705504 It fails on MRI 2.4.0-p0, but that's OK.

[wanabe (_ wanabe)]

• Related to Bug #13594: Segfault in gc_sweep_step 2.4.1 added

[nelhage (Nelson Elhage)]

Can this fix be backported to the Ruby 2.4 release series? 2.4.3 is still affected by this crash.

[hsbt (Hiroshi SHIBATA)]

• Backport changed from 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: UNKNOWN to 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: REQUIRED

[nagachika (Tomoyuki Chikanaga)]

• Backport changed from 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: REQUIRED to 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: DONE

ruby_2_4 r61414 merged revision(s) 57971.

[usa (Usaku NAKAMURA)]

• Backport changed from 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: DONE to 2.2: DONTNEED, 2.3: DONTNEED, 2.4: DONE