Re: Is this a -Weffc++ bug in gcc?
On 08/17/2012 12:41 AM, red floyd wrote:
On 8/16/2012 2:51 PM, Scott Lurndal wrote:
Victor Bazarov <v.bazarov@comcast.invalid> writes:
On 8/16/2012 4:53 PM, DeMarcus wrote:
If I compile the following with gcc 4.7.1 and -Weffc++ then I get a
compiler warning on the operator% but not the others. Is this a bug?
struct A
{
A& operator%( int i ) { return *this; }
A& operator++() { return *this; }
A& fnc( int i ) { return *this; }
};
int main()
{
return 0;
}
And for those of us who don't have [any intention to use] gcc, what
compiler warning do you get? Just out of curiosity, of course...
$ g++ -Weffc++ -o /tmp/a.o -c /tmp/a.cc
/tmp/a.cc:3:25: warning: 'A& A::operator%(int)' should return by value
[-Weffc++]
Because the semantics of a binary arithmetic operator such as "%"
indicate that the return should be an rvalue. Scott Meyers discusses
this in "Effective C++" (which is what the -Weffc++ looks for).
Consider: Does the following make sense?
int x, y;
(x % y) = 3;
If, however, the following does make sense:
A a;
(a + 3) = 7;
Then feel free to ignore the warning.
The problem is that the warning tries to solve two problems:
1. As Meyers says; "Unrelenting in their pursuit of pass-by-reference
purity, they invariably make a fatal mistake: they start to pass
references to objects that don't exist."
2. Prevent returning lvalues where it doesn't make sense.
In number 1, 'return *this;' should be excluded in such warning since it
guarantees that there exists an object (unless we've done 'delete this;'
on a heap object).
As far as I see in Effective C++, Meyers doesn't mention anything about
lvalues, so number 2 is something made up by someone else. It's a smart
thing, but it should be another kind of warning as I see it.
Anyway, Weffc++ is just a guideline, but it contains a good set of
warnings and it doesn't feel good to disable it back and forth to fit
certain features.
Thanks,
Daniel