Bug #2309: Segfault, repeatable testcase, possibly related to [2295], [2019], [1937].

• Status: Closed
• Assignee: -
• ruby -v: ruby 1.9.1p243 (2009-07-16 revision 24175) [x86_64-linux]

[jxtps (Jxtps 435)]

=begin
Ruby 1.9.1 segfaults when calling ActiveRecord::Base.transaction, dump attached (that's what the line with the error does).

Testcase is very repeatable (happens every time I run my specs), but slightly non-deterministic: changing a few lines of code will change in which particular test it happens (multiple tests go through the same code-path, doesn't always happen every time you go through the same path), and though it is usually in the same spot in the code it has sometimes happened in other spots (deeper within the same call-chain).

Unfortunately the code is for a commercial project and I can't share it. Sorry. I can work with someone from the ruby-core community to e.g. run debug versions of ruby on the test case if that is of use.

Environment:

• Ubuntu 9.04 on a basic intel box
• Rails 2.3.4
• DB: PostgreSQL, with ruby-pg binding

*** LOCAL GEMS ***

aasm (2.1.3)
actionmailer (2.3.4)
actionpack (2.3.4)
activerecord (2.3.4)
activeresource (2.3.4)
activesupport (2.3.4)
bluecloth (2.0.5)
builder (2.1.2)
capistrano (2.5.9)
cucumber (0.3.104, 0.3.103)
diff-lcs (1.1.2)
fastthread (1.0.7)
gemcutter (0.1.6)
haml (2.2.5)
highline (1.5.1)
hoe (2.3.3)
hpricot (0.8.1)
json (1.1.9)
json_pure (1.1.9)
mislav-will_paginate (2.3.11)
net-scp (1.0.2)
net-sftp (2.0.2)
net-ssh (2.0.15)
net-ssh-gateway (1.0.1)
nokogiri (1.3.3)
passenger (2.2.5)
pg (0.8.0)
polyglot (0.2.9)
postgres-pr (0.6.1)
rack (1.0.0)
rails (2.3.4)
rake (0.8.7)
right_aws (1.10.0)
right_http_connection (1.2.4)
rspec (1.2.8)
rspec-rails (1.2.7.1)
rubyforge (2.0.0)
rubygems-update (1.3.5)
spork (0.7.2)
syntax (1.0.0)
term-ansicolor (1.0.4)
test-unit (2.0.3, 1.2.3)
treetop (1.4.2)
webrat (0.5.3)
=end

[nobu (Nobuyoshi Nakada)]

=begin
Hi,

At Fri, 30 Oct 2009 09:13:36 +0900,
Jxtps 435 wrote in [ruby-core:26422]:

Testcase is very repeatable (happens every time I run my
specs), but slightly non-deterministic: changing a few lines
of code will change in which particular test it happens
(multiple tests go through the same code-path, doesn't always
happen every time you go through the same path), and though
it is usually in the same spot in the code it has sometimes
happened in other spots (deeper within the same call-chain).

Sounds like a GC bug.

Unfortunately the code is for a commercial project and I
can't share it. Sorry. I can work with someone from the
ruby-core community to e.g. run debug versions of ruby on the
test case if that is of use.

Could you show the backtrace with gdb?

--
Nobu Nakada

=end

[naruse (Yui NARUSE)]

• Status changed from Open to Feedback

=begin

=end

[jxtps (Jxtps 435)]

• File core_dump_backtrace.txt core_dump_backtrace.txt added
• File ruby_stack_trace_for_core_dump.txt ruby_stack_trace_for_core_dump.txt added

=begin
Sorry for the delay - despite watching this issue I did not get an email notification? (nothing in the spam folder)

In any case, attached is the result of connecting to the ruby process with gdb, continuing it and then issuing the "where" command - I hope it helps!

(the instructions on http://www.justskins.com/forums/error-compile-1-8-0-and-1-8-1-pre2-134754.html were helpful in getting the gdb backtrace - might be worth tidying up just a bit and put on the wiki for future gdb backtracing reference?)
=end

[jeremyevans0 (Jeremy Evans)]

• Description updated (diff)
• Status changed from Feedback to Closed