Marc Glisse wrote:
On Sun, 23 Aug 2009, Peter van Hoof wrote:
Nikos Chantziaras wrote:
On 08/23/2009 09:32 PM, Peter van Hoof wrote:
Hi,
My understanding is that in C++ two overloaded versions of strstr()
should be defined:
const char * strstr ( const char * str1, const char * str2 );
char * strstr ( char * str1, const char * str2 );
(http://www.cplusplus.com/reference/clibrary/cstring/strstr/ but also
Stroustrup's book section 20.4.1 (special edition))
The way I read this, the following C++ code should be invalid:
#include <cstring>
char *backdoor(const char *a, const char *b)
{
return std::strstr(a, b);
}
since it is trying to remove the const-ness of the std:strstr() result.
However, g++ accepts this. It doesn't even give a warning with g++
-ansi
-Wall -W -pedantic. Most compilers I tried accept this, but it is my
understanding that VS08 rejects this code.
So am I reading this wrong, or is there a bug in g++?
GCC 4.4.1:
In function 'char* backdoor(const char*, const char*)':
error: invalid conversion from 'const char*' to 'char*'
So it seems you're using an old GCC which has this bug.
Actually, I tried g++ 4.5.0, 4.4.1, 4.3.4, 4.2.4, 4.1.2, 4.0.4, and
3.3.6. They all accept the code with the command line switches I posted.
What command line switches did you use?
Peter.
The problem is with libc more than g++. Solaris libc has been doing the
right thing for years, but g++ doesn't take advantage of it. GNU libc
quite recently moved to the right prototypes (for some headers) and
g++>=4.4 takes advantage of it. You can find many discussions of this in
bugzilla.
I see. I not only need a fairly new g++, but also a newer Linux
installation... Well, at least the problem is clearer now...
Thanks!
Peter.
