From: Robert Feldt <feldt@...>
Date: 2000-10-27T23:40:50+09:00
Subject: [ruby-talk:5916] Re: [RFC] Towards a new synchronisation primitive


Continuing my previous message that was sent before complete...

On Fri, 27 Oct 2000, hipster wrote:

> > 4. Implement message queues and the rest of the common multi-threading
> > helper classes in extension 'multithreading'.
> > 
> > My knowledge of the current implementation of Ruby threads is somewhat
> > limited though, so I'm not sure how this would affect the deadlock
> > detection currently available?
> 
> Same here. Got some catching up to do  :\
> 
> IMO a good proposal, thanks.
> 
Thanks for your feedback.

Below is example code for Mutex and module Synchronized (using
test_and_set), using test_and_set, so that you can do:
    
class MySynchedClass
  include Synchronized
  def my_synched_method
    ...
  end
  synchronized :my_synched_method
end
    
It's only on the object level though. Finer granularity might be needed...

I think it correctly handles Dave's RingBuffer example but I'm not
sure. Can someone verify?

class Mutex
  def initialize
    @waiting_threads = []
    @locked = false             # Not locked yet...
    @waiting_threads.taint	# enable tainted comunication
    self.taint
  end

  # Since there is not yet a test_and_set in Ruby we use existing
  # Thread.critical= syntax.
  def test_and_set
    Thread.critical = true
    previous_value = @locked
    @locked = true
    Thread.critical = false
    previous_value
  end
  private :test_and_set

  def non_blocking_lock
    !test_and_set
  end
  alias try_lock non_blocking_lock

  def lock
    while test_and_set
      @waiting_threads.push Thread.current
      Thread.stop
    end
    self
  end

  def unlock
    return unless @locked
    # It's ok to release lock before releasing waiting thread
    # since Mutex do not garantuee an ordering of the threads.
    # Maybe we should supply different mutexes so that you can choose
    # the behavior you want (say BlockedQueueMutex that unlocks
    # threads FIFO-style, and BlockedSetMutex for the current impl?)
    @locked = false
    begin
      t = @waiting_threads.shift
      t.wakeup if t
    rescue ThreadError
      retry
    end
    self
  end

  # I can't see the use for exclusive_unlock but please enlighten me ;-)
end

# In the spirit of Masatoshi SEKI's MutexM:
module Synchronized
  def Synchronized.append_features(klass)
    super

    # Add synchronized (for synchronizing on the object level)
    # Maybe have way to synch on method level?
    def klass.synchronized(*methods)
      for method in methods
	begin
	  method = method.id2name
	rescue NameError
	end
	old = "_old_#{method}" # We need something smarter here to handle
methods including any of the characters [,] or =. Counter in class var?
	module_eval("alias #{old} #{method}")
	module_eval("def #{method}(*a,&b); @m.lock; #{old}(*a,&b); @m.unlock; end")
	private(old)
      end
    end

    return self
  end

  private

  def initialize(*args)
    ret = super
    # We should probably extend this to have locks/mutexes on finer
granularity
    # so that you can sync on the instance mutex or on individual methods
    # mutexes? Is it needed? If so we can generate the mutex
initialization 
    # code below dynamically?
    @m = Mutex.new
    return ret
  end

end  

Regards,

Robert