David Abrahams wrote:
I'm just looking through the variant library docs in detail and had the following commments:
Ah, if only boost::variant was actively maintained... I had several issues raised in http://lists.boost.org/Archives/boost/2006/06/105801.php and some time later I encountered another issue - I think the default constructor for variant is a bad idea. The thought came to me after I made a mistake of writing typedef variant< optional<X>, optional<Y> > SomeType; when I really meant was typedef optional< variant<X, Y> > SomeType; later in my code, thinking I had the right typedef, I constructed an instance of SomeType using its default constructor, assuming I had an empty optional<variant> in my hands, while I actually had a variant with an empty optional<X>. Unfortunately, it compiled, and that mistake was only found in run-time, with the help of the QA department. If variant didn't have a default constructor, my mistake wouldn't have compiled. IMO, this is also true from a more theoretical point of view - there is no reason to favor one of the variant types (the first one) over the others. Those types are basically equivalent and symmetric. If the user wishes to add a default constructor himself, he can derive from the variant, and add such a constructor, favoring the type of his choice. Or maybe the best solution would be to add a template parameter to boost::variant to indicate which of the types to use for the default constructor, with an option (the default one) to not supply a default constructor at all. Of course I'm just saying all this here for Google's sake. Nothing real would probably be done. Sad... Yuval