Re: [boost] [xint] Boost.XInt formal review

Subject: Re: [boost] [xint] Boost.XInt formal review
From: Ivan Le Lann (ivan.lelann_at_[hidden])
Date: 2011-03-04 02:56:23

Le vendredi 04 mars 2011 à 01:02 -0500, Chad Nelson a écrit :
> >>> integer_t<...>  &  operator+= (const integer_t<...>  b)
> >>> integer_t<...>  &  operator-= (const integer_t<...>  b)
> >>>
> >>> Why are these parameters passed by value?
> >>
> >> This was mentioned in an earlier message. I didn't recall the reason
> >> then, but I've looked into it: it's partly due to Dave Abraham's
> >> "Want Speed? Pass by Value." article (possibly by misunderstanding,
> >> I'll review that in the next few days to find out), and partly
> >> because the internal COW system makes copies very cheap -- I
> >> suspect, as cheap as pass-by-reference. Most of those pass-by-value
> >> functions should probably be changed, we'll see what I come up with.
> >
> > I'm pretty sure a COW copy is never cheaper than a reference copy,
> > and probably always less cheap (given the reference count that needs 
> > incrementing). 
> 
> Yes, I've reviewed the article, and it was a misunderstanding. I'll
> change the parameters to const references for the next release.
> 

As Mathias said, if your benchmarks used "pass by balue" everywhere,
it's not a surprise a priori that COW test case was faster.

Ivan