Re: [PATCH] c++: Fix access checking for __is_assignable and __is_constructible (c++/94197)
On 3/17/20 11:25 AM, Ville Voutilainen wrote: On Tue, 17 Mar 2020 at 16:52, Ville Voutilainen wrote: On Tue, 17 Mar 2020 at 16:42, Jason Merrill wrote: On 3/17/20 9:04 AM, Jonathan Wakely wrote: On 17/03/20 13:02 +, Jonathan Wakely wrote: Shouldn't the test use { dg-do compile { target c++11 } } instead of: +// { dg-do compile } +// { dg-options "-std=c++11" } Yes, good point. Ack, changing to { dg-do compile { target c++11 } } and committing with the (correct :P) expectation that the change is still ok with that change. Jason, are backports ok too? Yes. Jason
Re: [PATCH] c++: Fix access checking for __is_assignable and __is_constructible (c++/94197)
On Tue, 17 Mar 2020 at 16:52, Ville Voutilainen wrote: > > On Tue, 17 Mar 2020 at 16:42, Jason Merrill wrote: > > > > On 3/17/20 9:04 AM, Jonathan Wakely wrote: > > > On 17/03/20 13:02 +, Jonathan Wakely wrote: > > >> Shouldn't the test use { dg-do compile { target c++11 } } instead of: > > >> > > >> +// { dg-do compile } > > >> +// { dg-options "-std=c++11" } > > > > Yes, good point. > > Ack, changing to { dg-do compile { target c++11 } } and committing > with the (correct :P) expectation that > the change is still ok with that change. Jason, are backports ok too?
Re: [PATCH] c++: Fix access checking for __is_assignable and __is_constructible (c++/94197)
On Tue, 17 Mar 2020 at 16:42, Jason Merrill wrote: > > On 3/17/20 9:04 AM, Jonathan Wakely wrote: > > On 17/03/20 13:02 +, Jonathan Wakely wrote: > >> Shouldn't the test use { dg-do compile { target c++11 } } instead of: > >> > >> +// { dg-do compile } > >> +// { dg-options "-std=c++11" } > > Yes, good point. Ack, changing to { dg-do compile { target c++11 } } and committing with the (correct :P) expectation that the change is still ok with that change.
Re: [PATCH] c++: Fix access checking for __is_assignable and __is_constructible (c++/94197)
On 3/17/20 9:04 AM, Jonathan Wakely wrote: On 17/03/20 13:02 +, Jonathan Wakely wrote: Shouldn't the test use { dg-do compile { target c++11 } } instead of: +// { dg-do compile } +// { dg-options "-std=c++11" } ? With that change I see: UNSUPPORTED: g++.dg/ext/pr94197.C -std=c++98 PASS: g++.dg/ext/pr94197.C -std=c++14 (test for excess errors) PASS: g++.dg/ext/pr94197.C -std=c++17 (test for excess errors) PASS: g++.dg/ext/pr94197.C -std=c++2a (test for excess errors) rather than: PASS: g++.dg/ext/pr94197.C (test for excess errors) Yes, good point. Jason
Re: [PATCH] c++: Fix access checking for __is_assignable and __is_constructible (c++/94197)
On 17/03/20 13:02 +, Jonathan Wakely wrote: Shouldn't the test use { dg-do compile { target c++11 } } instead of: +// { dg-do compile } +// { dg-options "-std=c++11" } ? With that change I see: UNSUPPORTED: g++.dg/ext/pr94197.C -std=c++98 PASS: g++.dg/ext/pr94197.C -std=c++14 (test for excess errors) PASS: g++.dg/ext/pr94197.C -std=c++17 (test for excess errors) PASS: g++.dg/ext/pr94197.C -std=c++2a (test for excess errors) rather than: PASS: g++.dg/ext/pr94197.C (test for excess errors)
Re: [PATCH] c++: Fix access checking for __is_assignable and __is_constructible (c++/94197)
Shouldn't the test use { dg-do compile { target c++11 } } instead of: +// { dg-do compile } +// { dg-options "-std=c++11" } ?
