Re: [boost] [xint] Performance of fixed-size large integers

Subject: Re: [boost] [xint] Performance of fixed-size large integers
From: Artyom (artyomtnk_at_[hidden])
Date: 2011-03-05 04:43:22

----- Original Message ----
> From: Dave Abrahams <dave_at_[hidden]>
> To: boost_at_[hidden]
> Sent: Sat, March 5, 2011 12:21:50 AM
> Subject: Re: [boost] [xint] Performance of fixed-size large integers
> 
> At Fri, 4 Mar 2011 13:58:36 -0800 (PST),
> Artyom wrote:
> > 
> > I  personally think that COW approach would be very useful for most
> > of  users while move is "nice-to-have".
> 
> That's interesting, but it doesn't  help me form my own opinion.  Maybe
> if you told us on what basis you  came to that conclusion, it would be
> more helpful.  Have you done any  tests to see how easily/quickly a
> type implemented with move emulation works  as compared to one
> implemented with COW?  Have you got concrete  examples?
> 

I'm not talking about performance I'm talking about what users would do.

How many developers would write:

   {
     ...
     foo.set_value(boost::move(some_big_int));
     // some big int goes out of scope
   }

Over
  
   {
     ...
     foo.set_value(some_big_int);
     // some big int goes out of scope
   }

In case you really need a copy of the value and not reference/const reference.

We should remember that not all Boost audience is familiar
with "movable" concept. Even good C++ programmers
do not really live on the edge and use the-state-of-the-art tools.

And this is the vast majority of good C++ programmers.

Now if we take a look on thous who are actually going to use
Xint - algorithms developers, mathematicians and cryptographers 
who actually use and know C++ but do not live at the edge.

So I can say that COW would give an immediate benefit while
MOVE would give some benefit to some users.

I'm telling let's look on the wider picture and not the narrow
one.

Finally we write our libraries to be useful for much less
competent programmers. And unfortunately this is the reality.

I think it is likely smart users would rather use the old school move

  set_value(int_type &integer)
  {
      integer_.swap(integer);
  }

Then boost::move.

And BTW it would be more clear for vast majority of code
maintainers that are even less competent then the
mathematicians who had written the code at the begging.

It is sad but this is the reality, because I see at daily
basis what kind of code the real programmers write, and
Boost.Move is far beyond their capabilities.

That is why COW would be much better for real programmers
at least in near 5 years till boost::move or std::move would
become mainstream.

Artyom