Re: Strange behaviour of strchr in C++

[Joe Buck <Joe dot Buck at synopsys dot COM>]

Mateusz Åoskot wrote:
> > By the way, should it be reported to libstdc++ bugzilla?

On Tue, Feb 28, 2006 at 01:24:56AM +0100, Paolo Carlini wrote:
> I don't think so, really. The problem (and its relatives) is very well
> known, basically is subsumed by libstdc++/6257. Also, AFAIK, there are
> hopes that the next standard will make the life easier to the
> implementors wrt this kind of issues.

It might be possible to fix this issue without fixing all of 6257.
A <cstring> that does not include <string.h>, but that provides the
correct declarations, together with code to implement the two C++
strchr's as the C strchr with appropriate const_casts, would do the
job.  The current problem is that <cstring> contains <string.h>.

This could be extended to a partial fix for 6257 as a whole: the C
names would remain in the global namespace, but their global namespace
declarations would not appear in the standard headers.  For example,
we'd have declarations for std::printf, and appropriate magic so
that it is linked to the C printf global.

This leaves the problem that ISO C++ lets someone write

----------------
#include <cstdio>
// printf is not reserved, I can define it.
int printf = 42;

int main() {
    std::printf("The answer is %d\n", printf);
}
----------------

One possibility might be a kind of "swap".  For the C standard names,

C++ name	Linker name
std::printf  => printf
::printf     => __BLAH_printf

(Replace BLAH by something appropriate).  That is, swap the roles
of std:: and the global namespace for the standard C symbols that
ISO requires that C++ provide in the std:: namespace.

An issue is that this changes the ABI.  However, it might be that this
doesn't matter (if only programs that are problematic right now
are affected).