Re: [RFC] std::string(0);

[Martin Sebor <sebor at roguewave dot com>]

Daniel Frey wrote:

> Martin Sebor wrote:
>
> > A conforming implementation is not allowed to reject constructs
> > such as std::string (0) just because they might cause undefined
> > behavior at runtime. Consider this well-formed program:
> >
> >     #include <string>
> >     int main () {
> >         if (0) std::string (0);
> >     }
> >
> > Martin
>
>
> OK, so this means we can't go the "easy" way. The remaining options are 
> (AFAICS):
>
> a) Change the standard, make the above example illegal. Unlikely.

I would not be happy about such a change -- our implementation
defines useful behavior in this case :) (it allocates storage
for the string object without initializing it).

> b) Ask the compiler folks to detected these cases and emit a warning.
> Bonus points for suppressing the warning on dead code branches like 
> above. Sadly, I suppose it's ways more work than adding one or two lines 
> to the STL implementation. Any volunteers? ;)

I don't think this would work, either. The compiler may not be
able to tell whether a branch of code is dead or not. Consider
this:

     #include <string>
     int main (int argc, char**) {
         if (argc > 1) std::string (0);
     }

Martin