Bug #3606: Thread.stop and puts fail to work as documented.

• Status: Closed
• Assignee: kosaki (Motohiro KOSAKI)
• Target version: 2.0.0
• ruby -v: ruby 1.9.3dev (2010-07-22 trunk 28707) [i686-linux]

[phasis68 (Heesob Park)]

=begin
I have tested following three codes which are the sample code in RDoc Documentation.

t1.rb

a = Thread.new { puts "a"; Thread.stop; puts "c" }
Thread.pass
puts "Got here"
a.run
a.join
sleep 1

t2.rb

a = Thread.new { print "a"; Thread.stop; print "c" }
Thread.pass
print "b"
a.run
a.join
sleep 1

t3.rb

c = Thread.new { Thread.stop; puts "hey!" }
c.wakeup
sleep 1

Expected output
t1.rb
a
Got here
c

t2.rb
abc

t3.rb
hey!

on Ruby 1.8.6

$ ruby -v
ruby 1.8.6 (2009-06-08 patchlevel 369) [i686-linux]
$ ruby -v t1.rb
a
Got here
c
$ ruby -v t2.rb
abc$ ruby -v t3.rb
hey!
$

Result
t1.rb ==> success
t2.rb ==> success
t3.rb ==> success

on Ruby 1.9.1

$ ruby -v
ruby 1.9.1p378 (2010-01-10 revision 26273) [i686-linux]
$ ruby t1.rb
aGot here

c
$ ruby t2.rb
$ ruby t3.rb
$

Result
t1.rb ==> fail
t2.rb ==> success
t3.rb ==> fail

On Ruby 1.9.3dev

duometis02@Duo02:~$ ruby -v
ruby 1.9.3dev (2010-07-22 trunk 28707) [i686-linux]
$ ruby t1.rb
Got herea

t1.rb:5:in `join': deadlock detected (fatal)
	from t1.rb:5:in `<main>'
$ ruby t2.rb
bat2.rb:5:in `join': deadlock detected (fatal)
	from t2.rb:5:in `<main>'
$ ruby t3.rb
$

Result
t1.rb ==> fail
t2.rb ==> fail
t3.rb ==> fail

Interesingly, the following code sometimes works and sometimes failed on Ruby 1.9.3.dev.
t4.rb

c = Thread.new { Thread.stop; puts "hey!" }
puts "Hi!"
c.wakeup
sleep 1

$ ruby -v
ruby 1.9.3dev (2010-07-22 trunk 28707) [i686-linux]
$ ruby t4.rb
Hi!
$ ruby t4.rb
Hi!
hey!
$ ruby t4.rb
Hi!
hey!
$ ruby t4.rb
Hi!
$
=end

[nobu (Nobuyoshi Nakada)]

• Category changed from core to doc

=begin

=end

[raggi (James Tucker)]

=begin

On 23 Jul 2010, at 05:55, Heesob Park wrote:

on Ruby 1.8.6

$ ruby -v
ruby 1.8.6 (2009-06-08 patchlevel 369) [i686-linux]
$ ruby -v t1.rb
a
Got here
c
$ ruby -v t2.rb
abc$ ruby -v t3.rb
hey!
$

Result
t1.rb ==> success
t2.rb ==> success
t3.rb ==> success

This is because threads in 1.8.6 are not really threads. The engine is predictable within the realms of ruby frames.

on Ruby 1.9.1

$ ruby -v
ruby 1.9.1p378 (2010-01-10 revision 26273) [i686-linux]
$ ruby t1.rb
aGot here

c
$ ruby t2.rb
$ ruby t3.rb
$

Result
t1.rb ==> fail
t2.rb ==> success
t3.rb ==> fail

This could, or should vary, actually. The example code is not thread safe.

On Ruby 1.9.3dev

duometis02@Duo02:~$ ruby -v
ruby 1.9.3dev (2010-07-22 trunk 28707) [i686-linux]
$ ruby t1.rb
Got herea

t1.rb:5:in join': deadlock detected (fatal) from t1.rb:5:in '
$ ruby t2.rb
bat2.rb:5:in join': deadlock detected (fatal) from t2.rb:5:in '
$ ruby t3.rb
$

So what's happening here is that you're getting to thread.join before the other thread has got to thread.stop. This is a race condition in your code. Threads are hard, and it's regarded that tools like stop, run, join, and pass are very very hard to use in a thread safe way. Assuming that calling Thread.pass with force at least a single expression to run in another thread is not valid.

