From: Thomas Matelich (matelich_at_[hidden])
Date: 2005-08-09 08:00:27
On 8/9/05, Yuval Ronen <ronen_yuval_at_[hidden]> wrote:
> Thomas Matelich wrote:
> >
> > You'll note that stream >> std::ws is no longer called.
> >
> > One note that I must make is that I do not have the vanilla source for
> > lexical_cast. From the last time lexical_cast broke my source, I've
> > commented out stream.unsetf(std::ios::skipws); in the lexical_stream
> > constructor (see
> > http://listarchives.boost.org/MailArchives/boost-users/msg03656.php for the
> > reason).
> >
> > I just recalled a unit test we have that would expose the problem (and
> > it did). By adding a [stream >> std::ws &&] before the check for eof,
> > my problem was resolved. Is there a reason we want to fail with
> > trailing whitespace?
>
> I'm not the lexical_cast author, but I do know there were discussions
> about this in the Boost mailing lists (search the archives, one of these
> threads was started by me). The reason for changing it was symmetry. In
> Boost 1.32, trailing whitespaces were ignored, while leading whitespaces
> were not. This is clearly wrong because there's no reason for asymmetry
> here. So the decision (which I'm fully agree with, BTW) is to *not*
> ignore whitespaces at all.
>
> If you want to ignore whitespaces, then the (simple) suggestion I can
> give you is to trim (the string_algo Boost library) the string before
> sending it to lexical_cast.
Why do these changes always get made when I'm taking a break from the
list? Oh well, I didn't see the value in the first change, and I
don't see the value in this change. I'll add another item to my
"Things to do to new Boost versions before committing" list.
Thanks for the answer.
Tom