Re: [boost] Re: Optimising assignment using std::swap

From: Paul Moore (gustav_at_[hidden])
Date: 2001-01-23 17:29:42

On 23 Jan 2001, at 14:07, Daryle Walker wrote:

> on 1/21/01 10:33 PM, Paul Moore at gustav_at_[hidden] wrote:
> 
> [SNIP]
> > This could be replaced with
> > 
> > IntType r(n%m);
> > std::swap(n,m)
> > std::swap(m,r);
> 
> Does it have to be the "swap" from the "std" namespace?  Could it be
> generalized to just "swap" like
> 
> //=============================
> using std::swap;
> IntType  r( n % m );
> swap( n, m );
> swap( m, r );
> //=============================

Blast, damn, @*&%$! and other swear words. You're probably 
right... I thought for some reason that swap was one of the ones 
you could insert into std::. I'm not sure why I thought this.

This brings up all the nasty issues I had with abs(), on which 
subject I just posted a message explaining why I have decided to 
abandon my attempts to get rational's abs() to use a user-supplied 
abs().

At least with swap() there is a fallback which applies to all types in 
std::. So we don't have to worry about a built-in type without a 
swap() implementation. But we still have the worry of compilers 
which don't do Koenig lookup, the gcc bug where using statements 
at function scope are ignored, and so on.

I'll probably not bother with the swap() optimisation then. Maybe I'll 
put a comment in the code explaining the optimisation, plus a 
reminder to re-enable it once gcc and MSVC support Koenig 
lookup.

Paul.

PS In case people think I'm just being petluant, I should point out 
that without Koenig lookup, the swap() and abs() stuff simply 
doesn't work. With abs(), you get compile errors, and with swap() 
you get the wrong version being used (resulting in at best no 
improvement, and at worst a severe pessimization of the code). I'd 
rather leave it out than have broken code. And as I've said before, I 
have to use MSVC, so I won't implement something which doesn't 
work on that compiler.

PPS Is it only me who feels that the "using std::swap" incantation 
required to be able to use swap() unqualified and hence to get the 
benefit of Koenig lookup is clumsy, unintuitive and inelegant?