From -3504224610436976580 X-Google-Thread: f78e5,e1019ee5aaa65968 X-Google-NewGroupId: yes X-Google-Attributes: gid7894ca11fe,domainid0,public,usenet X-Google-Language: ENGLISH,ASCII-7-bit Path: g2news2.google.com!news2.google.com!news.glorb.com!news.alt.net From: Richard Smith Newsgroups: comp.std.c++ Subject: Re: Specializing a template template-parameters with a template template class. Date: Wed, 27 Apr 2011 10:54:22 CST Organization: http://groups.google.com Lines: 103 Sender: std-cpp-request@vandevoorde.com Approved: stephen.clamage@oracle.com Message-ID: References: <472cc17b-b5ee-4897-81e7-c52fe1b228ad@a11g2000pro.googlegroups.com> Content-Type: text/plain; charset=ISO-8859-1 To: undisclosed-recipients:; Return-Path: X-Original-Date: Wed, 27 Apr 2011 07:30:45 -0700 (PDT) X-Submission-Address: std-c++-submit@vandevoorde.com Xref: g2news2.google.com comp.std.c++:3275 On Mar 10, 2:21 pm, itaj sherman wrote: > Has the standard ever considered the option of defining template of > template of class? [...] > To demonstrate the problem: > > template< template< typename T> class> class A; > class B {}; > > template< typename T, typename U> class X {}; > //and hypothetic equivalent: > template< typename T> template< typename U> class Y {}; > //or > template< typename T> template< typename U> using Y = X< T, U>; > > Using the above hypothetical syntax, Y and X could be considered > equivalent X< T, U> == Y< T>< U>. > However, for any type T, Y is a template class (with 1 parameter), > and it might make sense to be able to specialize the first parameter > of A on Y: > > //hypothetically specialize: > template< typename T> class A< Y > {}; > //instanciated by A< Y > > > With X it's not possible to do such specialization. There's a caveat here: template arguments for template template parameters always refer to template declarations, and aren't the same even if the two declarations are equivalent. If the above were allowed, 'equivalent' alias template template specializations would naturally, by extension, refer to different alias templates. Given: template template using Z = X; Y and Z would be different templates, so the partial specialization for A> would not match the template specialization A>, so X and Y would not really be equivalent-up-to-currying in the desired sense. [Up until FDIS, the working draft contained this example: template class TT> struct X { }; template struct Y { }; template using Z = Y; X y; X z; declares y and z to be of the same type. ... which, if it weren't wrong, would imply that templates are compared by alpha-beta-eta equivalence (rather than by name), and that Y and Z above would be equivalent. I consider it unfortunate that the example was changed, rather than the normative wording!] > Nesting cannot solve that either: > > template< typename T> class Z > { > public: template< typename U> class R {}; > > }; > > template< typename T> class A< Z::R> {}; > //cannot instanciate by A< Z::R> > > It's not possible to deduce T in a member template class Z::R. Right; a nested name specifier is a non-deduced context. Alias templates, as per the FDIS, don't help here, since they don't change this rule. > Has this ever been considered as a c++ feature? I've been implementing support for alias templates in the clang compiler, and run across this limitation while building testcases for the feature. I've certainly considered proposing template templates as an extension for C++1x. As well as covering the deduction case, this change would have additional benefits: * Alias templates would gain the full computational power of the simply-typed lambda calculus (plus variadic types, but I think those can always be desugared into plain STLC). [Hence alias template expansion is still guaranteed to terminate.] * It would become simpler to express multiple variadic template parameter lists for a template: template template struct S; S<>> s; or template template struct Tuple; -- Richard -- [ comp.std.c++ is moderated. To submit articles, try posting with your ] [ newsreader. If that fails, use mailto:std-cpp-submit@vandevoorde.com ] [ --- Please see the FAQ before posting. --- ] [ FAQ: http://www.comeaucomputing.com/csc/faq.html ]