[c++-concepts] constrained friends
Added tests for constrained friends. No code, we already to the right thing. 2014-08-15 Andrew Sutton Add tests for constrained friends. * gcc/testsuite/g++.dg/concepts/friend1.C: New. * gcc/testsuite/g++.dg/concepts/friend2.C: New. Andrew Index: gcc/testsuite/g++.dg/concepts/friend2.C === --- gcc/testsuite/g++.dg/concepts/friend2.C (revision 0) +++ gcc/testsuite/g++.dg/concepts/friend2.C (revision 0) @@ -0,0 +1,20 @@ +// { dg-options "-std=c++1z" } + +template + concept bool Eq() { return requires(T t) { t == t; }; } + +template struct Foo { }; + +template + struct S { // { dg-error "constraint failure" } +template friend class Bar; + +friend class Foo; + }; + +struct X { }; + +int main() { + S si; // OK + S sx; +} Index: gcc/testsuite/g++.dg/concepts/friend1.C === --- gcc/testsuite/g++.dg/concepts/friend1.C (revision 0) +++ gcc/testsuite/g++.dg/concepts/friend1.C (revision 0) @@ -0,0 +1,33 @@ +// { dg-options "-std=c++1z" } + +template + concept bool Eq() { return requires(T t) { t == t; }; } + +struct Nt { + template friend void f(T) { } +} nt; + +template struct S; + +template + void proc(S*); + +template + struct S { +friend bool operator==(S, S) requires Eq() { return true; } + +friend void proc<>(S*); // { dg-error "does not match any template declaration" } + }; + +struct X { } x; + +int main() { + f(0); // OK + f(x); // { dg-error "cannot call" } + + S si; + si == si; // OK + + S sx; + sx == sx; // { dg-error "no match" } +}
Re: [c++-concepts] constrained friends redux
Hi Andrew, On 10/04/2013 07:36 PM, Andrew Sutton wrote: + if (!check_template_constraints (tmpl, args)) +{ + location_t loc = DECL_SOURCE_LOCATION (function); + error ("%qD is not a viable candidate", function); + diagnose_constraints (input_location, tmpl, args); + return error_mark_node; +} Nit: loc seems unused. Thanks, Paolo.
Re: [c++-concepts] constrained friends redux
OK. Jason
Re: [c++-concepts] constrained friends redux
>>> >Perhaps you mean that it must match a fully-instantiated function, so >>> > any >>> >constraints on the templates were considered during >>> >determine_specialization. > > >> This seems like a simple comment fix, but there's a longer explanation >> of what I want (see below). Would this be more appropriate? >> >>// Do not allow constrained friend template specializations. >> >> The intent is stronger than to say it must match something. > > > By "must" I meant that whatever it matches could only be a > fully-instantiated function. I see what you mean. I was caught up in the wrong part of the sentence.. But yes, that's right. > But I guess the main reason for disallowing constraints here is the same as > for explicit specializations and non-template functions; non-dependent > constraints don't really make any sense. That's the intent. Okay to commit? Andrew friends-3.patch Description: Binary data
Re: [c++-concepts] constrained friends redux
On 10/04/2013 09:20 AM, Andrew Sutton wrote: >Perhaps you mean that it must match a fully-instantiated function, so any >constraints on the templates were considered during >determine_specialization. This seems like a simple comment fix, but there's a longer explanation of what I want (see below). Would this be more appropriate? // Do not allow constrained friend template specializations. The intent is stronger than to say it must match something. By "must" I meant that whatever it matches could only be a fully-instantiated function. But I guess the main reason for disallowing constraints here is the same as for explicit specializations and non-template functions; non-dependent constraints don't really make any sense. Jason
Re: [c++-concepts] constrained friends redux
>> + // Do not permit the declaration of constrained friend >> + // function declarations. They cannot be instantiated since >> + // the resulting declaration would never match the definition, >> + // which must be a non-template and cannot be constrained. > > > You're in the template-id code here, so "must be a non-template" is > confusing: > > template void f(); > > struct A { > friend void f(); // matches a template > }; > > Perhaps you mean that it must match a fully-instantiated function, so any > constraints on the templates were considered during > determine_specialization. This seems like a simple comment fix, but there's a longer explanation of what I want (see below). Would this be more appropriate? // Do not allow constrained friend template specializations. The intent is stronger than to say it must match something. I don't want to allow any declarations of the form template struct X { friend void f<>(T x) requires C; // Error. }; This should never even get to determine_specialization since the original declaration is never actually pushed. We could use those constraints to match the specialization to one of several constrained overloads, as you mentioned earlier, but I'd rather avoid that for now. Solving that problem in general would require that we allow constrained (explicit) specializations and define a method of matching instantiated constraints to dependent constraints, and that we do so as an alternative to the usual constraint checking during template argument deduction. Maybe it's a useful feature, but it's really hard to gauge how much use it would actually get. We can always revisit that in the future. Somebody else can write that paper :) Andrew
