5c6
Subj : Re: Can C++ local static objects be made thread safe?
To   : comp.programming.threads
From : Alexander Terekhov
Date : Sat Jan 22 2005 01:43 pm


Gianni Mariani wrote:
[...]
> Is this below going to perform as advertised ?

Yeah. As long as it's advertised as totally busted.

> ThreadLocal, MemoryBarrierSyncReads and MemoryBarrierSyncWrites
> are mythical things at this time.

DCSI-TLS doesn't need extra barriers. Here's an illustration using 
Java's "ThreadLocal":

  class Foo {

    private static Foo theInstance;

    private static final ThreadLocal tlsInstance =
      new ThreadLocal() {
          protected synchronized Object initialValue() {
            if (theInstance == null)
              theInstance = new Foo();
            return theInstance;
          }
        };

    public static Foo getInstance() {
      return (Foo)tlsInstance.get();
    }  
  } 

In C++ (and dynamic scope),

  class stuff : private lazy_mutex { // "create/open named mutex" 
                                     // trick on windows 

    foo *                                m_ptr;
    thread_specific_ptr<foo, no_cleanup> m_tsp;

  public:

    /* ... */

    const foo & foo_instance() { 
      foo * ptr;
      if (!(ptr = m_tsp.get())) { 
        lazy_mutex::guard guard(this);
        if (!m_ptr) m_ptr = new foo();
        m_tsp.set(ptr = m_ptr);
      }
      return *ptr;
    }

regards,
alexander.

.

0