From: Edward Diener (eddielee_at_[hidden])
Date: 2005-05-03 12:13:39
Andrei Alexandrescu (See Website For Email) wrote:
> Edward Diener wrote:
>
>> Andrei Alexandrescu (See Website For Email) wrote:
>>
>>> By the way, it was FOREACH that inspired me to figure out a simple
>>> solution to max, which then Eric improved. It's basically like this:
>>>
>>> #define max(a, b) (max_fn(true ? (a) : (b), false ? (a) : (b)))
>>>
>>> which presto, nice-o passes the appropriate types to a template
>>> max_fn function that now can be smart about returning lvalues and so on.
>>
>>
>>
>> You have stumped me here. Why does one not do:
>>
>> #define max(a, b) (max_fn((a),(b))
>>
>> instead ?
>
>
> I thought nobody's gonna ask :o).
>
> The problem on the former code is that it puts a lot of aggravation on
> max_fn as far as deducing the proper type returned; see
> http://moderncppdesign.com/publications/cuj-04-2001.html for a
> discussion on why.
>
> On the contrary, the trick (used also, and inspired from, FOREACH):
>
> #define max(a, b) (max_fn(true ? (a) : (b), false ? (a) : (b)))
>
> lets the ?:'s rules figure out that type; now max_fn has two arguments
> of the same type, and can much more easily figure out what to return
> (only needs to deal with const and rvalues, stuff that was piece of cake
> for Eric to figure out).
I just re-read the conditional operator specs in 5.16 of the standard
and did not realize how much was there. Thanks for alerting me to this
area of the language.