the change at r12469 broke thread support (reverting that change fixes my problem). I get spurious "not owner"
errors during ConditionVariable#wait in the following context:

 mutex.synchronize do
   condition_variable.wait(mutex)
 end

I fail to see how the current thread can *not* be the owner

Date: 2007-06-20 05:39
Sender: Sylvain Joyeux

The patch works. Thanks.

Date: 2007-06-19 01:38
Sender: Nobuyoshi Nakada 

Oops, I missed to "Check to Upload & Attach File."

Date: 2007-06-18 08:38
Sender: Sylvain Joyeux

I think you forgot to attach the patch ;-)

Date: 2007-06-18 06:34
Sender: Nobuyoshi Nakada 

Is this patch OK?

Date: 2007-06-13 11:46
Sender: Sylvain Joyeux

OK ... Found something

I have a question about mutex_relock. How it is possible 
that the thread status in mutex_relock (which is called in 
an ensure block) is something different than RUNNABLE. I 
found that my problem was that the thread status is 
TO_KILL.

The thread *is* nonetheless running, and is processing 
ensure blocks. In particular, it processes mutex_relock in 
the ensure of wait_condvar and *then* processes 
mutex_unlock in the ensure block of rb_mutex_synchronize. 

Since the status was TO_KILL, mutex_relock did *not* 
relock and mutex_unlock complains. I personally don't see 
the point of mutex_relock. Could you explain ?

Date: 2007-06-13 11:16
Sender: Sylvain Joyeux

It seems that the problem is related to exception 
management. The error occurs when Thread#raise is used. 
Maybe it interrupts an ensure function.

The problem with multi-threading is that reproducing 
errors is very often NOT unit-testable as it is very much 
context-dependent. Your test, for instance, does not take 
into account the various context combinations the code can 
encounter

The only way to find this kind of bug is to run real MT 
apps (what I did), and then to review the code (what I 
also did for 1.8.6, I have unfortunately no time for this 
right now)

Since it is the second consecutive Ruby release that 
thread support is broken, and since it is not properly 
unit-testable, I'll advise that *any* change to thread.c 
is reviewed by *at least* two persons before commit. What 
do you think ?

Date: 2007-06-13 06:47
Sender: Nobuyoshi Nakada 

Can't you show complete small code that can reproduce the
problem?
This seems to work.

./ruby -e '
require "thread"
mutex = Mutex.new
cv = ConditionVariable.new
Thread.new{sleep 1; cv.signal}
mutex.synchronize do
  cv.wait(mutex)
end'