Re: [PATCH] c++: Fix access checking for __is_assignable and __is_constructible (c++/94197)
On 3/16/20 6:13 PM, Ville Voutilainen wrote: On Mon, 16 Mar 2020 at 23:25, Ville Voutilainen wrote: Tested on Linux-PPC64. This ain't no regression. But it seems to hamper attempts to fix library regressions (see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94033). It occurred to me that this can be done in one place. 2020-03-17 Ville Voutilainen gcc/ PR c++/94197 * cp/method.c (assignable_expr): Use cp_unevaluated. (is_xible_helper): Push a non-deferred access check for the stub objects created by assignable_expr and constructible_expr. It occurs to me that we don't need to defer access control in instantiate_template_1 if we're instantiating an alias template, but this patch is OK as is. Jason
Re: [PATCH] c++: Fix access checking for __is_assignable and __is_constructible (c++/94197)
On Mon, 16 Mar 2020 at 23:25, Ville Voutilainen wrote: > > Tested on Linux-PPC64. > > This ain't no regression. But it seems to hamper attempts to fix library > regressions (see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94033). It occurred to me that this can be done in one place. 2020-03-17 Ville Voutilainen gcc/ PR c++/94197 * cp/method.c (assignable_expr): Use cp_unevaluated. (is_xible_helper): Push a non-deferred access check for the stub objects created by assignable_expr and constructible_expr. testsuite/ PR c++/94197 * g++.dg/ext/pr94197.C: New. diff --git a/gcc/cp/method.c b/gcc/cp/method.c index 790d5704092..3427750f64e 100644 --- a/gcc/cp/method.c +++ b/gcc/cp/method.c @@ -1739,11 +1739,10 @@ check_nontriv (tree *tp, int *, void *) static tree assignable_expr (tree to, tree from) { - ++cp_unevaluated_operand; + cp_unevaluated cp_uneval_guard; to = build_stub_object (to); from = build_stub_object (from); tree r = cp_build_modify_expr (input_location, to, NOP_EXPR, from, tf_none); - --cp_unevaluated_operand; return r; } @@ -1806,6 +1805,7 @@ constructible_expr (tree to, tree from) static tree is_xible_helper (enum tree_code code, tree to, tree from, bool trivial) { + deferring_access_check_sentinel acs (dk_no_deferred); if (VOID_TYPE_P (to) || ABSTRACT_CLASS_TYPE_P (to) || (from && FUNC_OR_METHOD_TYPE_P (from) && (TYPE_READONLY (from) || FUNCTION_REF_QUALIFIED (from diff --git a/gcc/testsuite/g++.dg/ext/pr94197.C b/gcc/testsuite/g++.dg/ext/pr94197.C new file mode 100644 index 000..768bfbac0a6 --- /dev/null +++ b/gcc/testsuite/g++.dg/ext/pr94197.C @@ -0,0 +1,75 @@ +// { dg-do compile } +// { dg-options "-std=c++11" } + +template + T&& declval() noexcept; + +template +struct bool_constant +{ + static constexpr bool value = B; + using type = bool_constant; +}; + +using true_type = bool_constant; +using false_type = bool_constant; + +template + struct __is_nt_constructible_impl + : public false_type + { }; + +template + struct __is_nt_constructible_impl + : public bool_constant(declval()))> + { }; + +template + using __is_nothrow_constructible_impl += __is_nt_constructible_impl<__is_constructible(T, Arg), T, Arg>; + +template + struct __is_nothrow_copy_constructible_impl + : public __is_nothrow_constructible_impl + { }; + +template + struct is_nothrow_copy_constructible + : public __is_nothrow_copy_constructible_impl::type + { }; + +template + struct __is_nt_assignable_impl + : public false_type + { }; + +template + struct __is_nt_assignable_impl + : public bool_constant() = declval())> + { }; + +template + using __is_nothrow_assignable_impl += __is_nt_assignable_impl<__is_assignable(T, Arg), T, Arg>; + +template + struct __is_nothrow_copy_assignable_impl + : public __is_nothrow_assignable_impl + { }; + +template + struct is_nothrow_copy_assignable + : public __is_nothrow_copy_assignable_impl::type + { }; + +struct NType +{ + NType(); +private: + NType(const NType&); + NType& operator=(const NType&); +}; + + +static_assert( !is_nothrow_copy_constructible::value, "" ); +static_assert( !is_nothrow_copy_assignable::value, "" );
[PATCH] c++: Fix access checking for __is_assignable and __is_constructible (c++/94197)
Tested on Linux-PPC64. This ain't no regression. But it seems to hamper attempts to fix library regressions (see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94033). 2020-03-16 Ville Voutilainen gcc/ PR c++/94197 * cp/method.c (assignable_expr, constructible_expr): Push a deferred access check for the stub object. testsuite/ PR c++/94197 * g++.dg/ext/pr94197.C: New. diff --git a/gcc/cp/method.c b/gcc/cp/method.c index 790d5704092..b429ea48049 100644 --- a/gcc/cp/method.c +++ b/gcc/cp/method.c @@ -1739,11 +1739,11 @@ check_nontriv (tree *tp, int *, void *) static tree assignable_expr (tree to, tree from) { - ++cp_unevaluated_operand; + cp_unevaluated cp_uneval_guard; + deferring_access_check_sentinel acs (dk_no_deferred); to = build_stub_object (to); from = build_stub_object (from); tree r = cp_build_modify_expr (input_location, to, NOP_EXPR, from, tf_none); - --cp_unevaluated_operand; return r; } @@ -1759,6 +1759,7 @@ constructible_expr (tree to, tree from) { tree expr; cp_unevaluated cp_uneval_guard; + deferring_access_check_sentinel acs (dk_no_deferred); if (CLASS_TYPE_P (to)) { tree ctype = to; diff --git a/gcc/testsuite/g++.dg/ext/pr94197.C b/gcc/testsuite/g++.dg/ext/pr94197.C new file mode 100644 index 000..768bfbac0a6 --- /dev/null +++ b/gcc/testsuite/g++.dg/ext/pr94197.C @@ -0,0 +1,75 @@ +// { dg-do compile } +// { dg-options "-std=c++11" } + +template + T&& declval() noexcept; + +template +struct bool_constant +{ + static constexpr bool value = B; + using type = bool_constant; +}; + +using true_type = bool_constant; +using false_type = bool_constant; + +template + struct __is_nt_constructible_impl + : public false_type + { }; + +template + struct __is_nt_constructible_impl + : public bool_constant(declval()))> + { }; + +template + using __is_nothrow_constructible_impl += __is_nt_constructible_impl<__is_constructible(T, Arg), T, Arg>; + +template + struct __is_nothrow_copy_constructible_impl + : public __is_nothrow_constructible_impl + { }; + +template + struct is_nothrow_copy_constructible + : public __is_nothrow_copy_constructible_impl::type + { }; + +template + struct __is_nt_assignable_impl + : public false_type + { }; + +template + struct __is_nt_assignable_impl + : public bool_constant() = declval())> + { }; + +template + using __is_nothrow_assignable_impl += __is_nt_assignable_impl<__is_assignable(T, Arg), T, Arg>; + +template + struct __is_nothrow_copy_assignable_impl + : public __is_nothrow_assignable_impl + { }; + +template + struct is_nothrow_copy_assignable + : public __is_nothrow_copy_assignable_impl::type + { }; + +struct NType +{ + NType(); +private: + NType(const NType&); + NType& operator=(const NType&); +}; + + +static_assert( !is_nothrow_copy_constructible::value, "" ); +static_assert( !is_nothrow_copy_assignable::value, "" );