c++: unsynthesized defaulted constexpr fn [PR110122]
Checks
Commit Message
Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for
trunk?
-- >8 --
In the second testcase of PR110122, during regeneration of the generic
lambda with V=Bar{}, substitution followed by coerce_template_parms for
A<V>'s template argument naturally yields a copy of V in terms of Bar's
(implicitly) defaulted copy constructor.
This however happens inside a template context so although we introduced
a use of the copy constructor, mark_used didn't actually synthesize it,
which causes subsequent constant evaluation of the template argument to
fail with:
nontype-class58.C: In instantiation of ‘void f() [with Bar V = Bar{Foo()}]’:
nontype-class58.C:22:11: required from here
nontype-class58.C:18:18: error: ‘constexpr Bar::Bar(const Bar&)’ used before its definition
Conveniently we already make sure to instantiate eligible constexpr
functions before such (manifestly) constant evaluation, as per P0859R0.
So this patch fixes this by making sure to synthesize eligible defaulted
constexpr functions beforehand as well.
PR c++/110122
gcc/cp/ChangeLog:
* constexpr.cc (instantiate_cx_fn_r): Also synthesize defaulted
functions.
gcc/testsuite/ChangeLog:
* g++.dg/cpp2a/nontype-class58.C: New test.
---
gcc/cp/constexpr.cc | 7 ++++--
gcc/testsuite/g++.dg/cpp2a/nontype-class58.C | 23 ++++++++++++++++++++
2 files changed, 28 insertions(+), 2 deletions(-)
create mode 100644 gcc/testsuite/g++.dg/cpp2a/nontype-class58.C
Comments
On 6/6/23 14:29, Patrick Palka wrote:
> Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for
> trunk?
>
> -- >8 --
>
> In the second testcase of PR110122, during regeneration of the generic
> lambda with V=Bar{}, substitution followed by coerce_template_parms for
> A<V>'s template argument naturally yields a copy of V in terms of Bar's
> (implicitly) defaulted copy constructor.
>
> This however happens inside a template context so although we introduced
> a use of the copy constructor, mark_used didn't actually synthesize it,
> which causes subsequent constant evaluation of the template argument to
> fail with:
>
> nontype-class58.C: In instantiation of ‘void f() [with Bar V = Bar{Foo()}]’:
> nontype-class58.C:22:11: required from here
> nontype-class58.C:18:18: error: ‘constexpr Bar::Bar(const Bar&)’ used before its definition
>
> Conveniently we already make sure to instantiate eligible constexpr
> functions before such (manifestly) constant evaluation, as per P0859R0.
> So this patch fixes this by making sure to synthesize eligible defaulted
> constexpr functions beforehand as well.
We probably also want to do this in cxx_eval_call_expression, under
> /* We can't defer instantiating the function any longer. */
Jason
On Wed, 7 Jun 2023, Jason Merrill wrote:
> On 6/6/23 14:29, Patrick Palka wrote:
> > Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for
> > trunk?
> >
> > -- >8 --
> >
> > In the second testcase of PR110122, during regeneration of the generic
> > lambda with V=Bar{}, substitution followed by coerce_template_parms for
> > A<V>'s template argument naturally yields a copy of V in terms of Bar's
> > (implicitly) defaulted copy constructor.
> >
> > This however happens inside a template context so although we introduced
> > a use of the copy constructor, mark_used didn't actually synthesize it,
> > which causes subsequent constant evaluation of the template argument to
> > fail with:
> >
> > nontype-class58.C: In instantiation of ‘void f() [with Bar V =
> > Bar{Foo()}]’:
> > nontype-class58.C:22:11: required from here
> > nontype-class58.C:18:18: error: ‘constexpr Bar::Bar(const Bar&)’ used
> > before its definition
> >
> > Conveniently we already make sure to instantiate eligible constexpr
> > functions before such (manifestly) constant evaluation, as per P0859R0.
> > So this patch fixes this by making sure to synthesize eligible defaulted
> > constexpr functions beforehand as well.
>
> We probably also want to do this in cxx_eval_call_expression, under
Makes sense, like so? I'm not sure if it's possible to write a test
for which this code path makes an observable difference, but I verified
the code path is hit a couple of times throughout the testsuite (mainly
from fold_non_dependent_expr called from build_non_dependent_expr).
Bootstrapped and regtested on x86_64-pc-linux-gnu.
-->8 --
PR c++/110122
gcc/cp/ChangeLog:
* constexpr.cc (cxx_eval_call_expression): Also synthesize
eligible defaulted functions.
(instantiate_cx_fn_r): Likewise.
gcc/testsuite/ChangeLog:
* g++.dg/cpp2a/nontype-class58.C: New test.
