Bug #15766

Crash in 2.4, 2.5 involving at_exit

Added by (Christian Boos) 6 days ago. Updated 1 day ago.

Target version:



I've written a piece of code which reproducibly triggers a crash in some versions of Ruby (2.4 and 2.5, but not 2.2, 2.3 nor 2.6).

As this involves Minitest, I've started by filing a bug report on GitHub:

However, as this triggers a crash, I've been advised to report it here as well. Also, I wasn't able to find a similar bug report.

I have not looked too deep, but it seems that the way Minitest deals with atexit handlers exposes a problem when there's a pending LocalJumpError.

You will find all the details in the Minitest bug 789, but let me summarize them again here:

require 'minitest/autorun'
require 'minitest/spec'

describe 'a simple crash reproducer' do
  before { return } # /!\
  it 'checks stuff' do

Executing the above with the default minitest (or any newer version) won't work as a local return is not expected from within a before call (same thing for after).

But depending on the version, we may get a crash in addition to the error report:

  • ruby 2.2.5p319 (2016-04-26 revision 54774) [x64-mingw32] => LocalJumpError: unexpected return (no crash)
  • ruby 2.3.3p222 (2016-11-21 revision 56859) [x64-mingw32] => LocalJumpError: unexpected return (no crash)
  • ruby 2.4.2p198 (2017-09-14 revision 59899) [x64-mingw32] => unexpected return (+ CRASH)
  • ruby 2.4.5p335 (2018-10-18 revision 65137) [x64-mingw32] => unexpected return (+ CRASH)
  • ruby 2.5.0p0 (2017-12-25 revision 61468) [x64-mingw32] => unexpected return (+ CRASH)
  • ruby 2.5.3p105 (2018-10-18 revision 65156) [x64-mingw32] => unexpected return (+ CRASH)
  • ruby 2.6.0p0 (2018-12-25 revision 66547) [x64-mingw32] => unexpected return (no crash)
  • ruby 2.6.1p33 (2019-01-30 revision 66950) [x64-mingw32] => unexpected return (no crash)

I'm mainly developing on Windows, and use the pre-built versions from,
but in order to be sure that this is not a Windows-specific issue, I also tried Ruby 2.5.0p0 and 2.5.1p57 on Linux and I get the same crash.

That crash always takes the same form:

D:/Workspace/install/windows64/ruby-2.4/lib/ruby/gems/2.4.0/gems/minitest-5.10.3/lib/minitest.rb:60: [BUG] Segmentation fault

-- Control frame information -----------------------------------------------
c:0003 p:---- s:0011 e:000010 CFUNC  :exit
c:0002 p:0027 s:0006 e:000005 BLOCK  D:/Workspace/install/windows64/ruby-2.4/lib/ruby/gems/2.4.0/gems/minitest-5.10.3/lib/minitest.rb:60 [FINISH]
c:0001 p:0000 s:0003 E:000520 (none) [FINISH]

As the problem doesn't happen with 2.6, I did hope that this problem would also be fixed in recent 2.4 and 2.5, but the problem is still present for 2.4.5p335 and 2.5.3p105 at least (latest versions from



Updated by (Christian Boos) 6 days ago

  • Subject changed from Crash in 2.4, 2.5 involving atexit to Crash in 2.4, 2.5 involving at_exit

Updated by MSP-Greg (Greg L) 6 days ago

Using a recent build (ruby 2.5.5p157 (2019-03-15 revision 67260) [x64-mingw32]), the code ran fine, or just like 2.6. I haven't checked on 2.4.

Using the current RubyInstaller release (ruby 2.5.3p105 (2018-10-18 revision 65156) [x64-mingw32]) I had the same behavior you reported, a SEGV.

Hence, it would seem that whatever issue is causing the SEGV, the fix has been backported...

Updated by MSP-Greg (Greg L) 1 day ago

I can repo the issues with

ruby 2.4.6p354 (2019-04-01 revision 67394) [x64-mingw32]

2.5 & 2.6 do not SEGV.

Also available in: Atom PDF