Interesingly, the following code sometimes works and sometimes failed on Ruby 1.9.3.dev.
t4.rb

c = Thread.new { Thread.stop; puts "hey!" }
puts "Hi!"
c.wakeup
sleep 1

$ ruby -v
ruby 1.9.3dev (2010-07-22 trunk 28707) [i686-linux]
$ ruby t4.rb
Hi!
$ ruby t4.rb
Hi!
hey!
$ ruby t4.rb
Hi!
hey!
$ ruby t4.rb
Hi!

As above, this is a race condition. If c.wakeup is called before Thread.stop, then the code will never restart the thread after it stops itself. This is an error in the code, not in the interpreter.

=end

[raggi (James Tucker)]

=begin

On 23 Jul 2010, at 05:55, Heesob Park wrote:

Bug #3606: Thread.stop and puts fail to work as documented.
http://redmine.ruby-lang.org/issues/show/3606

Author: Heesob Park
Status: Open, Priority: Normal
Category: core, Target version: 1.9.x
ruby -v: ruby 1.9.3dev (2010-07-22 trunk 28707) [i686-linux]

I have tested following three codes which are the sample code in RDoc Documentation.

I see your point regarding this being code from the documentation. Examples as given in the documentation cannot work reliably with a preemptive scheduler and should be removed.

I would also say that use of Thread.stop and Thread.pass for any kind of "synchrony" should be very strongly discouraged, we have locking primitives.

=end

[phasis68 (Heesob Park)]

=begin
2010/7/23 James Tucker [e-mail]:

On 23 Jul 2010, at 05:55, Heesob Park wrote:

Bug #3606: Thread.stop and puts fail to work as documented.
http://redmine.ruby-lang.org/issues/show/3606

Author: Heesob Park
Status: Open, Priority: Normal
Category: core, Target version: 1.9.x
ruby -v: ruby 1.9.3dev (2010-07-22 trunk 28707) [i686-linux]

I have tested following three codes which are the sample code in RDoc Documentation.

I see your point regarding this being code from the documentation. Examples as given in the documentation cannot work reliably with a preemptive scheduler and should be removed.

I would also say that use of Thread.stop and Thread.pass for any kind of "synchrony" should be very strongly discouraged, we have locking primitives.

Thanks for your explanation.

I think that the sample code should be modified like this.

t1.rb

a = Thread.new { puts "a"; Thread.stop; puts "c" }
sleep 0.1 while a.status!='sleep'
Thread.pass
puts "Got here"
a.run
a.join
sleep 1

t2.rb

a = Thread.new { print "a"; Thread.stop; print "c" }
sleep 0.1 while a.status!='sleep'
Thread.pass
print "b"
a.run
a.join
sleep 1

t3.rb

c = Thread.new { Thread.stop; puts "hey!" }
sleep 0.1 while c.status!='sleep'
c.wakeup
sleep 1

Regards,
Park Heesob

=end

[raggi (James Tucker)]

=begin

On 24 Jul 2010, at 01:52, Heesob Park wrote:

I think that the sample code should be modified like this.

t1.rb

a = Thread.new { puts "a"; Thread.stop; puts "c" }
sleep 0.1 while a.status!='sleep'
Thread.pass
puts "Got here"
a.run
a.join
sleep 1

Hmm status, yeah, I can't think of a better way there (I wish there was)

Maybe a.join(0.1) instead of sleep 0.1?

Thread.pass is not required anymore.

In docs for Thread.pass, I'm not sure what a good use case is.

=end

[naruse (Yui NARUSE)]

• Status changed from Open to Assigned
• Assignee set to kosaki (Motohiro KOSAKI)

[kosaki (Motohiro KOSAKI)]

• Status changed from Assigned to Closed
• % Done changed from 0 to 100

This issue was solved with changeset r32298.
Heesob, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.

---

• thread.c (rb_thread_run): change RDoc. The old example is buggy
  and may cause deadlock. The patch is suggested by Heesob Park
  [e-mail]. Thank you! [Bug #3606][ruby-core:31454]