g++ 2.5.2 does not catch reference to local variable error.

[Jonathan Wakely]

On 4 August 2011 11:31, LIM Fung-Chai wrote:
> Hi,
>
> On Thu, Aug 4, 2011 at 5:29 PM, Miles Bader <miles@gnu.org> wrote:
>> Jonathan Wakely <jwakely.gcc@gmail.com> writes:
>>>> "g++ -Wall -Wextra ..." should flag a warning on the following code
>>>> but does not.
>>>
>>> Thanks for the apology, but it should still be reported to bugzilla
>>> not to this list.
>
> I was hoping someone could submit a bugzilla report.  I don't want to
> create an account which I may never use again.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49974

>>
>> BTW, it should only warn if given -Weffc++, right?
>
> -Weffc++ does not catch the dangerous error in my code.  -Wall caught
> another Item21 error somewhere else in my code:
>
> std::vector<XYZ*>&
> get_XYYs()
> {
>    std::vector<XYZ*> result;
>    return result;
> }
>
> I notice -Weffc++ produces more warnings on my code, but -Wall knows
> about Item21.  Both -Weffc++ and -Wall only see get_XYZs() as an
> Item21 violation but not get_XYZ_data() which returns the local
> variable as a reference in an std::pair.

Because -Weffc++ issues warnings about style guidelines (specifically,
Items 11, 12, 14, 15, 23 from Effective C++ and Items 6 and 67 from
More Effective C++) that not everyone agrees with in all cases.

Item 21 is about correctness, not style, returning a dangling
reference always warrants a warning.