[Boost-users] Re: tokenizer and wstring with VC7.1

From: Keith MacDonald (boost_at_[hidden])
Date: 2004-02-22 15:36:54

Hmmm.  I've been trying to use various members of the boost library as black
boxes, but this issue highlights the danger of doing so.  I suppose a
language keyword is needed to specify when a non-temporary object is
required as an actual parameter.  Given that there's no such thing, perhaps
it would be safer to eliminate such convenience constructors from the
library?

Keith MacDonald

"Bronek Kozicki" <brok_at_[hidden]> wrote in message
news:1d55128kvroww.cbh559gfhwom.dlg_at_40tude.net...
> On Sat, 21 Feb 2004 09:10:43 -0000, Keith MacDonald wrote:
> >     MyTokenizer token(string_t(_T("abacadaeafag")), sep);
>
> if you take a look into tokenizer constructor
>
>     template <typename Container>
>     tokenizer(const Container& c,const TokenizerFunc& f)
>       : first_(c.begin()), last_(c.end()), f_(f) { }
>
> you will notice that it's not storing copy of its string argument;
> instead it stores only its begin and end iterator. When string variable
> is destroyed (and in your example it's temporary variable; thus its
> destroyed at the end of expression) these iterators are no longer valid.
> Problem you are experiencing here is unusual manifestation of undefined
> behaviour - you are working in invalid interators. I think that program
> crash would be better indicatation that you have serious problem, but
> undefined behaviour may manifest in any other way - this time it's just
> as if tokenizer is empty. Of course, slight change of program or
> compilation options may result in crash (or anything else), until you
> remove undefined behaviour:
>
> >     string_t    s(_T("abacadaeafag"));
> >     MyTokenizer token(s, sep);
>
>
> B.