Re: [boost] Re: Boost.Threads - thread_desc concept

From: Greg Colvin (gcolvin_at_[hidden])
Date: 2001-06-28 16:48:42

From: John Max Skaller <skaller_at_[hidden]>
> Greg Colvin wrote:
> 
> > To be clear, just stopping a thread "out of nowhere" is
> > a bad idea, as there is then no way to get out of any
> > monitors it has entered.  Better is a way to throw an
> > exception on the thread, so as to let destructors run.
> 
> You did not meet the requirements.
> The thread is blocked. It cannot throw anything.

Right.  You would have to unblock it first.

> It is a core language issue whether destructors
> are run when a thread is cancelled.

Yes, which is one reason I think we want to punt.

> > Sometimes it is even better to spawn processes instead
> > of threads, so the OS can take care of cleaning up a
> > killed process.
> 
> Sometimes it is. And sometimes not.
> I have seen the application in question, a very sophisticated
> database. The designer is an expert. He swears by cancellation.

Yes, it can be effective.

> FYI the design supports distribution, replication,
> and recovery, all while the DB remains operational.

A familiar set of requirements.  Oracle uses process on
some platforms, threads on others, depending on what is
available and how well it performs.