Re: Problem with using char* to return string by reference

From:

"Ben Voigt [C++ MVP]" <rbv@nospam.nospam>

Newsgroups:

microsoft.public.vc.language

Date:

Wed, 11 Jun 2008 13:06:19 -0500

Message-ID:

<uxgeA1#yIHA.4808@TK2MSFTNGP02.phx.gbl>

Alf P. Steinbach wrote:

* Doug Harrison [MVP]:

On Wed, 11 Jun 2008 00:09:34 +0200, Norbert Unterberg
<nunterberg@newsgroups.nospam> wrote:

While we are at it, what is the best method to fill a std::string
with what functiosn like RegQueryValueEx()? MFC's CString class has
GetBuffer/ReleaseBuffer to fill the string buffer without doing an
extra copy or allocating extra memory, but what do you experts do
with a std::string?

The "official" answer is to use a std::vector and copy it to a
std::string. In practice, you can probably get away with something
like: std::string s;
// n and x conform to usual Windows API definition.
int n = maximum length including nul terminator;
s.resize(n);
int x = SomeAPI(&s[0], n);
// Assume x < n and no error...
s.resize(x);

At least I don't know of any implementation for which this would
fail, but people will object that std::string may not use contiguous
storage.

It's safe, and contigous storage for string will be guaranteed in
C++0x (voted into WP a few years ago, at Lillehammer -- I recall
the location because I'm Norwegian!, and forget the exact year
because time just slips by... :-) ).

Not entirely sure that is sufficient to make it safe. Does the new wording
also prohibit string pooling/copy-on-write?

Oh, I'm being paranoid. Using the non-const operator[] would force the copy
to be made. Never mind.

As of late last year (draft N2461) the wording was, in ?21.3.1/3,

"The char-like objects in a basic_string object shall be stored
contiguously. That is, for any basic_string object s, the identity
&*(s.begin() + n) == &*s.begin() + n shall hold for all values of n
such that 0 <= n < s.size()."

Cheers, & hth.,

- Alf