On 31 October 2013 11:16, Christopher Jefferson <chris_at_[hidden]>wrote:

> Me and my wife teach C and C++ at Scottish Universities. I asked her
> what is the most important bit of UB she would like to see removed to
> make C++ easier to teach. Her suggestion (which I agree with) is how
> easy it is to no initialise variables, both in classes and on the
> stack. There are 2 main current ways I am aware of detecting UB from
> uninitalised variables:
>
> 1) Many compilers warn about stack variables being read before they
> are written to, with an appropriate warning level (but none provide
> these warnings with no warning flag, and some deductions require
> optimisation, which then means you have to compile twice to get a
> debugging version)
>
> 2) Tools like 'valgrind' will warn of more variable misuses.
>
> However, Rightly or wrongly, students will not use these tools
> correctly, and will get difficult to fix bugs.
>
> Her suggested fix is to introduce a new notation for constructors
> which means "I am not initialising this member on purpose". This will
> allow compilers to then add a warning at a high warning level that a
> member is missed from a constructor.
>
> I also wonder, what is people's opinion on telling students from day 1
> of a C++ class to always run their programs through valgrind, even
> their first "hello world" program? This would help them catch a much
> wider range of bugs earlier, but also teaches them a safety net they
> may come to excessively rely on.
>
> Chris
> _______________________________________________
> ub mailing list
> ub_at_[hidden]
> http://www.open-std.org/mailman/listinfo/ub
>

We have had discussions about this, in at least c++std-ext-12346
and probably in some core reflector discussions as well.

I don't think telling people to run valgrind is good advice. I think
building
with ubsan is better advice. I'd like to have a safer default, but I don't
think we can have it due to its breaking existing code.