Bug #2330: Non systematic segmentation fault with autoload rubyspec

• Status: Closed
• Assignee: marcandre (Marc-Andre Lafortune)
• Target version: 1.9.2
• ruby -v: ruby 1.9.2dev (2009-11-04 trunk 25641) [x86_64-darwin10.0.0]

[marcandre (Marc-Andre Lafortune)]

=begin
Starting from r25601, the following rubyspec will crash about half of the time (just repeat a couple of times):

$ mspec -t rubydev core/kernel/autoload_spec.rb
ruby 1.9.2dev (2009-11-04 trunk 25641) [x86_64-darwin10.0.0]
.........../Users/work/mspec/lib/mspec/mocks/proxy.rb:8: [BUG] Segmentation fault
ruby 1.9.2dev (2009-11-04 trunk 25641) [x86_64-darwin10.0.0]

-- control frame ----------
c:0036 p:0024 s:0121 b:0121 l:000120 d:000120 METHOD /Users/work/mspec/lib/mspec/mocks/proxy.rb:8
c:0035 p:---- s:0115 b:0115 l:000114 d:000114 FINISH
c:0034 p:---- s:0113 b:0113 l:000112 d:000112 CFUNC :autoload
c:0033 p:0057 s:0108 b:0108 l:000208 d:000107 BLOCK /Users/work/ruby/dev/spec/rubyspec/core/kernel/autoload_spec.rb:105
c:0032 p:---- s:0105 b:0105 l:000104 d:000104 FINISH
c:0031 p:---- s:0103 b:0103 l:000102 d:000102 CFUNC :instance_eval
c:0030 p:0017 s:0100 b:0100 l:000099 d:000099 METHOD /Users/work/mspec/lib/mspec/runner/mspec.rb:67
c:0029 p:0021 s:0094 b:0094 l:000081 d:000093 BLOCK /Users/work/mspec/lib/mspec/runner/context.rb:135
c:0028 p:---- s:0091 b:0091 l:000090 d:000090 FINISH
c:0027 p:---- s:0089 b:0089 l:000084 d:000088 IFUNC :instance_variable_get
c:0026 p:---- s:0087 b:0087 l:000086 d:000086 CFUNC :each
c:0025 p:---- s:0085 b:0085 l:000084 d:000084 CFUNC :all?
c:0024 p:0053 s:0082 b:0082 l:000081 d:000081 METHOD /Users/work/mspec/lib/mspec/runner/context.rb:135
c:0023 p:0101 s:0076 b:0076 l:000065 d:000075 BLOCK /Users/work/mspec/lib/mspec/runner/context.rb:163
c:0022 p:---- s:0071 b:0071 l:000070 d:000070 FINISH
c:0021 p:---- s:0069 b:0069 l:000068 d:000068 CFUNC :each
c:0020 p:0127 s:0066 b:0066 l:000065 d:000065 METHOD /Users/work/mspec/lib/mspec/runner/context.rb:155
c:0019 p:0109 s:0063 b:0063 l:000062 d:000062 METHOD /Users/work/mspec/lib/mspec/runner/mspec.rb:36
c:0018 p:0029 s:0056 b:0056 l:000055 d:000055 METHOD /Users/work/mspec/lib/mspec/runner/object.rb:11
c:0017 p:0167 s:0049 b:0049 l:000208 d:000208 TOP /Users/work/ruby/dev/spec/rubyspec/core/kernel/autoload_spec.rb:69
c:0016 p:---- s:0047 b:0047 l:000046 d:000046 FINISH

mspec version: 1.5.12
rubyspec: current
=end

[nobu (Nobuyoshi Nakada)]

=begin
Hi,

At Wed, 4 Nov 2009 14:36:14 +0900,
Marc-Andre Lafortune wrote in [ruby-core:26523]:

Starting from r25601, the following rubyspec will crash about half of the time (just repeat a couple of times):

$ mspec -t rubydev core/kernel/autoload_spec.rb
ruby 1.9.2dev (2009-11-04 trunk 25641) [x86_64-darwin10.0.0]
.........../Users/work/mspec/lib/mspec/mocks/proxy.rb:8: [BUG] Segmentation fault
ruby 1.9.2dev (2009-11-04 trunk 25641) [x86_64-darwin10.0.0]

Worked fine 20 times.

$ make MSPECOPT=core/kernel/autoload_spec.rb test-rubyspec
./miniruby -I../src/lib -I../ext/common ../src/tool/runruby.rb --extout=../ext -- ../src/spec/mspec/bin/mspec run -B ../src/spec/default.mspec core/kernel/autoload_spec.rb
ruby 1.9.2dev (2009-11-02 trunk 25629) [x86_64-darwin9.8.0]
..............

