[ruby-talk:5914] Re: [RFC] Towards a new synchronisation primitive

In message <[e-mail]>
[e-mail] writes:

> > deadlock could be avoided.  The only time you cannot get into your house
> > is if somebody ELSE has the only key; if you have it that is OK.  Or is
> > that too simplistic?  This area is full of subtleties....
> 
> I like this house+key metaphor, let's use it. The Java lang-spec 2.0
> uses the same semantics IIRC. And simplicity is good: "Things should
> be as simple as possible, but not simpler." -- AE.

Uh, hmm.  Threads cannot split.  But such a program is rare.... at
least I can't get any meaningful situation.

An artificial example I thoght is concurrent traversal of graph.

    class synchronized Node
      attr :neighbors, true
    
      # details are snipped.
    
      def traverse
        queries = @neighbors.collect {|n| Thread.new {n.traverse}}
        ThreadsWait.wait_all(queries)
        some_work_with_query_results
      end
    end

    # If graph has cycles, an originator can be traversed by successor
    # threads, that would be blocked since they are not the originator
    # thread itself.

This type of deadlock can be avoided by threads remembering its parent
though.  The interpreter can trace all ancestors from a known
immediate parent.

# E. A. Brewer and C. A. Waldspurger, ``Preventing Recursion Deadlock
# In Concurrent Object-Oriented Systems,'' Massachusetts Institute of
# Technology, Laboratory for Computer Science,
# MIT-LCS//MIT/LCS/TR-526, February 1992.



-- 
[e-mail]                               October 27, 2000
So many men, so many minds.