---
gcc/cp/constexpr.cc | 14 ++++++++----
gcc/testsuite/g++.dg/cpp2a/nontype-class58.C | 23 ++++++++++++++++++++
2 files changed, 33 insertions(+), 4 deletions(-)
create mode 100644 gcc/testsuite/g++.dg/cpp2a/nontype-class58.C
diff --git a/gcc/cp/constexpr.cc b/gcc/cp/constexpr.cc
index 8f7f0b7d325..9122a5efa65 100644
--- a/gcc/cp/constexpr.cc
+++ b/gcc/cp/constexpr.cc
@@ -2897,7 +2897,7 @@ cxx_eval_call_expression (const constexpr_ctx *ctx, tree t,
/* We can't defer instantiating the function any longer. */
if (!DECL_INITIAL (fun)
- && DECL_TEMPLOID_INSTANTIATION (fun)
+ && (DECL_TEMPLOID_INSTANTIATION (fun) || DECL_DEFAULTED_FN (fun))
&& !uid_sensitive_constexpr_evaluation_p ())
{
location_t save_loc = input_location;
@@ -2905,7 +2905,10 @@ cxx_eval_call_expression (const constexpr_ctx *ctx, tree t,
++function_depth;
if (ctx->manifestly_const_eval == mce_true)
FNDECL_MANIFESTLY_CONST_EVALUATED (fun) = true;
- instantiate_decl (fun, /*defer_ok*/false, /*expl_inst*/false);
+ if (DECL_TEMPLOID_INSTANTIATION (fun))
+ instantiate_decl (fun, /*defer_ok*/false, /*expl_inst*/false);
+ else
+ synthesize_method (fun);
--function_depth;
input_location = save_loc;
}
@@ -8110,11 +8113,14 @@ instantiate_cx_fn_r (tree *tp, int *walk_subtrees, void */*data*/)
&& DECL_DECLARED_CONSTEXPR_P (*tp)
&& !DECL_INITIAL (*tp)
&& !trivial_fn_p (*tp)
- && DECL_TEMPLOID_INSTANTIATION (*tp)
+ && (DECL_TEMPLOID_INSTANTIATION (*tp) || DECL_DEFAULTED_FN (*tp))
&& !uid_sensitive_constexpr_evaluation_p ())
{
++function_depth;
- instantiate_decl (*tp, /*defer_ok*/false, /*expl_inst*/false);
+ if (DECL_TEMPLOID_INSTANTIATION (*tp))
+ instantiate_decl (*tp, /*defer_ok*/false, /*expl_inst*/false);
+ else
+ synthesize_method (*tp);
--function_depth;
}
else if (TREE_CODE (*tp) == CALL_EXPR
diff --git a/gcc/testsuite/g++.dg/cpp2a/nontype-class58.C b/gcc/testsuite/g++.dg/cpp2a/nontype-class58.C
new file mode 100644
index 00000000000..6e40698da2f
--- /dev/null
+++ b/gcc/testsuite/g++.dg/cpp2a/nontype-class58.C
@@ -0,0 +1,23 @@
+// PR c++/110122
+// { dg-do compile { target c++20 } }
+
+struct Foo {
+ Foo() = default;
+ constexpr Foo(const Foo&) { }
+};
+
+struct Bar {
+ Foo _;
+};
+
+template<Bar V>
+struct A { };
+
+template<Bar V>
+void f() {
+ [](auto){ A<V> d; }(0); // { dg-bogus "used before its definition" }
+};
+
+int main() {
+ f<Bar{}>();
+}
On 6/8/23 15:54, Patrick Palka wrote:
> On Wed, 7 Jun 2023, Jason Merrill wrote:
>
>> On 6/6/23 14:29, Patrick Palka wrote:
>>> Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for
>>> trunk?
>>>
>>> -- >8 --
>>>
>>> In the second testcase of PR110122, during regeneration of the generic
>>> lambda with V=Bar{}, substitution followed by coerce_template_parms for
>>> A<V>'s template argument naturally yields a copy of V in terms of Bar's
>>> (implicitly) defaulted copy constructor.
>>>
>>> This however happens inside a template context so although we introduced
>>> a use of the copy constructor, mark_used didn't actually synthesize it,
>>> which causes subsequent constant evaluation of the template argument to
>>> fail with:
>>>
>>> nontype-class58.C: In instantiation of ‘void f() [with Bar V =
>>> Bar{Foo()}]’:
>>> nontype-class58.C:22:11: required from here
>>> nontype-class58.C:18:18: error: ‘constexpr Bar::Bar(const Bar&)’ used
>>> before its definition
>>>
>>> Conveniently we already make sure to instantiate eligible constexpr
>>> functions before such (manifestly) constant evaluation, as per P0859R0.
>>> So this patch fixes this by making sure to synthesize eligible defaulted
>>> constexpr functions beforehand as well.
>>
>> We probably also want to do this in cxx_eval_call_expression, under
>
> Makes sense, like so? I'm not sure if it's possible to write a test
> for which this code path makes an observable difference, but I verified
> the code path is hit a couple of times throughout the testsuite (mainly
> from fold_non_dependent_expr called from build_non_dependent_expr).
> Bootstrapped and regtested on x86_64-pc-linux-gnu.
OK.
> -->8 --
>
> PR c++/110122
>
> gcc/cp/ChangeLog:
>
> * constexpr.cc (cxx_eval_call_expression): Also synthesize
> eligible defaulted functions.
> (instantiate_cx_fn_r): Likewise.
>
> gcc/testsuite/ChangeLog:
>
> * g++.dg/cpp2a/nontype-class58.C: New test.