Finished in 0.038309 seconds

1 file, 14 examples, 16 expectations, 0 failures, 0 errors

--
Nobu Nakada

=end

[marcandre (Marc-Andre Lafortune)]

• Priority changed from 6 to 5

=begin
Can't reproduce on PowerBook with Leopard, nor on a MacBook Pro with Leopard.

On my Mac Pro with Snow Leopard, though, it's pretty systematic. I rebuilt the nightly download from scratch and got the same crash Maybe a strange 64 bit issue?

I'll try tomorrow on a different machine with Snow Leopard.
=end

[marcandre (Marc-Andre Lafortune)]

=begin
Reproduced on a MacBook Pro (thus a different processor) running Snow Leopard, same error.

Is there a compile option I can use to compile ruby in 32 bit instead of 64?
=end

[naruse (Yui NARUSE)]

• Status changed from Open to Assigned
• Assignee changed from matz (Yukihiro Matsumoto) to marcandre (Marc-Andre Lafortune)
• Priority changed from 5 to 3

=begin

=end

[marcandre (Marc-Andre Lafortune)]

• Assignee changed from marcandre (Marc-Andre Lafortune) to naruse (Yui NARUSE)
• Priority changed from 3 to 5

=begin
I'm quite flattered to have been assigned this issue, but I'm afraid my skills are being overestimated.

Here's the minimal test case to reproduce:
$ cat load_test.rb
load 'test.rb'

$ cat test.rb
def Kernel.load
end
call_method_that_doesnt_exist

$ rubydev load_test.rb
test.rb:3: [BUG] Segmentation fault
ruby 1.9.2dev (2009-12-27 trunk 26184) [x86_64-darwin10.2.0]

-- control frame ----------
c:0006 p:0024 s:0014 b:0014 l:000013 d:000013 TOP test.rb:3
c:0005 p:---- s:0012 b:0012 l:000011 d:000011 FINISH
^Z
[4]+ Stopped rubydev temp.rb
$ kill -9 %
[4]+ Killed rubydev temp.rb

Crash happens on line 751 of vm.c (vm_backtrace_each) because cfp->me->def is 0:
if ((*iter)(arg, file, line_no, rb_id2str(cfp->me->def->original_id))) break;

As stated previously, this bug is present since r25601.

Note: I've taken the liberty to reset the priority of this issue to High, but maybe I'm mistaken in thinking that any reproducible crashing bug should be at least at that priority?
=end

[nobu (Nobuyoshi Nakada)]

=begin
Hi,

At Sun, 27 Dec 2009 17:00:43 +0900,
Marc-Andre Lafortune wrote in [ruby-core:27335]:

$ rubydev load_test.rb
test.rb:3: [BUG] Segmentation fault
ruby 1.9.2dev (2009-12-27 trunk 26184) [x86_64-darwin10.2.0]

It doesn't reproduce on darwin9.8.0. Maybe gcc version
dependent?

Does the binary compiled on darwin10 segfault on darwin9, or
vice versa?

--
Nobu Nakada

=end

[marcandre (Marc-Andre Lafortune)]

• Assignee changed from naruse (Yui NARUSE) to marcandre (Marc-Andre Lafortune)

=begin

=end

[nobu (Nobuyoshi Nakada)]

=begin
Hi,

At Sat, 30 Jan 2010 06:58:25 +0900,
Marc-Andre Lafortune wrote in [ruby-core:27959]:

1. The minimal test case
   I wonder why I didn't copy-paste my actual original code. In any case,
   the bug happens if the instance method #load is redefined in any way
   (and not the module method, as in my posted example). Thus the file
   'load.rb' should read:

module Kernel
def load
end
end
generate_any_exception

It seems like same as [ruby-dev:40234].

I think it should use called_id instead of original_id to
traverse the backtrace, but currently called_id is not set
always properly. I loose a patch to fix it now.

Index: vm.c

--- vm.c (revision 26470)
+++ vm.c (working copy)
@@ -734,5 +734,5 @@ vm_backtrace_each(rb_thread_t *th, int l
else if (RUBYVM_CFUNC_FRAME_P(cfp)) {
if (NIL_P(file)) file = rb_str_new_cstr("ruby");

•  if ((*iter)(arg, file, line_no, rb_id2str(cfp->me->def->original_id))) break;
  

•  if ((*iter)(arg, file, line_no, rb_id2str(cfp->me->called_id))) break;
  

  }
  cfp = RUBY_VM_NEXT_CONTROL_FRAME(cfp);
  

--
Nobu Nakada

=end

[marcandre (Marc-Andre Lafortune)]

• Status changed from Assigned to Closed

=begin
Crash resolved by r26718
=end