[c++-pthreads] Re: What are the real issues?

[Fergus Henderson]

On 06-Jan-2004, Howard Hinnant <hinnant at twcny.rr.com> wrote:
> In the scenario I've outlined, code written by different parties would 
> work together in a thread/cancel environment as long as:
> 
> 1.  The code met basic exception safety requirements.
> 2.  The code was not compute-bound (and thus ignored cancellation 
> requests).

It would also require

	3.  The code doesn't block for I/O.

wouldn't it?

> For example, I could've sprinkled my helloworld with std::lib calls, 
> and it still would've been ok because the std::lib maintains basic 
> exception safety.

Won't it fail to cancel the threads if the output blocks, e.g. because
it is redirected to Unix pipe that fills up?

> Admittedly if a std::lib call took an inordinate 
> amount of time, cancellation requests would be ignored during that 
> call.  But it seems we already have the risk of ignoring cancellation 
> requests anyway.

That risk of cancellations being ignored can be avoided by ensuring
that all the code that you call is strongly cancellation-safe.

The problem with your suggested approach is that unless the cancellation
framework is part of the standard library, in general it's not possible
for third parties to write strongly cancellation-safe code!
Writing strongly cancellation-safe code requires (a) being able to poll
for cancellation and (b) being able to do blocking I/O in a way that
will wake up when a cancellation request is received.

-- 
Fergus Henderson <fjh at cs.mu.oz.au>  |  "I have always known that the pursuit
The University of Melbourne         |  of excellence is a lethal habit"
WWW: <http://www.cs.mu.oz.au/~fjh>  |     -- the last words of T. S. Garp.