Re: failure to warn about an obvious error

[Andrew Koenig <ark at research dot att dot com>]

Martin> As you probably know, it is very tricky to reliably tell whether a
Martin> function can "fall off the end". For example,

Martin>  #include <vector>

Martin>  std::vector<int> f(bool b) { 
Martin>    if (b)
Martin>      return std::vector<int>();
Martin>  }

Martin> is well-defined if this function is never invoked with "false".

Indeed, I do know.  Personally, I would consider a warning from the
example above to be a Good Thing.  However, I agree with you that
there are some reasonable programs that a compiler cannot detect as
probably erroneous.

What do you think about the idea of warning if a function is defined
as returning a value but there is no return statement anywhere in it
(and the function is not main)?  That would catch lots of common cases.