Boost logo

Boost :

From: Rob Stewart (stewart_at_[hidden])
Date: 2004-07-08 14:13:05

From: brangdon_at_[hidden] (Dave Harris)
> In-Reply-To: <200407072103.i67L3Jk28273_at_[hidden]>
> stewart_at_[hidden] (Rob Stewart) wrote (abridged):
> > > v << foo() << bar();
> >
> > I don't like that. It suggests formatted insertion rather than
> > initialization.
> It's not intended to be initialisation. It does a push_back(), so you can
> do it at any time and not just on empty vectors. I guess I'd pick a
> different name for the library, but as written that includes += so it too
> is not just for initialisation.

The name misleads one into thinking myopically about its use.

> As I said, I don't think formatting is an essential characteristic of
> operator<<(). It happens to be necessary for streams because they are text
> oriented. The serialisation library uses operator<<() without formatting.

A reasonable point, though not having used the serialization
library, I still think of operator<< and operator>> as doing
formatted I/O.

> > As you point out, conversion is possible, but one can't overload
> > operator<< to customize the initialization.
> Why not?
> vector<Foo> &operator<<( vector<Foo> &v, const Bar &bar ) {
> return v << bar.toFoo(); // Or whatever conversion you want.
> }
> Or whatever the right types are.

I overstated the case. How many types will one have to overload
operator<< (and operator>>) for? The combinatorial explosion
(containers X types to use with the initialization library X 2)
will be overwhelming.

If there is a standard representation one wants in those
contexts, one might get away with using a function template to
genericize the container type, but then you'd need a way for that
template to be used only in specific contexts.

Also, what if you want different formatting in different
containers, even with the same type?

(No doubt I'm trying to make difficult what's supposed to be a
simplistic task.)

> > Thus, it doesn't fit neatly with IOStreams.
> It's not polymorphic with them, but I think it fits about as neatly as the
> serialisation library.

I can't agree, at least not yet.

Rob Stewart                           stewart_at_[hidden]
Software Engineer           
Susquehanna International Group, LLP  using std::disclaimer;

Boost list run by bdawes at, gregod at, cpdaniel at, john at