Re: [Boost-users] [ASIO] Problem with reading of serial ports using ASIO

Subject: Re: [Boost-users] [ASIO] Problem with reading of serial ports using ASIO
From: Jim Donaldson (jim.donaldson_at_[hidden])
Date: 2010-10-20 13:30:19

Igor R <boost.lists <at> gmail.com> writes:

> 
> > http://gist.github.com/634618
> >
> > We regularly call the function
> > PCTimeoutUart::wait( iseSize_t len, uint32 timeoutInMS ) which wraps the
> > calls to the boost asio libraries.
> >
> > This creates a deadline_timer, and binds it to our timeout callback
> > (PCTimeoutUart::finishedTimer)
> > It also calls async_read_some on the boost::asio::serial_port object  m_port
> > to read the required length, and binds that to the
> > PCTimeoutUart::finishedRead callback.
> >
> > We call the runone() function on m_io, which is declared in the header as a
> > boost::asio::io_service and has been passed to both the timer and serial
> > port's constructors.
> >
> > We continually call this in a loop until either m_timedOut has been set by
> > the asynchronous call back to PCTimeoutUart::finishedTimer
> > or m_finishedRead has been set via the call back to
> > PCTimeoutUart::finishedRead
> >
> > Normally, this works well. However, there are certain scenarios when it
> > fails to write anything to the buffer, despite the fact that finishedRead()
> > is called with error = 0 and the correct length .
> 
> I don't know whether it's related or not (most likely not), but it
> seems that the following code is dangerous, as you delete the buffer
> without waiting for the async read completion. You assume that after
> you call cancel(), the buffer won't be used anymore by the async
> operation -- which is not obvious, imho.
> 
> if( m_timedOut )		
> {			
>   m_port.cancel();			
>   delete[] buffer;			
>   return false;		
> }
> _______________________________________________
> Boost-users mailing list
> Boost-users <at> lists.boost.org
> http://lists.boost.org/mailman/listinfo.cgi/boost-users

Thanks Igor and you were virtually bang on. 

For those who are interested, the problems are at line 52 and line 72 - i.e.
once we've had a callback from either the timer or the serial port, we
immediately return from the function without waiting for the other async process
to complete. 

I believe that what was happening was that we were receiving data at about the
same time as the timeout fired and thus it was interrupting the reception of
data. Because we immediately returned, we didn't give the in-process read
request chance to finish. 

I've changed this now to only exit the 
while(m_io.run_one()) loop (line 46) 

once we've received confirmation that both the timer and port callbacks have
fired. (One will have fired normally, and the second will fire as a result of
calling cancel() in the relevant callback.) 

Only at that point do we return with the received data. 

Thanks again for your help!

Jim