std::wcout and const char *

[Erik Knudsen]

Hi!

In converting applications from ansi to unicode, it is a problem that
the std::wcout accepts a const char * without complaining compile time.

This compiles and runs:
------------------------------
const char *lookup()
{
return "Hello world";
}
std::wcout << lookup();
------------------------------

I would like it to fail compile time, so I can properly convert cases
like this to unicode in a meaningful way.

Any ideas?


Regards,
Erik Knudsen

 

[Stefan Naewe]

Hi!

In converting applications from ansi to unicode, it is a problem that
the std::wcout accepts a const char * without complaining compile time.

This compiles and runs:
------------------------------
const char *lookup()
{
return "Hello world";
}
std::wcout << lookup();
------------------------------

I would like it to fail compile time, so I can properly convert cases
like this to unicode in a meaningful way.

Any ideas?

Declare and do not define std:perator<<(wostream&, char const*) yourself?

namespace std
{
inline
wostream& operator<<(wostream& wos, const char*);
}


Just an idea...


S.

 

[Erik Knudsen]

Declare and do not define std:perator<<(wostream&, char const*) yourself?

namespace std
{
inline
wostream& operator<<(wostream& wos, const char*);
}


Just an idea...

Thanks, but wouldn't that just give me a linker error, with no reference
to code lines?


Regards,
Erik Knudsen

 

[Stefan Naewe]

Thanks, but wouldn't that just give me a linker error, with no reference
to code lines?

Yes...but:

If I call 'g++ -g' the linker gives me the file and line number:


g++ -g -Wall wchar.cc -o wchar
/tmp/ccLf8VPM.o: In function `main':
wchar.cc:37: undefined reference to `std:perator<<(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >&, char const*)'
collect2: ld returned 1 exit status
make: *** [wchar] Fehler 1


HTH

S.

 

[Stefan Naewe]

Yes...but:

If I call 'g++ -g' ...

and remove the 'inline'...

the linker gives me the file and line number:


g++ -g -Wall wchar.cc -o wchar
/tmp/ccLf8VPM.o: In function `main':
wchar.cc:37: undefined reference to `std:perator<<(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >&, char const*)'
collect2: ld returned 1 exit status
make: *** [wchar] Fehler 1

I don't know why I put 'inline' there a all...

S.

 

[James Kanze]

On 12/11/2007 10:16 AM, Erik Knudsen wrote:

Declare and do not define std:perator<<(wostream&, char
const*) yourself?

namespace std
{
inline
wostream& operator<<(wostream& wos, const char*);
}

Just an idea...

That's undefined behavior. You're not allowed to define new
free functions in a std::.

The standard std::wostream defines an operator<< for char
const*. You have to assume that some of the other operator<<
for wostream may use it, so you really have to leave it.

 

[Stefan Naewe]

That's undefined behavior. You're not allowed to define new
free functions in a std::.

I knew someone would say that...

But really: Who cares in this case if it just works (with some
modifications and only (?) with g++ (see my other post)) ?

The standard std::wostream defines an operator<< for char
const*. You have to assume that some of the other operator<<
for wostream may use it, so you really have to leave it.

S.

 

[James Kanze]

On 12/12/2007 1:41 PM, James Kanze wrote:

I knew someone would say that...

But really: Who cares in this case if it just works (with some
modifications and only (?) with g++ (see my other post)) ?

Does it work? It might, or it might not. Or it might work some
of the time, and not others. There are things you can
reasonably define in std::, and expect to get away with it. But
this isn't one of them.

And that is why. If you declare your version of the function,
either:

-- The only version in the standard library is the template
version. Your version overloads it, and since it is not a
template, has precedence in operator overload resolution.
If the library has overloaded the << operator for, say,
complex with something like:

template< typename Float, typename CharT ... >
std::basic_ostream< CharT ... >&
operator<<( std::basic_ostream< CharT ... >& dest,
std::complex< Float > const& z )
{
dest << "(" << z.real() << "," << z.imag() << ")" ;
return dest ;
}

then this function may call your version, rather than the
standard one; which one will be chosen more or less
randomly.

-- If the standard library has also declared the function
(which would seem a more or less likely "optimization" for
the usual cases of std:stream and std::wostream), then
your declaration just duplicates the one in the standard
library, and doesn't change anything.

The standard limits what you can do in std:: for a very real
reason.

 

[Pete Becker]

Does it work? It might, or it might not. Or it might work some
of the time, and not others. There are things you can
reasonably define in std::, and expect to get away with it. But
this isn't one of them.

But as a short-term debugging aid, this is fine: if it detects
problems, you fix them. Once you've done that, you remove it. I've
occasionally gone so far as to add #error statements to standard
headers when I couldn't figure out why some header was being included
in a bogus context. (One of the Windows headers pulls in <stdlib.h>
while inside an extern "C" block. Sigh.)

 

[Erik Knudsen]

But as a short-term debugging aid, this is fine: if it detects problems,
you fix them. Once you've done that, you remove it. I've occasionally
gone so far as to add #error statements to standard headers when I
couldn't figure out why some header was being included in a bogus
context. (One of the Windows headers pulls in <stdlib.h> while inside an
extern "C" block. Sigh.)

I will only use it as a short term porting/debugging aid. However, I use
VC++ 2005, and it doesn't give me the line numbers for a link error.


Regards,
Erik Knudsen

 

[Erik Knudsen]

Ok, so far the best approach has been to comment out the std stream
operators for char *, void * and bool from the stl include files - that
made the compiler complain when we tried to stream a char *.

However, I'm still looking for a nicer approach...


Regards,
Erik Knudsen