From: Howard Hinnant (hinnant_at_[hidden])
Date: 1999-11-29 11:47:13
Beman Dawes wrote on 11/29/99 10:23 AM
>Unless I am misunderstanding you, that would cause a reference to a
>reference, which is not allowed by the language rules.
I was just talking about changing the value of by_value, nothing more.
As good example uses have yet to be proposed for by_value, I was probably
premature to suggest the change.
>That was the
>case that caused Steve Cleary to develop the empty_member constructor
>argument workaround that eventually morphed into call_traits.
I think I got excited about call_traits for the wrong reason
(optimization). I think the true value is exactly what you're stating
here (avoidance of reference to a reference). And also if it can help
with the Dietmar/make_pair dance.
If I'm right about the optimization part, then that means that the list
of bool, char, short... is superfluous. It doesn't hurt, but call_traits
is just as good without the "by_value" specializations for the built-in
types.
Then again, the answer to the optimization part my vary from platform to
platform. I'm not sure. (so let's defintely leave it in so we can
explore that!)
-Howard