Re: [Boost-users] [asio] asynchronous reading/writing to the tcp stream

Subject: Re: [Boost-users] [asio] asynchronous reading/writing to the tcp stream
From: Michael Chisholm (chisholm_at_[hidden])
Date: 2012-07-11 13:58:13

On 7/11/2012 12:42 PM, Ovanes Markarian wrote:
> On Wed, Jul 11, 2012 at 6:27 PM, Nat Linden <nat_at_[hidden]> wrote:
>> On Wed, Jul 11, 2012 at 12:05 PM, Ovanes Markarian
>> <om_boost_at_[hidden]> wrote:
>>
>>> On Wed, Jul 11, 2012 at 5:10 PM, Nat Linden <nat_at_[hidden]> wrote:
>>
>>>> When you want code that appears to block, but is actually driven by
>>>> asynchronous I/O, that's a use case for Boost.Coroutine.
>>>> Interestingly, the new Coroutine is supposed to be reviewed very soon:
>>>> http://olk.bplaced.net/boost-coroutine.zip
>>
>>> Nat thanks for the quick reply... But what happens with the
>>> tcp::iostream if another thread starts to read from it after a
>>> connection was open, but no response is there? Would not it block? Is
>>> it even safe to write requests to the tcp::iostream from one thread
>>> and read it from the other? As I stated before, I would like to solve
>>> the issue with ASIO only. I might be able to review coroutine lib, but
>>> I don't think this an option for us in this project.
>>
>> coroutine != thread.  :-)
>>
>> A coroutine is user-space context switching, and it happens at
>> well-defined times. It's just a way of organizing code that must
>> otherwise be structured to receive intermittent calls -- as with an
>> ASIO completion handler.
>
> Nat,
>
> I know what coroutines, fibers or user space threads are... But using
> them I still need to do some additional impl, which I don't want. I
> first want to know if I can read/write to tcp::iostream from two
> different threads and if yes, than how should I do it best.
>
>
> Thanks,
> Ovanes
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>

In my experience, no.  tcp::iostream is not thread-safe.  What I did was 
use boost::iostreams to create a separate istream and ostream which 
wrapped the same raw socket descriptor (since boost::iostreams is not 
thread-safe either).

Andy