Re: Draft of generation 2 thread class

From: bill_kempf (williamkempf_at_[hidden])
Date: 2002-01-21 08:32:27

--- In boost_at_y..., "Peter Dimov" <pdimov_at_m...> wrote:
> From: "bill_kempf" <williamkempf_at_h...>
> > --- In boost_at_y..., "Peter Dimov" <pdimov_at_m...> wrote:
> > > From: "bill_kempf" <williamkempf_at_h...>
> > > > I've uploaded the draft document for the next generation of
> > > > boost::thread for discussion.
> > >
> > > One thing that immediately jumps at me is:
> > >
> > > Requires: category() != foreign and there has been no previous
> > calls to
> > > either join() or detach() on any thread objects representing the
> > same thread
> > > of execution as *this.
> > >
> > > but I (the user) have no means to detect whether the 
requirements
> > are met.
> >
> > To detect, no, but know, possibly.  It depends heavily on the 
design
> > of the application.
> 
> Consider this example:
> 
> void f(boost::thread & t)
> {
> // ...
>     t.join();
> }
> 
> I'd like to assert() on the precondition before calling t.join().
> 
> Isn't it possible to make the thread object remember whether join() 
or
> detach() have been called?

For a single instance of the thread object, yes.  To do so for all 
thread objects, not with out going to a full blown thread_ref design.

I don't consider this much of a problem, however.  This is really no 
different then not being able to call delete twice for an object 
(whether from the same pointer or from different pointers to the same 
object).  A well designed application will ensure this never occurs 
simply through the normal tracking of ownership.

Bill Kempf