| Message ID | 20200501120310.GA498115@redhat.com |
|---|---|
| State | New |
| Headers | show |
| Series | libstdc++: Replace deduced return type in ranges::iter_move (PR 92894) | expand |
On 01/05/20 13:03 +0100, Jonathan Wakely wrote: >The deduced return type causes the instantiation of the function body, >which can then require the instantiation of std::projected::operator* >which is intentionally not defined. > >This patch uses a helper trait to define the return type, so that the >function body doesn't need to be instantiated. That helper trait can >then also be used in other places that currently check the return type >of ranges::iter_move (iter_rvalue_reference_t and indirectly_readable). > >2020-05-01 Jonathan Wakely <jwakely@redhat.com> > Patrick Palka <ppalka@redhat.com> > > PR libstdc++/92894 > * include/bits/iterator_concepts.h (ranges::__cust_imove::_IMove): > Add trait to determine return type and an alias for it. > (ranges::__cust_imove::_IMove::operator()): Use __result instead of > deduced return type. > (iter_rvalue_reference_t): Use _IMove::__type instead of checking > the result of ranges::iter_move. > (__detail::__indirectly_readable_impl): Use iter_rvalue_reference_t > instead of checking the result of ranges::iter_move. > * testsuite/24_iterators/indirect_callable/92894.cc: New test. > >Patrick, I prefer this to the patch you added to the bug. Avoiding >doing overload resolution on ranges::iter_move(x) seems worthwhile. > >What do you think? > >I *think* the changes to __indirectly_readable_impl are equivalent. As >far as I can tell, the point of the original { *in } >and { ranges::iter_move(in) } constraints are to check that const In >and In do the same thing. We could leave the { *in } on, but for >consistency it seems better to change both. Here's a slightly improved patch which I'm committing now. Diffs from the previous patch are removing the useless parameter list from the __indirectly_readable_impl concept: @@ -460,7 +461,7 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION using __iter_concept = typename __iter_concept_impl<_Iter>::type; template<typename _In> - concept __indirectly_readable_impl = requires(const _In __in) + concept __indirectly_readable_impl = requires { typename iter_value_t<_In>; typename iter_reference_t<_In>; Fixing 24_iterators/indirect_callable/92894.cc so it actually passes, and adding 24_iterators/customization_points/92894.cc to verify that iter_move is fixed (which the other tests don't do because indirectly_readable no longer uses ranges::iter_move). Tested powerpc64le-linux, committed to master.
On 01/05/20 14:28 +0100, Jonathan Wakely wrote: >On 01/05/20 13:03 +0100, Jonathan Wakely wrote: >>The deduced return type causes the instantiation of the function body, >>which can then require the instantiation of std::projected::operator* >>which is intentionally not defined. >> >>This patch uses a helper trait to define the return type, so that the >>function body doesn't need to be instantiated. That helper trait can >>then also be used in other places that currently check the return type >>of ranges::iter_move (iter_rvalue_reference_t and indirectly_readable). >> >>2020-05-01 Jonathan Wakely <jwakely@redhat.com> >> Patrick Palka <ppalka@redhat.com> >> >> PR libstdc++/92894 >> * include/bits/iterator_concepts.h (ranges::__cust_imove::_IMove): >> Add trait to determine return type and an alias for it. >> (ranges::__cust_imove::_IMove::operator()): Use __result instead of >> deduced return type. >> (iter_rvalue_reference_t): Use _IMove::__type instead of checking >> the result of ranges::iter_move. >> (__detail::__indirectly_readable_impl): Use iter_rvalue_reference_t >> instead of checking the result of ranges::iter_move. >> * testsuite/24_iterators/indirect_callable/92894.cc: New test. >> >>Patrick, I prefer this to the patch you added to the bug. Avoiding >>doing overload resolution on ranges::iter_move(x) seems worthwhile. >> >>What do you think? >> >>I *think* the changes to __indirectly_readable_impl are equivalent. As >>far as I can tell, the point of the original { *in } >>and { ranges::iter_move(in) } constraints are to check that const In >>and In do the same thing. We could leave the { *in } on, but for >>consistency it seems better to change both. > >Here's a slightly improved patch which I'm committing now. > >Diffs from the previous patch are removing the useless parameter list >from the __indirectly_readable_impl concept: > >@@ -460,7 +461,7 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION > using __iter_concept = typename __iter_concept_impl<_Iter>::type; > > template<typename _In> >- concept __indirectly_readable_impl = requires(const _In __in) >+ concept __indirectly_readable_impl = requires > { > typename iter_value_t<_In>; > typename iter_reference_t<_In>; > >Fixing 24_iterators/indirect_callable/92894.cc so it actually passes, >and adding 24_iterators/customization_points/92894.cc to verify that >iter_move is fixed (which the other tests don't do because >indirectly_readable no longer uses ranges::iter_move). This is the patch I'm testing for the gcc-10 branch (if it gets RM approval). It doesn't do the compile-time optimization for iter_rvalue_reference_t and __indirectly_readable_impl, it just fixes the bug in ranges::iter_move. The optimization isn't required for correctness.
