Re: [PATCH] c++: unifying FUNCTION_DECLs [PR93740]

2023-12-12 Thread Jason Merrill

On 12/12/23 13:40, Patrick Palka wrote:

Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look
OK for trunk?


OK.


I considered removing the is_overloaded_fn test now as
well, but it could in theory be hit (and not subsumed by the
type_unknown_p test) for e.g. OVERLOAD of a single FUNCTION_DECL.  I
wonder if that's something we'd see here?  If not, I can remove the
test.  It seems safe to remove as far as the testsuite is concerned.


Next stage 1, sure.


-- >8 --

unify currently always returns success when unifying two FUNCTION_DECLs
(due to the is_overloaded_fn deferment within the default case), which
means for the below testcase unify incorrectly matches ::foo with
::bar, which leads to deduction failure for the index_of calls due to
a bogus base class ambiguity.

This patch makes us instead handle unification of FUNCTION_DECL like
other decls, i.e. according to their identity.

PR c++/93740

gcc/cp/ChangeLog:

* pt.cc (unify) : Handle it like FIELD_DECL
and TEMPLATE_DECL.

gcc/testsuite/ChangeLog:

* g++.dg/template/ptrmem34.C: New test.
---
  gcc/cp/pt.cc |  1 +
  gcc/testsuite/g++.dg/template/ptrmem34.C | 27 
  2 files changed, 28 insertions(+)
  create mode 100644 gcc/testsuite/g++.dg/template/ptrmem34.C

diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
index c2ddbff405b..a8966e223f1 100644
--- a/gcc/cp/pt.cc
+++ b/gcc/cp/pt.cc
@@ -24967,6 +24967,7 @@ unify (tree tparms, tree targs, tree parm, tree arg, 
int strict,
gcc_unreachable ();
  
  case FIELD_DECL:

+case FUNCTION_DECL:
  case TEMPLATE_DECL:
/* Matched cases are handled by the ARG == PARM test above.  */
return unify_template_argument_mismatch (explain_p, parm, arg);
diff --git a/gcc/testsuite/g++.dg/template/ptrmem34.C 
b/gcc/testsuite/g++.dg/template/ptrmem34.C
new file mode 100644
index 000..c349ca55639
--- /dev/null
+++ b/gcc/testsuite/g++.dg/template/ptrmem34.C
@@ -0,0 +1,27 @@
+// PR c++/93740
+// { dg-do compile { target c++11 } }
+
+struct A {
+  void foo();
+  void bar();
+};
+
+template 
+struct const_val{};
+
+template 
+struct indexed_elem{};
+
+using mem_fun_A_foo = const_val;
+using mem_fun_A_bar = const_val;
+
+struct A_indexed_member_funcs
+  : indexed_elem<0, mem_fun_A_foo>,
+indexed_elem<1, mem_fun_A_bar>
+{};
+
+template 
+constexpr int index_of(indexed_elem) { return N; }
+
+static_assert(index_of(A_indexed_member_funcs{}) == 0, "");
+static_assert(index_of(A_indexed_member_funcs{}) == 1, "");




[PATCH] c++: unifying FUNCTION_DECLs [PR93740]

2023-12-12 Thread Patrick Palka
Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look
OK for trunk?  I considered removing the is_overloaded_fn test now as
well, but it could in theory be hit (and not subsumed by the
type_unknown_p test) for e.g. OVERLOAD of a single FUNCTION_DECL.  I
wonder if that's something we'd see here?  If not, I can remove the
test.  It seems safe to remove as far as the testsuite is concerned.

-- >8 --

unify currently always returns success when unifying two FUNCTION_DECLs
(due to the is_overloaded_fn deferment within the default case), which
means for the below testcase unify incorrectly matches ::foo with
::bar, which leads to deduction failure for the index_of calls due to
a bogus base class ambiguity.

This patch makes us instead handle unification of FUNCTION_DECL like
other decls, i.e. according to their identity.

PR c++/93740

gcc/cp/ChangeLog:

* pt.cc (unify) : Handle it like FIELD_DECL
and TEMPLATE_DECL.

gcc/testsuite/ChangeLog:

* g++.dg/template/ptrmem34.C: New test.
---
 gcc/cp/pt.cc |  1 +
 gcc/testsuite/g++.dg/template/ptrmem34.C | 27 
 2 files changed, 28 insertions(+)
 create mode 100644 gcc/testsuite/g++.dg/template/ptrmem34.C

diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
index c2ddbff405b..a8966e223f1 100644
--- a/gcc/cp/pt.cc
+++ b/gcc/cp/pt.cc
@@ -24967,6 +24967,7 @@ unify (tree tparms, tree targs, tree parm, tree arg, 
int strict,
   gcc_unreachable ();
 
 case FIELD_DECL:
+case FUNCTION_DECL:
 case TEMPLATE_DECL:
   /* Matched cases are handled by the ARG == PARM test above.  */
   return unify_template_argument_mismatch (explain_p, parm, arg);
diff --git a/gcc/testsuite/g++.dg/template/ptrmem34.C 
b/gcc/testsuite/g++.dg/template/ptrmem34.C
new file mode 100644
index 000..c349ca55639
--- /dev/null
+++ b/gcc/testsuite/g++.dg/template/ptrmem34.C
@@ -0,0 +1,27 @@
+// PR c++/93740
+// { dg-do compile { target c++11 } }
+
+struct A {
+  void foo();
+  void bar();
+};
+
+template 
+struct const_val{};
+
+template 
+struct indexed_elem{};
+
+using mem_fun_A_foo = const_val;
+using mem_fun_A_bar = const_val;
+
+struct A_indexed_member_funcs
+  : indexed_elem<0, mem_fun_A_foo>,
+indexed_elem<1, mem_fun_A_bar>
+{};
+
+template 
+constexpr int index_of(indexed_elem) { return N; }
+
+static_assert(index_of(A_indexed_member_funcs{}) == 0, "");
+static_assert(index_of(A_indexed_member_funcs{}) == 1, "");
-- 
2.43.0.76.g1a87c842ec