Here's how to provide platform neutrality while keeping typesafety and avoiding #ifdef:
 
I will fabricate an example loosely based on  boost::thread.
 
 
// thread.hpp
 
namespace boost {
 
    class thread {
        public:
            //...
        private:
            void runAsynchronousMethod () {
                // run our happy boost::function
                // deal with exceptions
            }
 
            boost::shared_ptr<class PlatformSpecificThread> psThread; // must be last attribute to make sure "this" is fully constructed before thread starts.
            friend class PlatformSpecificThread;
    };
}
 
// thread.cpp - win32 version in an isolated subdirectory
 
#include <boost/thread.hpp>
#include <windows.h>
 
using boost::thread;
 
class PlatformSpecificThread { // everything private!
    PlatformSpecificThread (boost::thread* const thread)
        : handle (CreateThread (0, 0, &run), reinterpret_cast<void*> (thread), 0, 0)) {
    }
 
    ~PlatformSpecificThread () {
        CloseHandle (handle);
    }
 
    static DWORD WINAPI run (void* typelessPointer) { //nothrow
        if (0 == typelessPointer) return FALSE;
 
        reinterpret_cast<boost::thread*> (typelessPointer)->runAsynchronousMethod (); // can be implemented in the header because it's platform neutral :)
        return TRUE;
    }
       
    HANDLE handle;
    friend class boost::thread;
    // needs to be a friend of checked_delete as well for use with shared_ptr
}
 
thread::thread (/*boost::function*/)
    : psThread (new PlatformSpecificThread (this)) {
}
 
// thread specific methods like thread::join, etc, only the few platform specific methods are necessary here
 
----- Original Message -----
From: Moore, Dave
Newsgroups: gmane.comp.lib.boost.devel
Sent: Monday, 2002:August:05 12:59 PM
Subject: RE: Re: Threads & Exceptions


> My complaints on the ifdefs were on the underside of the thread. C++ alone
provides plenty of
> mechanisms for mapping the common interface to different platforms, with
typesafety (instead of
> the reinterpret_casts<>!)--again, the topic of a different discussion, at
which point it would be
> easier just to submit a replacement as example.

Fair enough, but it's a hard problem to solve w/o resorting to (1)
#including os-specific files like <windows.h> in the interface which wreak
havoc on namespaces, macros, etc., or (2) having a complete shadow set of
dynamically allocated Pimpl classes.  It's always interesting to see
alternative approaches, though.


> There is no reason to make users of thread handle their own exceptions
because there is only one
> mechanism to do it, and every user will have to repeat this code in their
implementation. 
> in thread processes have always been ambiguous to me, it seems that no
framework out there
> supportws them properly and just expects them to be caught and sometimes
ignored.

> If you'll take note of the advanced_thread usage, it not provides a
generic way to run any  
> boost::function asynchronously, get its return value and deal with
exceptions it may throw--a
>complete solution and something currently not yet available. Of course one
could take the
>perspective that advanced_thread is really an asynchronous function
adapter, but that deprecates the >need to support what is currently offered
by boost::thread--which is why I suggested a possible
>replacement.

I agree with your statements about reuse and flexibitly for asynchronous
function calls and flexibility in handling exceptions.  I'm just not sure
that an inheritance from (or replacement of) boost:thread is the best
solution.

Consider an alternative: a thread_pool object which can enqueue
boost::function calls and distribute them to a managed pool of threads ready
to execute them.  boost::function gives you the hook to implement -any-
scenario of argument passing, return value capture, and exception handling
you wish.  This is a problem (IMHO) that suggests a solution via composition
of two existing classes, not inheritance.

See:
http://groups.yahoo.com/group/boost/files/thread_pool.zip

For a draft of a solution which may (hopefully) find its way into
Boost.Threads.

Regards,
Dave

_______________________________________________
Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost