Re: [PATCH] c++: equivalence of non-dependent calls [PR107461]

[Jason Merrill <jason@redhat.com>]

On 2/5/23 09:57, Patrick Palka wrote:
> On Sat, 4 Feb 2023, Jason Merrill wrote:
> 
>> On 2/4/23 20:41, Jason Merrill wrote:
>>> On 2/4/23 20:08, Patrick Palka wrote:
>>>> On Sat, 4 Feb 2023, Jason Merrill wrote:
>>>>
>>>>> On 2/4/23 15:31, Patrick Palka wrote:
>>>>>> After r13-5684-g59e0376f607805 the (pruned) callee of a non-dependent
>>>>>> CALL_EXPR is a bare FUNCTION_DECL rather than ADDR_EXPR of
>>>>>> FUNCTION_DECL.
>>>>>> This innocent change revealed that cp_tree_equal doesn't first check
>>>>>> dependentness of a CALL_EXPR before treating the callee as a dependent
>>>>>> name, which manifests as us incorrectly accepting the first two
>>>>>> testcases below and rejecting the third:
>>>>>>
>>>>>>     * In the first testcase, cp_tree_equal incorrectly returns true for
>>>>>>       the two non-dependent CALL_EXPRs f(0) and f(0) (whose
>>>>>> CALL_EXPR_FN
>>>>>>       are different FUNCTION_DECLs) and so we treat #2 as a
>>>>>> redeclaration
>>>>>>       of #1.
>>>>>>
>>>>>>     * Same issue in the second testcase, for f<int*>() and f<char>().
>>>>>>
>>>>>>     * In the third testcase, cp_tree_equal incorrectly returns true for
>>>>>>       f<int>() and f<void(*)(int)>() which causes us to conflate the
>>>>>> two
>>>>>>       dependent specializations A<decltype(f<int>()(U()))> and
>>>>>>       A<decltype(f<void(*)(int)>()(U()))>, leading to a bogus error.
>>>>>>
>>>>>> This patch fixes this by making called_fns_equal treat two callees as
>>>>>> dependent names only if the CALL_EXPRs in question are dependent.
>>>>>>
>>>>>> Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK
>>>>>> for
>>>>>> trunk/12?  Patch generated with -w to ignore noisy whitespace changes.
>>>>>>
>>>>>>      PR c++/107461
>>>>>>
>>>>>> gcc/cp/ChangeLog:
>>>>>>
>>>>>>      * pt.cc (iterative_hash_template_arg) <case CALL_EXPR>: Treat
>>>>>>      the callee as a dependent name only if the CALL_EXPR is
>>>>>>      dependent.
>>>>>>      * tree.cc (called_fns_equal): Take two CALL_EXPRs instead of
>>>>>>      CALL_EXPR_FNs thereof.  As above.
>>>>>>      (cp_tree_equal) <case CALL_EXPR>: Adjust call to called_fns_equal.
>>>>>>
>>>>>> gcc/testsuite/ChangeLog:
>>>>>>
>>>>>>      * g++.dg/cpp0x/overload5.C: New test.
>>>>>>      * g++.dg/cpp0x/overload5a.C: New test.
>>>>>>      * g++.dg/cpp0x/overload6.C: New test.
>>>>>> ---
>>>>>>     gcc/cp/pt.cc                            |  1 +
>>>>>>     gcc/cp/tree.cc                          | 33
>>>>>> ++++++++++++++-----------
>>>>>>     gcc/testsuite/g++.dg/cpp0x/overload5.C  | 12 +++++++++
>>>>>>     gcc/testsuite/g++.dg/cpp0x/overload5a.C | 10 ++++++++
>>>>>>     gcc/testsuite/g++.dg/cpp0x/overload6.C  | 16 ++++++++++++
>>>>>>     5 files changed, 58 insertions(+), 14 deletions(-)
>>>>>>     create mode 100644 gcc/testsuite/g++.dg/cpp0x/overload5.C
>>>>>>     create mode 100644 gcc/testsuite/g++.dg/cpp0x/overload5a.C
>>>>>>     create mode 100644 gcc/testsuite/g++.dg/cpp0x/overload6.C
>>>>>>
>>>>>> diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
>>>>>> index 255332dc0c1..c9360240cd2 100644
>>>>>> --- a/gcc/cp/pt.cc
>>>>>> +++ b/gcc/cp/pt.cc
>>>>>> @@ -1841,6 +1841,7 @@ iterative_hash_template_arg (tree arg, hashval_t
>>>>>> val)
>>>>>>         case CALL_EXPR:
>>>>>>           {
>>>>>>         tree fn = CALL_EXPR_FN (arg);
>>>>>> +    if (TREE_TYPE (arg) == NULL_TREE)
>>>>>
>>>>> How about changing dependent_name to take the CALL_EXPR rather than the
>>>>> CALL_EXPR_FN?  That would mean some changes to write_expression to move
>>>>> the
>>>>> dependent_name handling into the CALL_EXPR handling, but that doesn't
>>>>> seem
>>>>> like a bad thing.  Other callers seem like a trivial change.
