Bug #13107: def_delegators causes random errors in MRI 2.4.0 - Ruby master - Ruby Issue Tracking System

Actions

Copy link

Bug #13107

closed

def_delegators causes random errors in MRI 2.4.0

Added by twalpole@gmail.com (Thomas Walpole) almost 8 years ago. Updated over 7 years ago.

Status:

Closed

Assignee:

Target version:

ruby -v:

Backport:

2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: DONE

[ruby-core:78981]

Description

In the Capybara project we use the rack_test gem which uses def_delegators to forward flast_reponse to @rack_mock_session - https://github.com/brynary/rack-test/blob/master/lib/rack/test.rb#L29

When running the Capybara test test suite calling last_reponse on a rack_test Session object will sporadically result in a TypeError from forwardable.rb

TypeError:
       wrong argument type Integer (expected Proc)

Patching around the delegator with

class Rack::Test::Session
  def last_response
    @rack_mock_session.last_response
  end
end

stops the error from occurring, so it appears that something def_delegators is doing is causing problems. Unfortunately I have not yet been able to establish exactly what causes the issues to produce a simple test case.

Files

Download all files

0001-forwardable.rb-use-public_send-to-call-method.patch (789 Bytes) 0001-forwardable.rb-use-public_send-to-call-method.patch		eugeneius (Eugene Kenny), 01/08/2017 01:48 AM
0001-forwardable-impl.rb-include-trace-instruction.patch (773 Bytes) 0001-forwardable-impl.rb-include-trace-instruction.patch		eugeneius (Eugene Kenny), 01/08/2017 02:20 AM

Actions

Copy link

#1 [ruby-core:78982]

Updated by twalpole@gmail.com (Thomas Walpole) almost 8 years ago

Forgot to add the location reported for the error

/home/travis/.rvm/rubies/ruby-2.4.0/lib/ruby/2.4.0/forwardable.rb:228:in `last_response'

Actions

Copy link

#2 [ruby-core:79010]

Updated by eugeneius (Eugene Kenny) almost 8 years ago

File 0001-forwardable.rb-use-public_send-to-call-method.patch 0001-forwardable.rb-use-public_send-to-call-method.patch added

The mongo gem is also affected by this bug (https://jira.mongodb.org/browse/RUBY-1191). Running the tests in the spec/mongo/grid/stream/read_spec.rb file reliably reproduces it.

I bisected the changes to forwardable.rb between v2.3.0 and v2.4.0, and determined that the bug was introduced in r55376.

I've attached a patch that fixes the problem, although it effectively reverts the optimisation from r55376, so there may be a performance impact.

Actions

Copy link

#3 [ruby-core:79012]

Updated by eugeneius (Eugene Kenny) almost 8 years ago

File 0001-forwardable-impl.rb-include-trace-instruction.patch 0001-forwardable-impl.rb-include-trace-instruction.patch added

Here's another patch that also fixes the problem, while keeping the optimisation from r55376.

I'll admit that I have no idea what this option does or how it could cause this bug - I noticed that the "portable" implementation of forwardable/impl.rb wasn't affected, and fiddled with the MRI-specific one until it worked. Please review this patch carefully if you intend to use it!

Removing tailcall_optimization also appears to work, although I think we need that option to get a sensible stack trace. Maybe trace_instruction: false, tailcall_optimization: true is an invalid combination?

Actions

Copy link

#4 [ruby-core:79015]

Updated by nobu (Nobuyoshi Nakada) almost 8 years ago

Status changed from Open to Feedback

Enabling trace instruction disables tail call optimization, too.
It seems like a bug of the optimization and GC.
How can I reproduce it briefly?

Actions

Copy link

#5 [ruby-core:79016]

Updated by nobu (Nobuyoshi Nakada) almost 8 years ago

Description updated (diff)

Actions

Copy link

Updated by nobu (Nobuyoshi Nakada) almost 8 years ago

Status changed from Feedback to Closed

Applied in changeset r57293.

vm_insnhelper.c: block argument at tailcall

vm_insnhelper.c (vm_call_iseq_setup_tailcall): check interrupts
after set up the new frame, not the passed block to be clobbered
by invoked finalizers and so on. [ruby-core:78981] [Bug #13107]

Actions

Copy link

#7 [ruby-core:79018]

Updated by nobu (Nobuyoshi Nakada) almost 8 years ago

When interrupted by a finalizer during setting up tailcall frame, a block argument on the stack was overwritten by the magic number in the finalizer frame which looks an Integer.
I tried in vain to make a test case, but a failure in mongo-ruby-driver seems fixed.
Could you try the latest trunk?

Actions

Copy link

#8 [ruby-core:79030]