Re: [boost] [thread] condition suffering from lost wakup bug?

From: Roland Schwarz (roland.schwarz_at_[hidden])
Date: 2005-09-15 08:50:43

Alexander Terekhov schrieb:

>Roland Schwarz wrote:
>[...]
>  
>
>>Altough I would very much appreciate if someone could tell me why it is
>>not a
>>problem when the sequence: "wait then signal"  does occur, during the
>>gate is closed.
>>It might very well be that this indeed _is_ no problem, I just can't see
>>why.
>>    
>>
>
>Because incoming waiter holds external mutex while waiting on gate.
>  
>
This means that there can be at most only one thread beeing blocked on 
the gate, correct?

But this still doesn't me get going. I am considering the following 
sequence of events:

Given sem_wait(semBlockQueue) currently is waiting with nWaitersBlocked 
== N.
The gate is open.

1) signal all takes place, causing
    closing the gate
    nWaitersBlocked = 0
    nWaitersToUnblock = N

2) sem_wait is getting awake and starts procssing,
    needing some time to proceed while the gate is closed

3) a new waiter arrives while the gate still is closed
    he will block on the gate and not increment nWaitersBlocked yet

4) a signal arrives, while gate still is closed
    Now is 0 < nWaitersToUnblock < N, correct?
    But 0 == nWaitersBlocked
    the signal results in a NO-OP and consequently the
    waiter from 3) will never get it to see - hence it is lost ??

I am studying the pthread_cond_wait.c pseudo code taken from
SNAPSHOT 2004-05-16.

What do I miss here? I see this is not going to happen if the thread
that signals, is holding the external mutex when trying to signal. But 
this isn't
a requirement, is it?

Thank you for your kind help.
Roland