>>>>
>>>> Indeed changing dependent_name seems best, but I'm worried about such a
>>>> refactoring to write_expression causing unintended mangling changes at
>>>> this stage.  Because it seems the CALL_EXPR case of write_expression
>>>> isn't the user of the dependent_name branch of write_expression, at
>>>> least according to the following patch which causes us to ICE on
>>>> mangle{37,57,58,76}.C:
>>>
>>> Yeah, I tried the same thing.  Maybe for GCC 13 better to add a new function
>>> rather than change the current one.
> 
> Sounds good, like so?  Only regtested so far.  Full bootstrap and
> regtest running on x86_64-pc-linux-gnu.
> 
> -- >8 --
> 
> Subject: [PATCH] c++: equivalence of non-dependent calls [PR107461]
> 
> After r13-5684-g59e0376f607805 the (pruned) callee of a non-dependent
> CALL_EXPR is a bare FUNCTION_DECL rather than ADDR_EXPR of FUNCTION_DECL.
> This innocent change revealed that cp_tree_equal doesn't first check
> dependentness of a CALL_EXPR before treating a FUNCTION_DECL callee as a
> dependent name, which manifests as us incorrectly accepting the first
> two testcases below and rejecting the third:
> 
>   * In the first testcase, cp_tree_equal incorrectly returns true for
>     the two non-dependent CALL_EXPRs f(0) and f(0) (whose CALL_EXPR_FN
>     are different FUNCTION_DECLs) and so we treat #2 as a redeclaration
>     of #1.
> 
>   * Same issue in the second testcase, for f<int*>() and f<char>().
> 
>   * In the third testcase, cp_tree_equal incorrectly returns true for
>     f<int>() and f<void(*)(int)>() which causes us to conflate the two
>     dependent specializations A<decltype(f<int>()(U()))> and
>     A<decltype(f<void(*)(int)>()(U()))>, leading to a bogus error.
> 
> This patch fixes this by making called_fns_equal treat two callees as
> dependent names only if the overall CALL_EXPRs are dependent, via a new
> convenience function call_expr_dependent_name that is like dependent_name
> but also checks dependence of the overall CALL_EXPR.
> 
> Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for
> trunk/12?  Patch generated with -w to ignore noisy whitespace changes.

OK, thanks.

> 	PR c++/107461
> 
> gcc/cp/ChangeLog:
> 
> 	* cp-tree.h (call_expr_dependent_name): Declare.
> 	* pt.cc (iterative_hash_template_arg) <case CALL_EXPR>: Use
> 	call_expr_dependent_name instead of dependent_name.
> 	* tree.cc (call_expr_dependent_name): Define.
> 	(called_fns_equal): Adjust to take two CALL_EXPRs instead of
> 	CALL_EXPR_FNs thereof.  Use call_expr_dependent_name instead
> 	of dependent_name.
> 	(cp_tree_equal) <case CALL_EXPR>: Adjust call to called_fns_equal.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* g++.dg/cpp0x/overload5.C: New test.
> 	* g++.dg/cpp0x/overload5a.C: New test.
> 	* g++.dg/cpp0x/overload6.C: New test.
> ---
>   gcc/cp/cp-tree.h                        |  1 +
>   gcc/cp/pt.cc                            |  2 +-
>   gcc/cp/tree.cc                          | 24 +++++++++++++++++++-----
>   gcc/testsuite/g++.dg/cpp0x/overload5.C  | 12 ++++++++++++
>   gcc/testsuite/g++.dg/cpp0x/overload5a.C | 10 ++++++++++
>   gcc/testsuite/g++.dg/cpp0x/overload6.C  | 16 ++++++++++++++++
>   6 files changed, 59 insertions(+), 6 deletions(-)
>   create mode 100644 gcc/testsuite/g++.dg/cpp0x/overload5.C
>   create mode 100644 gcc/testsuite/g++.dg/cpp0x/overload5a.C
>   create mode 100644 gcc/testsuite/g++.dg/cpp0x/overload6.C
> 
> diff --git a/gcc/cp/cp-tree.h b/gcc/cp/cp-tree.h
> index 00b2bffc85c..ef601182d4b 100644
> --- a/gcc/cp/cp-tree.h
> +++ b/gcc/cp/cp-tree.h
> @@ -7902,6 +7902,7 @@ extern tree lookup_maybe_add			(tree fns, tree lookup,
>   extern int is_overloaded_fn			(tree) ATTRIBUTE_PURE;
>   extern bool really_overloaded_fn		(tree) ATTRIBUTE_PURE;
>   extern tree dependent_name			(tree);
> +extern tree call_expr_dependent_name		(tree);
>   extern tree maybe_get_fns			(tree) ATTRIBUTE_PURE;
>   extern tree get_fns				(tree) ATTRIBUTE_PURE;
>   extern tree get_first_fn			(tree) ATTRIBUTE_PURE;
> diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
