Re: [boost] Review request: Boost.FSM

From: Andrey Semashev (andysem_at_[hidden])
Date: 2007-01-06 18:29:11

Hello Steven,

Sunday, January 7, 2007, 2:02:10 AM, you wrote:

> AMDG

> Andrey Semashev wrote:
>> This is just wrong. It is perfectly legal
>> > for the compiler to reorder it to be
>> >         if (!m_fInitialized)
>> >         {
>> >           m_fInitialized = true;
>> >           // Fill m_StatesInfo array for all states
>> >           const root_type* pRoot = pStates;
>> >           pStates->__init_states_info(reinterpret_cast< const char* 
>>  >>(pRoot), m_StatesInfo);
>>   
>>> >         }
>>> > then
>>> > Thread A
>>> >    if (!m_fInitialized)
>>> >    {
>>> >      m_fInitialized = true;
>>> > Thread B
>>> >    if (!m_fInitialized)
>>> >    {
>>> >    }
>>> >    //use m_StatesInfo.
>>> >    //die.
>>>     
>> I'd rather say, your scenario is possible here but not necessarily it 
>> will happen. A compiler _may_ be smart enough to reorder expressions 
>> if it may prove that it won't break the code behavior (which is true 
>> here) and it will gain something (performance, for example) from it. 
>> And even if it does so consider that Thread A in your example may be 
>> interleaved by Thread B right between it checks "m_fInitialized" and 
>> stores "true" in it. So it is actually possible that more than one 
>> thread will execute the "true" branch of the "if" statement 
>> concurrently. But we're on the safe side here because only simple 
>> operations on POD types are involved in it.
> I see that it is ok for  m_StatesInfo to be initialized more than once.
> My point was
> that it can be used before it is initialized unless you make everything
> involved volatile.

Oh, you're right. But isn't it enough just to mark "m_fInitialized" as
"volatile"? According to 1.9/7 it would enforce the compiler to
initialize "m_StatesInfo" prior to setting "m_fInitialized".

-- 
Best regards,
 Andrey                            mailto:andysem_at_[hidden]