unclosed subscribe segfaults the rubyvm

[eqv]

This sample will segfault the ruby vm:

require 'redis'
require 'celluloid/io'
require 'celluloid/redis'

    class Runner
        include Celluloid::IO
        attr_accessor :redis
        def initialize
            async.run
        end

        def run()
            @redis = ::Redis.new(:driver => :celluloid)
            @redis.subscribe("foo", "bar") {}
        end
    end
    r = Runner.new
    exit #crashes since r.redis is still subscribed

[tarcieri]

Pretty much every Celluloid-implicated segv so far has been due to MRI
bugs. Can you post the version of MRI you're using and the complete stack
trace including the segv components?

Also: try MRI 2.0 or JRuby and see if that solves the segvs

Tony Arcieri

On Jun 5, 2013, at 1:34 AM, ranmrdrakono [email protected] wrote:

This sample will segfault the ruby vm:

require 'redis'require 'celluloid/io'require 'celluloid/redis'

class Runner
    include Celluloid::IO
    attr_accessor :redis
    def initialize
        async.run
    end

    def run()
        @redis = ::Redis.new(:driver => :celluloid)
        @redis.subscribe("foo", "bar") { |on|
            on.message do |channel, message| redis.unsubscribe end
        }
    end
end
r = Runner.new
exit

—
Reply to this email directly or view it on
GitHubhttps://github.com//issues/8
.

[eqv]

crash log for 1.9.3
https://gist.github.com/ranmrdrakono/304a07d514488b08c3be

ruby version:

ruby -v 
ruby 1.9.3p194 (2012-04-20 revision 35410) [x86_64-linux]

gem versions:

gem list --local

*** LOCAL GEMS ***
 [...]
celluloid (0.14.1)
celluloid-io (0.14.1)
celluloid-redis (0.0.2)
 [...]
nio4r (0.4.6)
 [...]
rubygems-bundler (1.2.0, 1.1.0)
rvm (1.11.3.8, 1.11.3.5)

also: problem remains in ruby2.0:
https://gist.github.com/ranmrdrakono/b48334e74e55d833a162

[tarcieri]

Okay, thank you, this looks like it might actually be an nio4r bug at a cursory glance

[eqv]

should I resubmit the issue at nio4r?

[tarcieri]

Sure, that'd be great. It'd also be great if you can try JRuby and see what problems (if any) you're having there

[eqv]

while it does not produce a segfault under JRuby, it still crashes, which I assume to be a bug as well:
(https://gist.github.com/ranmrdrakono/c5dad69ed388f990f176)

[tarcieri]

This make it look like we aren't safely detecting when something is trying register IO objects on a closed selector. That's likely a corner case in the nio4r API that isn't tested, and breaking in different ways on both the libev and Java NIO backends of nio4r.

Closing this issue in favor of the nio4r one