From: Steven Watanabe (steven_at_[hidden])
Date: 2007-10-21 21:22:04
AMDG
Mathias Gaunard <mathias.gaunard <at> etu.u-bordeaux1.fr> writes:
> The problem with small object optimization is that you get the same
> issue that boost.variant has: it's hard to maintain the never-empty
> guarantee in operator=, because the copy constructor of a new value can
> throw and thus you lose the old value.
When I was trying to do this I only used the small
object optimization for types that I knew to have a no-throw
copy constructor (i.e. builtin types at the moment). In C++Ox
this can be loosened to having a no-throw move constructor.
> I wonder, by the way, how boost.function solves that issue.
It doesn't. Boost.Function can be empty and assignment for
Boost.Function does not provide the strong guarantee of
exception safety.
In Christ,
Steven Watanabe