NetNews Usenet Archive 1992 #30/NN_1992_30.iso/spool/comp/std/cplus/1757

Text File | 1992-12-12 | 1.8 KB | 56 lines

Path: sparky!uunet!newsserver.pixel.kodak.com!laidbak!tellab5!mcdchg!att!att!allegra!alice!ark
From: [e-mail] (Andrew Koenig)
Newsgroups: comp.std.c++
Subject: Re: char * to const char*&
Message-ID: <[e-mail]>
Date: 10 Dec 92 13:26:41 GMT
Article-I.D.: alice.24377
References: <[e-mail]>
Reply-To: [e-mail] ()
Distribution: comp
Organization: AT&T Bell Laboratories, Murray Hill NJ
Lines: 42

In article <[e-mail]> [e-mail] (Jonathan Polito) writes:

> I am curious to what the current wisdom is on accepting char* as a
> valid argument to a function expecting const char*&

It's invalid, though the reason may not be obvious at first glance.

The main thing to realize is that const char*& means "reference to
pointer to const char" and not "reference to constant pointer to
char" as one might think at first.

That means that in order to accept a char* as an argument, it would
be necessary to convert the char* to a temporary of type const char*
and then bind a reference to that temporary.  However, the reference
does not refer to a const object (because a pointer to const is not
itself a const), so it must be bound to a variable of the same type and
not a temporary.

The thing that makes this example tricky is that if we define

    typedef char* T;

then const T and const char* represent different types! -- const T is
"constant pointer to char" and const char* is "pointer to constant char".

The way to write "constant pointer to char" without a typedef it
char *const&, so the following should be OK:

    extern void f(char *const&);

    main()
    {
        char* p = /* some value */;
        f(p);
        // ...
    }

Here, p is of type "pointer to char" and there is no problem binding to
p a "reference to constant pointer to char".
--
                --Andrew Koenig
                  [e-mail]