Re: [boost] Re: future of boost.threads

From: Peter Dimov (pdimov_at_[hidden])
Date: 2004-02-10 08:44:48

Darryl Green wrote:
>> -----Original Message-----
>> From: Peter Dimov [mailto:pdimov_at_[hidden]]
> [snip]
>> Now, in C++ functions actually can return either a value or
>> an exception.
>> That is,
>>     int f();
>> can return an int or throw an exception. Hence, when you
>> execute that f in a
>> thread, you should be able later to call
>>
>>     int thread::join();
>>
>> and it should return whatever f returned.
>
> Yes. I think you mentioned this last time exception
> propogation across threads was discussed,
> and I agree.

Indeed I did. :-)

> I seem to recall that William Kempf was keen
> to shift the burden of these sorts of things to some form
> of wrapper/function object
> (see http://article.gmane.org/gmane.comp.lib.boost.devel/14237).

The problem is that boost::threads does not give you access to the function
object. If it did you could've emulated the functionality from user space
with something like

    boost::thread th( wrap_function(f) );

and later

    wrapped_function<int> * pw = th.join();
    int r = pw->return_value();

>> Note that you can't add exception marshalling on a higher
>> level (with the
>> current interface) because you can't access a thread's state
>> (to retrieve
>> its return value or exception) from another thread.
>
> Can't some function object wrapper catch and store the exception
> somewhere other than in the "thread state" for later access
> from some sort of async call object regardless of the underlying
> thread implementation?

The function object is part of the thread state. In theory, as
boost::threads is about equivalent in expressive power to pthreads, you can
implement boost::threads2 on top of the low level boost::threads. But that's
just doing the work twice for no apparent benefit.