Re: [lmi] Re: tests checking for compilation failure

[Greg Chicares]

On 2010-12-16 13:22Z, Vadim Zeitlin wrote:
> On Thu, 16 Dec 2010 12:59:45 +0000 Greg Chicares <address@hidden> wrote:
> 
> GC> +// If lmi provided unit tests that deliberately fail to compile, then
> GC> +// this could be used:
> GC> +//
> GC> +// #include "uncopyable_lmi.hpp"
> GC> +//
> GC> +// class X : private lmi::uncopyable<X> {};
> GC> +//
> GC> +// int main()
> GC> +// {
> GC> +//     X x;
> GC> +//     X y(x); // Error: cannot copy.
> GC> +//     x = y;  // Error: cannot assign.
> GC> +// }
> 
>  Just an idle question that arose while reading the commit email: why
> doesn't LMI have such tests? It looks like it oughtn't be too difficult to
> implement a GNU make rule/canned sequence which could be used to verify
> that the given file fails to compile (possibly even checking for the
> presence of some string in the error output). It's difficult to do this
> portably but if you can rely on GNU make features it looks perfectly
> doable.

It has just never seemed worth the trouble (not that it would be a lot
of trouble). But today I needed at least a throwaway test; and, having
written it, I preserved it in a comment, against the day when creating
a general facility for such tests may seem worthwhile. In that day,
we'll get this one for free...

...just as this idea:
  
http://svn.savannah.nongnu.org/viewvc/lmi/trunk/view_ex.hpp?root=lmi&r1=5127&r2=5128
long lay dormant, but was convenient to awaken now that this usage:
    #include <boost/utility.hpp>
    // ...let some class inherit from boost::noncopyable...
has become untenable anyway with the latest Comeau compiler.