Boost logo

Boost :

From: William E. Kempf (williamkempf_at_[hidden])
Date: 2002-08-12 13:13:21


----- Original Message -----
From: "David Bergman" <davidb_at_[hidden]>
> Bill,
>
> Your design is good, probably the only rational one for all possible
> thread cases (worker threads, asynchronous calls etc.) I have no problem
> with that, although I would rather have a more "embedded" variant, which
> would be quite hard to define so that the behavior is consistent in all
> the above thread cases.
>
> As an example that there are alternative applications of the C++
> Standard onto MT C++: one, fundamentalist, reading of The C++ Standard
> gives that each thread should be born with the default behavior for
> uncaught exceptions (i.e., unexpected --> terminate --> ending thread).
> Thus, the handler vector would need to be reset for the thread. The code
> run by the thread should be able to count on that default behavior.
> I.e., one should not be allowed to change the unexpected--terminate
> behavior before triggering that thread...
>
> So, your design is good, but not the only logical extension of C++ to MT
> C++. Another logical extension would be to override "set_XXX_handler" so
> that each thread hosts its own handler vector, making them reside in a
> C++ compliant environment.

Again, correct me if I'm wrong, but I can't see how to implement this idea
portably.

Bill Kempf


Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk