thread-safety definition

[David Abrahams]

David Abrahams <dave at boost-consulting.com> writes:

> Ted Baker <baker at cs.fsu.edu> writes:
>
>> Yes, but it need not be a *huge* amount of cycles.  Protection
>> against cancellation can be done via the equivalent of setting one
>> bit in a thread descriptor
>
> Setting one bit per nontrivial destructor call would be a performance
> disaster for many programs.

But fortunately we all seem to be coming to the conclusion that it can
happen at virtually no cost (see Ted B's "static mapping" remark).

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com