diff --git a/libstdc++-v3/include/bits/iterator_concepts.h b/libstdc++-v3/include/bits/iterator_concepts.h index b598532089e..d84e8639ea8 100644 --- a/libstdc++-v3/include/bits/iterator_concepts.h +++ b/libstdc++-v3/include/bits/iterator_concepts.h @@ -89,6 +89,21 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION struct _IMove { private: + template<typename _Tp> + struct __result + { using type = iter_reference_t<_Tp>; }; + + template<typename _Tp> + requires __adl_imove<_Tp> + struct __result<_Tp> + { using type = decltype(iter_move(std::declval<_Tp>())); }; + + template<typename _Tp> + requires (!__adl_imove<_Tp>) + && is_lvalue_reference_v<iter_reference_t<_Tp>> + struct __result<_Tp> + { using type = remove_reference_t<iter_reference_t<_Tp>>&&; }; + template<typename _Tp> static constexpr bool _S_noexcept() @@ -100,16 +115,19 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION } public: - template<typename _Tp> - requires __adl_imove<_Tp> || requires(_Tp& __e) { *__e; } - constexpr decltype(auto) + // The result type of iter_move(std::declval<_Tp>()) + template<std::__detail::__dereferenceable _Tp> + using __type = typename __result<_Tp>::type; + + template<std::__detail::__dereferenceable _Tp> + constexpr __type<_Tp> operator()(_Tp&& __e) const noexcept(_S_noexcept<_Tp>()) { if constexpr (__adl_imove<_Tp>) return iter_move(static_cast<_Tp&&>(__e)); - else if constexpr (is_reference_v<iter_reference_t<_Tp>>) - return std::move(*__e); + else if constexpr (is_lvalue_reference_v<iter_reference_t<_Tp>>) + return static_cast<__type<_Tp>>(*__e); else return *__e; } @@ -123,10 +141,9 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION } // namespace ranges template<__detail::__dereferenceable _Tp> - requires requires(_Tp& __t) - { { ranges::iter_move(__t) } -> __detail::__can_reference; } + requires __detail::__can_reference<ranges::__cust_imove::_IMove::__type<_Tp&>> using iter_rvalue_reference_t - = decltype(ranges::iter_move(std::declval<_Tp&>())); + = ranges::__cust_imove::_IMove::__type<_Tp&>; template<typename> struct incrementable_traits { }; @@ -448,8 +465,10 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION typename iter_value_t<_In>; typename iter_reference_t<_In>; typename iter_rvalue_reference_t<_In>; - { *__in } -> same_as<iter_reference_t<_In>>; - { ranges::iter_move(__in) } -> same_as<iter_rvalue_reference_t<_In>>; + requires same_as<iter_reference_t<const _In>, + iter_reference_t<_In>>; + requires same_as<iter_rvalue_reference_t<const _In>, + iter_rvalue_reference_t<_In>>; } && common_reference_with<iter_reference_t<_In>&&, iter_value_t<_In>&> && common_reference_with<iter_reference_t<_In>&&, diff --git a/libstdc++-v3/testsuite/24_iterators/indirect_callable/92894.cc b/libstdc++-v3/testsuite/24_iterators/indirect_callable/92894.cc new file mode 100644 index 00000000000..78c4585923f --- /dev/null +++ b/libstdc++-v3/testsuite/24_iterators/indirect_callable/92894.cc @@ -0,0 +1,59 @@ +// Copyright (C) 2020 Free Software Foundation, Inc. +// +// This file is part of the GNU ISO C++ Library. This library is free +// software; you can redistribute it and/or modify it under the +// terms of the GNU General Public License as published by the +// Free Software Foundation; either version 3, or (at your option) +// any later version. + +// This library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. + +// You should have received a copy of the GNU General Public License along +// with this library; see the file COPYING3. If not see +// <http://www.gnu.org/licenses/>. + +// { dg-options "-std=gnu++2a" } +// { dg-do compile { target c++2a } } + +#include <iterator> + +using std::projected; +using std::identity; +using std::indirect_unary_predicate; + +template<typename T, + indirect_unary_predicate<projected<T*, identity>> Pred> + constexpr void + all_of(T*, Pred) + { } + +void +test01() +{ + // PR libstdc++/92894 + struct X { }; + X x; + all_of(&x, [](X&) { return false; }); +} + +using std::iterator_t; +using std::ranges::input_range; +using std::ranges::borrowed_range; + +template<input_range R, class Proj = identity, + indirect_unary_predicate<projected<iterator_t<R>, Proj>> Pred> + constexpr borrowed_iterator_t<R> + find_if(R&& r, Pred pred, Proj proj = {}) + { } + +void +test02() +{ + // PR 94241 + struct s { int m; }; + s r[] = { s{0}, s{1}, s{2}, s{3} }; + find_if(r, [](auto const) { return true; }); +}
The deduced return type causes the instantiation of the function body, which can then require the instantiation of std::projected::operator* which is intentionally not defined. This patch uses a helper trait to define the return type, so that the function body doesn't need to be instantiated. That helper trait can then also be used in other places that currently check the return type of ranges::iter_move (iter_rvalue_reference_t and indirectly_readable). 2020-05-01 Jonathan Wakely <jwakely@redhat.com> Patrick Palka <ppalka@redhat.com> PR libstdc++/92894 * include/bits/iterator_concepts.h (ranges::__cust_imove::_IMove): Add trait to determine return type and an alias for it. (ranges::__cust_imove::_IMove::operator()): Use __result instead of deduced return type. (iter_rvalue_reference_t): Use _IMove::__type instead of checking the result of ranges::iter_move. (__detail::__indirectly_readable_impl): Use iter_rvalue_reference_t instead of checking the result of ranges::iter_move. * testsuite/24_iterators/indirect_callable/92894.cc: New test. Patrick, I prefer this to the patch you added to the bug. Avoiding doing overload resolution on ranges::iter_move(x) seems worthwhile. What do you think? I *think* the changes to __indirectly_readable_impl are equivalent. As far as I can tell, the point of the original { *in } and { ranges::iter_move(in) } constraints are to check that const In and In do the same thing. We could leave the { *in } on, but for consistency it seems better to change both. commit 312bf4457effa0ce99c7c4cb0ebe5e3fb0fd3747 Author: Jonathan Wakely <jwakely@redhat.com> Date: Fri May 1 10:50:50 2020 +0100 libstdc++: Replace deduced return type in ranges::iter_move (PR 92894) The deduced return type causes the instantiation of the function body, which can then require the instantiation of std::projected::operator* which is intentionally not defined. This patch uses a helper trait to define the return type, so that the function body doesn't need to be instantiated. That helper trait can then also be used in other places that currently check the return type of ranges::iter_move (iter_rvalue_reference_t and indirectly_readable). 2020-05-01 Jonathan Wakely <jwakely@redhat.com> Patrick Palka <ppalka@redhat.com> PR libstdc++/92894 * include/bits/iterator_concepts.h (ranges::__cust_imove::_IMove): Add trait to determine return type and an alias for it. (ranges::__cust_imove::_IMove::operator()): Use __result instead of deduced return type. (iter_rvalue_reference_t): Use _IMove::__type instead of checking the result of ranges::iter_move. (__detail::__indirectly_readable_impl): Use iter_rvalue_reference_t instead of checking the result of ranges::iter_move. * testsuite/24_iterators/indirect_callable/92894.cc: New test.