On 20 Aug 2009, at 23:57, Scott McMurray wrote:
2009/8/20 Edward Grace <ej.grace@imperial.ac.uk>:
if O is compatible with the concept of a functor and O returns a type that can be converted to a float, then we are good to go -- otherwise I'd like an intelligible error.
If, for example, O is of the type 'void (*)()' I'd like it (the compiler) to spit out, "The function pointer of type O must of signature float (*)()"
Sounds like you want something like the MPL_ASSERT example at [1] along with the result_of metafunction[2].
[1] http://www.boost.org/doc/libs/1_39_0/libs/mpl/doc/refmanual/ assert.html
[2] http://www.boost.org/doc/libs/1_39_0/libs/utility/ utility.htm#result_of
HTH, ~ Scott
Thanks for that tip Scott. Knowing where to look is always half the battle. I think I want to do something like the following (ignore _TEMPLATE_ and _TIMER_, they just expand to the usual template and scope for a template class). _TEMPLATE_ template <class O1, class O2> void _TIMER_ measure_percentage_speedup(O1 fa, O2 fb, double &MinPercent, double &MedPercent, double &MaxPercent) { // We want to make sure that the functions return something. // the following doesn't seem to work. typedef typename boost::result_of<O1()>::type result_of_fa; typedef typename boost::result_of<O2()>::type result_of_fb; BOOST_MPL_ASSERT(( boost::is_same<result_of_fa, unsigned> )); BOOST_MPL_ASSERT(( boost::is_same<result_of_fb, unsigned> )); ... ... No joy --- I can stick in whatever function type I want, with whatever return type (e.g. void) and the asserts don't complain. Is the above the type of thing you had in mind or have I got the wrong end of the stick? For what it's worth, I get an error if returning void from the function f() because of this subsequent line, in another part of the code, // Time a set of function calls. t0 = chrono(); for (counter n=0; n < iterations; ++n) test::live_code += f(); t1 = chrono(); however I'd like to make this requirement of a non-void return something more obvious. Thanks, -ed