Re: Boost.Threads - thread_desc concept

From: williamkempf_at_[hidden]
Date: 2001-06-27 08:44:39

--- In boost_at_y..., <llee1_at_l...> wrote:
> On Tue, 26 Jun 2001, Jeremy Siek wrote:
> 
> >
> > Here's an alternative interface, mostly due to Rich Lee.
> 
> In fact, I have changed the interface from free functions to class 
as
> follows:
> 
>   template <class Obj,
> 	    class Feature1=default_template_arg,
> 	    class Feature2=default_template_arg,
> 	    class Feature3=default_template_arg >
>   class thread {
>    public:
>     thread(Obj* o_);
>     void run();
> 
>     void join();
>     void detach();
> 
>     void yield();
> 
>     ~thread();
>   };
> 
> Where Feature1, Feature2, and Feature3 are for ContentionScope,
> DetachState, and SchedulePolicy.

These "features" are currently left out on purpose.  They are not 
necessary for most purposes, and they don't exist on all platforms.  
I'm shooting for a minimal yet complete interface to begin with.  
Advanced concepts can be added as demand and analysis on portability 
can be done properly.

You also opted for the two phased construction that I don't think is 
a good idea.  It doesn't add any benefit (at least that's been 
pointed out to me yet), while it complicates usage.

This design also doesn't provide a self() method, which I think may 
be necessary.

Also, yield should be a static method, especially if self() does not 
exist.

There's also a few bugs in the implementation.  The detached state is 
not synchronized, making this class not thread safe.  Join sets the 
state to not-detached, when in fact there should be a third state, 
lest some operations result in errors.

After all of this, this is in essence the third design I suggested.

Bill Kempf