> ---
> gcc/cp/constexpr.cc | 14 ++++++++----
> gcc/testsuite/g++.dg/cpp2a/nontype-class58.C | 23 ++++++++++++++++++++
> 2 files changed, 33 insertions(+), 4 deletions(-)
> create mode 100644 gcc/testsuite/g++.dg/cpp2a/nontype-class58.C
>
> diff --git a/gcc/cp/constexpr.cc b/gcc/cp/constexpr.cc
> index 8f7f0b7d325..9122a5efa65 100644
> --- a/gcc/cp/constexpr.cc
> +++ b/gcc/cp/constexpr.cc
> @@ -2897,7 +2897,7 @@ cxx_eval_call_expression (const constexpr_ctx *ctx, tree t,
>
> /* We can't defer instantiating the function any longer. */
> if (!DECL_INITIAL (fun)
> - && DECL_TEMPLOID_INSTANTIATION (fun)
> + && (DECL_TEMPLOID_INSTANTIATION (fun) || DECL_DEFAULTED_FN (fun))
> && !uid_sensitive_constexpr_evaluation_p ())
> {
> location_t save_loc = input_location;
> @@ -2905,7 +2905,10 @@ cxx_eval_call_expression (const constexpr_ctx *ctx, tree t,
> ++function_depth;
> if (ctx->manifestly_const_eval == mce_true)
> FNDECL_MANIFESTLY_CONST_EVALUATED (fun) = true;
> - instantiate_decl (fun, /*defer_ok*/false, /*expl_inst*/false);
> + if (DECL_TEMPLOID_INSTANTIATION (fun))
> + instantiate_decl (fun, /*defer_ok*/false, /*expl_inst*/false);
> + else
> + synthesize_method (fun);
> --function_depth;
> input_location = save_loc;
> }
> @@ -8110,11 +8113,14 @@ instantiate_cx_fn_r (tree *tp, int *walk_subtrees, void */*data*/)
> && DECL_DECLARED_CONSTEXPR_P (*tp)
> && !DECL_INITIAL (*tp)
> && !trivial_fn_p (*tp)
> - && DECL_TEMPLOID_INSTANTIATION (*tp)
> + && (DECL_TEMPLOID_INSTANTIATION (*tp) || DECL_DEFAULTED_FN (*tp))
> && !uid_sensitive_constexpr_evaluation_p ())
> {
> ++function_depth;
> - instantiate_decl (*tp, /*defer_ok*/false, /*expl_inst*/false);
> + if (DECL_TEMPLOID_INSTANTIATION (*tp))
> + instantiate_decl (*tp, /*defer_ok*/false, /*expl_inst*/false);
> + else
> + synthesize_method (*tp);
> --function_depth;
> }
> else if (TREE_CODE (*tp) == CALL_EXPR
> diff --git a/gcc/testsuite/g++.dg/cpp2a/nontype-class58.C b/gcc/testsuite/g++.dg/cpp2a/nontype-class58.C
> new file mode 100644
> index 00000000000..6e40698da2f
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp2a/nontype-class58.C
> @@ -0,0 +1,23 @@
> +// PR c++/110122
> +// { dg-do compile { target c++20 } }
> +
> +struct Foo {
> + Foo() = default;
> + constexpr Foo(const Foo&) { }
> +};
> +
> +struct Bar {
> + Foo _;
> +};
> +
> +template<Bar V>
> +struct A { };
> +
> +template<Bar V>
> +void f() {
> + [](auto){ A<V> d; }(0); // { dg-bogus "used before its definition" }
> +};
> +
> +int main() {
> + f<Bar{}>();
> +}
@@ -8110,11 +8110,14 @@ instantiate_cx_fn_r (tree *tp, int *walk_subtrees, void */*data*/)
&& DECL_DECLARED_CONSTEXPR_P (*tp)
&& !DECL_INITIAL (*tp)
&& !trivial_fn_p (*tp)
- && DECL_TEMPLOID_INSTANTIATION (*tp)
+ && (DECL_TEMPLOID_INSTANTIATION (*tp) || DECL_DEFAULTED_FN (*tp))
&& !uid_sensitive_constexpr_evaluation_p ())
{
++function_depth;
- instantiate_decl (*tp, /*defer_ok*/false, /*expl_inst*/false);
+ if (DECL_TEMPLOID_INSTANTIATION (*tp))
+ instantiate_decl (*tp, /*defer_ok*/false, /*expl_inst*/false);
+ else
+ synthesize_method (*tp);
--function_depth;
}
else if (TREE_CODE (*tp) == CALL_EXPR
new file mode 100644
@@ -0,0 +1,23 @@
+// PR c++/110122
+// { dg-do compile { target c++20 } }
+
+struct Foo {
+ Foo() = default;
+ constexpr Foo(const Foo&) { }
+};
+
+struct Bar {
+ Foo _;
+};
+
+template<Bar V>
+struct A { };
+
+template<Bar V>
+void f() {
+ [](auto){ A<V> d; }(0); // { dg-bogus "used before its definition" }
+};
+
+int main() {
+ f<Bar{}>();
+}