Re: [c++-concepts] constrained friends redux
On 10/02/2013 09:05 AM, Andrew Sutton wrote: + // Do not permit the declaration of constrained friend + // function declarations. They cannot be instantiated since + // the resulting declaration would never match the definition, + // which must be a non-template and cannot be constrained. You're in the template-id code here, so "must be a non-template" is confusing: template void f(); struct A { friend void f(); // matches a template }; Perhaps you mean that it must match a fully-instantiated function, so any constraints on the templates were considered during determine_specialization. + error("constrained friend does not depend on template parameters"); Space before (. +// Returns true if FN is a non-template member function. +static inline bool is_non_template_member_fn (tree fn) { return DECL_FUNCTION_MEMBER_P (fn) && @@ -1829,6 +1829,21 @@ is_non_template_member_fn (tree fn) !DECL_MEMBER_TEMPLATE_P (DECL_TI_TEMPLATE (fn)); } +// Returns true if FN is a non-template friend definition. +static inline bool +is_non_template_friend (tree fn) These names/comments fail to make it clear that they return true only for non-template members/friends *of class template specializations*. So my preference would be to open-code them into is_constrainable_non_template_fn. Jason
[c++-concepts] constrained friends redux
This patch implements constrained friends and disallows declarations of constrained friend template specialization. There was a previous question about whether I was doing the right thing in determine_specialization. I'm looking at that issue separately. 2013-10-01 Andrew Sutton * gcc/cp/parser.c (cp_parser_member_declaration): Check that a constrained friend definition is valid. * gcc/cp/decl.c (grokfndecl): Disallow constrained friend template specializations. * gcc/cp/constraints.cc (check_constrained_friend): New. * gcc/cp/typeck.c (cp_build_function_call_vec): Diagnose constraints in the presence of the failure of a single candidate. * gcc/cp/cp-tree.h (check_constrained_friend): New. * gcc/cp/call.c (is_non_template_member_fn): Make inline. (is_non_template_friend), (is_constrainable_non_template_fn): New. (add_function_candidate): Predicate check on is_constrainable_non_template_fn. Andrew Sutton friends-2.patch Description: Binary data
Re: [c++-concepts] Constrained friends
On 09/21/2013 08:52 AM, Andrew Sutton wrote: It is wrong, but not for the reasons I gave. This only happens when you try to constrain a friend function that declares a specialization, which happens to be completely separate from the previously declared template. I'm going to disallow the ability to declare constrained friend specializations. They don't really make sense. Hmm, it seems to me the constraints on the specialization help to select the template the specialization belongs to. I'm not sure that would be useful, but I don't see a need to specifically prevent it. Jason
Re: [c++-concepts] Constrained friends
I'm going to rewrite this patch tomorrow morning. The semantics aren't quite right --- they should be simpler. >> Previously, if constraints were not >> satisfied, we would not record the template as a candidate. However, >> this causes errors in class template instantiation if there are >> constrained friend declarations whose constraints are not satisfied >> ("no matching template declaration"). > > Is that wrong? We normally give errors about friend declarations that don't > match any template. Why don't we want this error when the friend > declaration is looking for a constrained template? It is wrong, but not for the reasons I gave. This only happens when you try to constrain a friend function that declares a specialization, which happens to be completely separate from the previously declared template. I'm going to disallow the ability to declare constrained friend specializations. They don't really make sense. >> + if (is_non_template_member_fn (fn) || is_non_template_friend (fn)) > > > Let's have one predicate instead of two; the condition here is a temploid > that is not a specialization of a primary template. Agreed. > >> + if (!check_template_constraints (tmpl, args) && (complain & >> tf_error)) >> { >>reason = template_constraint_failure (tmpl, args); >>viable = false; > > > Why add the complain check? A constraint failure should make a candidate > non-viable even in SFINAE context. What have I done? That's awful... >> + // If we're not instantiating a friend function, then we need to >> + // ensure the specialization of the best template satisfies its >> + // constraints. > > > Surely we need to check constraints in the earlier loop, so that we don't > treat as a candidate a template that doesn't satisfy the constraints; > otherwise if we have two templates > > template T f(T) requires Thing; > template T* f(T*); > > and our specialization requires Thing, we would select the second (because > it is otherwise more specialized) and then give an error about constraint > mismatch; I would think we want to select the first. I believe that's what the previous version did, and we'll go back to that. This change was part of the semantics that I didn't get right. >> +// If there is an overload with the same type and >> +// constraints, then this is a good declaration. >> +if (same_type_p (TREE_TYPE (fn), TREE_TYPE (f))) >> + if (equivalent_constraints (constr, get_constraints (f))) >> +return; > > > It seems that this will only allow friend declarations that match the > template exactly, not friend declarations that are more specialized than the > matching template. It looks like you're trying to implement a subset of > determine_specialization here, which I think is a mistake. I agree. It's a mistake. This is also related to the semantics that I got wrong. Effectively, the only changes needed for constrained friends are that: a) you can't constrain non-dependent friends b) you can't constraint non-template friend functions that declare a specialization and c) that we check non-template friends the same way as non-template member fns There should be no changes to any of the rules for determining specializations. Andrew
Re: [c++-concepts] Constrained friends
On 09/13/2013 12:21 PM, Andrew Sutton wrote: Previously, if constraints were not satisfied, we would not record the template as a candidate. However, this causes errors in class template instantiation if there are constrained friend declarations whose constraints are not satisfied ("no matching template declaration"). Is that wrong? We normally give errors about friend declarations that don't match any template. Why don't we want this error when the friend declaration is looking for a constrained template? + if (is_non_template_member_fn (fn) || is_non_template_friend (fn)) Let's have one predicate instead of two; the condition here is a temploid that is not a specialization of a primary template. + if (!check_template_constraints (tmpl, args) && (complain & tf_error)) { reason = template_constraint_failure (tmpl, args); viable = false; Why add the complain check? A constraint failure should make a candidate non-viable even in SFINAE context. + // If we're not instantiating a friend function, then we need to + // ensure the specialization of the best template satisfies its + // constraints. Surely we need to check constraints in the earlier loop, so that we don't treat as a candidate a template that doesn't satisfy the constraints; otherwise if we have two templates template T f(T) requires Thing; template T* f(T*); and our specialization requires Thing, we would select the second (because it is otherwise more specialized) and then give an error about constraint mismatch; I would think we want to select the first. +// If there is an overload with the same type and +// constraints, then this is a good declaration. +if (same_type_p (TREE_TYPE (fn), TREE_TYPE (f))) + if (equivalent_constraints (constr, get_constraints (f))) +return; It seems that this will only allow friend declarations that match the template exactly, not friend declarations that are more specialized than the matching template. It looks like you're trying to implement a subset of determine_specialization here, which I think is a mistake. Jason
[c++-concepts] Constrained friends
This patch implements semantics for constrained friend templates and classes. The only significant changes are in determine_specializaiton and check_constrained_friend. Unless a friend function is defined, a constraints on friend declarations are never actually checked. The checking happens during overload resolution against the actual (non-friend) declarations. The change to determine_specialization is interesting. We have cases where a programmer has written an explicit specialization and it's being matched to a template. Previously, if constraints were not satisfied, we would not record the template as a candidate. However, this causes errors in class template instantiation if there are constrained friend declarations whose constraints are not satisfied ("no matching template declaration"). With this patch, we defer constraints checks until we've selected the best template for the specialization. And then we only check the constraints if the declaration is a non-friend. 2013-09-13 Andrew Sutton * gcc/cp/cp-tree.h (check_constrained_friend): New. * gcc/cp/pt.c (determine_specialization): Only check constraints after determining which template the declaraiton is a specialization of. Don't check constraints for friends during class template instantiation. (fn_type_unification): New parameter to determine if constraints should be checked. (more_specialized_class): Update for interface change. (get_bindings): New parameter to determine if constraints should be checked during fn_type_unification. New overload that checks constraints by default. * gcc/cp/parser.c (cp_parser_member_declaration): Check constrained friends after parsing. * gcc/cp/class.c (resolve_address_of_overloaded_function): Update for interface change. * gcc/cp/call.c (is_non_template_friend): New. (add_function_candidate): Check constraints on constrained friend templates. (add_template_candidate_real): Update for interface change. * gcc/cp/constraint.c (check_constrained_friend): New. Andrew Sutton friends.patch Description: Binary data