Re: __gnu_cxx::hash_set efficient union

From:

tragomaskhalos <dave.du.vergier@logicacmg.com>

Newsgroups:

comp.lang.c++

Date:

Fri, 17 Aug 2007 02:01:51 -0700

Message-ID:

<1187341311.587776.9480@k79g2000hse.googlegroups.com>

On Aug 16, 4:58 pm, mark.duf...@gmail.com wrote:

hi all,

as part of an 'implicitly statically typed Python' to C++ compiler
called Shedskin I'm working on (http://mark.dufour.googlepages.com),
I'm using __gnu_cxx::hash_set to implement Python sets. unfortunately
I can't get the performance to quite match that of set. for example,
this is how I thought set union would be reasonably fast:

template<class T> set<T> *set<T>::_union(set<T> *s) {
    set<T> *c = new set<T>();
    c->units = units;

    it2 = s->units.end();
    for(it1 = s->units.begin(); it1 != it2; it1++)
        c->units.insert(*it1);

    return c;

}

I'm sure the effect is negligible, but is there any reason
you're dynamically allocating the set here? I'd have though
set itself is a fairly small object so creating it on the
stack and returning that (and relying on RVO) might be better.

Also, I have read that support for multithreading, injudiciously
applied by library providers, can have a ruinous effect upon
performance of standard containers. So it may be worth having a
look at the documentation (or on google) to see if there are
any such issues surrounding __gnu_cxx::hash_set.