> index 255332dc0c1..9f3fc1fa089 100644
> --- a/gcc/cp/pt.cc
> +++ b/gcc/cp/pt.cc
> @@ -1841,7 +1841,7 @@ iterative_hash_template_arg (tree arg, hashval_t val)
>       case CALL_EXPR:
>         {
>   	tree fn = CALL_EXPR_FN (arg);
> -	if (tree name = dependent_name (fn))
> +	if (tree name = call_expr_dependent_name (arg))
>   	  {
>   	    if (TREE_CODE (fn) == TEMPLATE_ID_EXPR)
>   	      val = iterative_hash_template_arg (TREE_OPERAND (fn, 1), val);
> diff --git a/gcc/cp/tree.cc b/gcc/cp/tree.cc
> index c1da868732b..880bd4f9bcf 100644
> --- a/gcc/cp/tree.cc
> +++ b/gcc/cp/tree.cc
> @@ -2608,6 +2608,18 @@ dependent_name (tree x)
>     return NULL_TREE;
>   }
>   
> +/* Like dependent_name, but takes the overall CALL_EXPR and checks its
> +   dependence.  */
> +
> +tree
> +call_expr_dependent_name (tree x)
> +{
> +  if (TREE_TYPE (x) != NULL_TREE)
> +    /* X isn't dependent, so its callee isn't a dependent name.  */
> +    return NULL_TREE;
> +  return dependent_name (CALL_EXPR_FN (x));
> +}
> +
>   /* Returns true iff X is an expression for an overloaded function
>      whose type cannot be known without performing overload
>      resolution.  */
> @@ -3870,16 +3882,18 @@ decl_internal_context_p (const_tree decl)
>     return !TREE_PUBLIC (decl);
>   }
>   
> -/* Subroutine of cp_tree_equal: t1 and t2 are the CALL_EXPR_FNs of two
> -   CALL_EXPRS.  Return whether they are equivalent.  */
> +/* Subroutine of cp_tree_equal: t1 and t2 are two CALL_EXPRs.
> +   Return whether their CALL_EXPR_FNs are equivalent.  */
>   
>   static bool
>   called_fns_equal (tree t1, tree t2)
>   {
>     /* Core 1321: dependent names are equivalent even if the overload sets
>        are different.  But do compare explicit template arguments.  */
> -  tree name1 = dependent_name (t1);
> -  tree name2 = dependent_name (t2);
> +  tree name1 = call_expr_dependent_name (t1);
> +  tree name2 = call_expr_dependent_name (t2);
> +  t1 = CALL_EXPR_FN (t1);
> +  t2 = CALL_EXPR_FN (t2);
>     if (name1 || name2)
>       {
>         tree targs1 = NULL_TREE, targs2 = NULL_TREE;
> @@ -4037,7 +4051,7 @@ cp_tree_equal (tree t1, tree t2)
>   	if (KOENIG_LOOKUP_P (t1) != KOENIG_LOOKUP_P (t2))
>   	  return false;
>   
> -	if (!called_fns_equal (CALL_EXPR_FN (t1), CALL_EXPR_FN (t2)))
> +	if (!called_fns_equal (t1, t2))
>   	  return false;
>   
>   	call_expr_arg_iterator iter1, iter2;
> diff --git a/gcc/testsuite/g++.dg/cpp0x/overload5.C b/gcc/testsuite/g++.dg/cpp0x/overload5.C
> new file mode 100644
> index 00000000000..e05b1594f51
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp0x/overload5.C
> @@ -0,0 +1,12 @@
> +// PR c++/107461
> +// { dg-do compile { target c++11 } }
> +
> +int f(...);
> +template<class T> decltype(T() + f(0)) g(); // #1
> +
> +char f(int);
> +template<class T> decltype(T() + f(0)) g(); // #2, distinct from #1
> +
> +int main() {
> +  g<int>(); // { dg-error "ambiguous" }
> +}
> diff --git a/gcc/testsuite/g++.dg/cpp0x/overload5a.C b/gcc/testsuite/g++.dg/cpp0x/overload5a.C
> new file mode 100644
> index 00000000000..037114f199c
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp0x/overload5a.C
> @@ -0,0 +1,10 @@
> +// PR c++/107461
> +// { dg-do compile { target c++11 } }
> +
> +template<class T> T f();
> +template<class T> decltype(T() + f<int*>()) g(); // #1
> +template<class T> decltype(T() + f<char>()) g(); // #2, distinct from #1
> +
> +int main() {
> +  g<int>(); // { dg-error "ambiguous" }
> +}
> diff --git a/gcc/testsuite/g++.dg/cpp0x/overload6.C b/gcc/testsuite/g++.dg/cpp0x/overload6.C
> new file mode 100644
> index 00000000000..1fbee0501de
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp0x/overload6.C
> @@ -0,0 +1,16 @@
> +// PR c++/107461
> +// { dg-do compile { target c++11 } }
> +
> +template<class T> T f();
> +
> +template<class> struct A { };
> +
> +template<class T> struct B {
> +  template<class U, class = A<decltype(f<T>()(U()))>>
> +  static void g(U);
> +};
> +
> +int main() {
> +  B<int> b;
> +  B<void(*)(int)>::g(0); // { dg-bogus "no match" }
> +}