Please refer to https://www.ruby-forum.com/topic/143840 for background. The consensus in that thread appears to be that Thread#raise, Thread#kill, and timeout.rb are unsafe. Aside from the issue of providing safe alternatives, would it not be a great service to the Ruby community, especially newcomers, to document this lack of safety in the relevant classes? If such documentation is thought to be worthwhile and we can agree on suitable wording, I will happily submit a pull request. Here's a first draft of the wording: # Warning: this method can produce unpredictable behaviour and its use # is not recommended. # See https://www.ruby-forum.com/topic/143840 for more information.
on 2013-09-12 10:08
on 2013-09-12 22:06
Thread from 2008? Seriously ruby was in 1.8 at the time. Ruby has check and wall already implemented as well as a bug trac system
on 2013-09-12 22:36
On Thu, Sep 12, 2013 at 1:08 AM, Glyn Normington <firstname.lastname@example.org>wrote: > Please refer to https://www.ruby-forum.com/topic/143840 for background. > The consensus in that thread appears to be that Thread#raise, > Thread#kill, and timeout.rb are unsafe. > Yes, they are unsafe. Ruby 2.0 has done some work to try to improve their safety at the cost of performance, but they are still unsafe on other Ruby VMs, particularly those without a GIL, and are probably still buggy on Ruby 2.0 as well. Aside from the issue of providing safe alternatives, would it not be a > great service to the Ruby community, especially newcomers, to document > this lack of safety in the relevant classes? Considering the work that has gone into features like Thread.handle_interrupt, ruby-core is probably unlikely to accept such a label, although it would probably make sense.
on 2013-09-13 10:04
Thanks Tony. I think the difficulty with Thread.handle_interrupt is that it, or something similar, would need to be used correctly across Ruby's class libraries, e.g. in the networking classes, in order for Thread#raise, thread#kill and timeout.rb to be safe for general use. I couldn't see any evidence of such activity in the last couple of years in https://github.com/ruby/ruby, although maybe I was unlucky in my search. Perhaps the majority of the libraries had already been written with timeout.rb's limitations in mind, but it's hard to know. But I take the point that the blanket warning is not likely to be acceptable.
on 2013-09-13 10:09
Yes, Stu, I know this is an old problem, but that doesn't necessarily mean it has been fixed. I'm relatively new to Ruby and just trying to work out how much to trust it.