[PATCH] c++: Fix up maybe_warn_for_constant_evaluated calls [PR114580]

[PATCH] c++: Fix up maybe_warn_for_constant_evaluated calls [PR114580]

[Jakub Jelinek]

Hi!

When looking at maybe_warn_for_constant_evaluated for the trivial
infinite loops patch, I've noticed that it can emit weird diagnostics
for if constexpr in templates, first warn that std::is_constant_evaluted()
always evaluates to false (because the function template is not constexpr)
and then during instantiation warn that std::is_constant_evaluted()
always evaluates to true (because it is used in if constexpr condition).
Now, only the latter is actually true, even when the if constexpr
is in a non-constexpr function, it will still always evaluate to true.

So, the following patch fixes it to call maybe_warn_for_constant_evaluated
always with IF_STMT_CONSTEXPR_P (if_stmt) as the second argument rather than
true if it is if constexpr with non-dependent condition etc.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2024-04-05  Jakub Jelinek  <jakub@redhat.com>

	PR c++/114580
	* semantics.cc (finish_if_stmt_cond): Call
	maybe_warn_for_constant_evaluated with IF_STMT_CONSTEXPR_P (if_stmt)
	as the second argument, rather than true/false depending on if
	it is if constexpr with non-dependent constant expression with
	bool type.

	* g++.dg/cpp2a/is-constant-evaluated15.C: New test.

--- gcc/cp/semantics.cc.jj	2024-04-03 09:58:33.407772541 +0200
+++ gcc/cp/semantics.cc	2024-04-04 12:11:36.203886572 +0200
@@ -1126,6 +1126,9 @@ tree
 finish_if_stmt_cond (tree orig_cond, tree if_stmt)
 {
   tree cond = maybe_convert_cond (orig_cond);
+  maybe_warn_for_constant_evaluated (cond,
+				     /*constexpr_if=*/
+				     IF_STMT_CONSTEXPR_P (if_stmt));
   if (IF_STMT_CONSTEXPR_P (if_stmt)
       && !type_dependent_expression_p (cond)
       && require_constant_expression (cond)
@@ -1134,16 +1137,11 @@ finish_if_stmt_cond (tree orig_cond, tre
 	 converted to bool.  */
       && TYPE_MAIN_VARIANT (TREE_TYPE (cond)) == boolean_type_node)
     {
-      maybe_warn_for_constant_evaluated (cond, /*constexpr_if=*/true);
       cond = instantiate_non_dependent_expr (cond);
       cond = cxx_constant_value (cond);
     }
-  else
-    {
-      maybe_warn_for_constant_evaluated (cond, /*constexpr_if=*/false);
-      if (processing_template_decl)
-	cond = orig_cond;
-    }
+  else if (processing_template_decl)
+    cond = orig_cond;
   finish_cond (&IF_COND (if_stmt), cond);
   add_stmt (if_stmt);
   THEN_CLAUSE (if_stmt) = push_stmt_list ();
--- gcc/testsuite/g++.dg/cpp2a/is-constant-evaluated15.C.jj	2024-04-04 12:23:36.706962932 +0200
+++ gcc/testsuite/g++.dg/cpp2a/is-constant-evaluated15.C	2024-04-04 12:22:29.915882859 +0200
@@ -0,0 +1,28 @@
+// PR c++/114580
+// { dg-do compile { target c++17 } }
+// { dg-options "-Wtautological-compare" }
+
+namespace std {
+  constexpr inline bool
+  is_constant_evaluated () noexcept
+  {
+#if __cpp_if_consteval >= 202106L
+    if consteval { return true; } else { return false; }
+#else
+    return __builtin_is_constant_evaluated ();
+#endif
+  }
+}
+
+template <typename T>
+void foo ()
+{
+  if constexpr ((T) std::is_constant_evaluated ())	// { dg-warning "'std::is_constant_evaluated' always evaluates to true in 'if constexpr'" }
+    ;							// { dg-bogus "'std::is_constant_evaluated' always evaluates to false in a non-'constexpr' function" }
+}
+
+void
+bar ()
+{
+  foo <bool> ();
+}

	Jakub

Re: [PATCH] c++: Fix up maybe_warn_for_constant_evaluated calls [PR114580]

[Marek Polacek]

On Fri, Apr 05, 2024 at 09:40:48AM +0200, Jakub Jelinek wrote:
> Hi!
> 
> When looking at maybe_warn_for_constant_evaluated for the trivial
> infinite loops patch, I've noticed that it can emit weird diagnostics
> for if constexpr in templates, first warn that std::is_constant_evaluted()
> always evaluates to false (because the function template is not constexpr)
> and then during instantiation warn that std::is_constant_evaluted()
> always evaluates to true (because it is used in if constexpr condition).
> Now, only the latter is actually true, even when the if constexpr
> is in a non-constexpr function, it will still always evaluate to true.
> 
> So, the following patch fixes it to call maybe_warn_for_constant_evaluated
> always with IF_STMT_CONSTEXPR_P (if_stmt) as the second argument rather than
> true if it is if constexpr with non-dependent condition etc.
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
> 
> 2024-04-05  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR c++/114580
> 	* semantics.cc (finish_if_stmt_cond): Call
> 	maybe_warn_for_constant_evaluated with IF_STMT_CONSTEXPR_P (if_stmt)
> 	as the second argument, rather than true/false depending on if
> 	it is if constexpr with non-dependent constant expression with
> 	bool type.
> 
> 	* g++.dg/cpp2a/is-constant-evaluated15.C: New test.
> 
> --- gcc/cp/semantics.cc.jj	2024-04-03 09:58:33.407772541 +0200
> +++ gcc/cp/semantics.cc	2024-04-04 12:11:36.203886572 +0200
> @@ -1126,6 +1126,9 @@ tree
>  finish_if_stmt_cond (tree orig_cond, tree if_stmt)
>  {
>    tree cond = maybe_convert_cond (orig_cond);
> +  maybe_warn_for_constant_evaluated (cond,
> +				     /*constexpr_if=*/
> +				     IF_STMT_CONSTEXPR_P (if_stmt));

I don't think we need the comment anymore since it's clear what the
argument does, and then the whole call can fit on a single line.

But either way, the patch looks good, thanks.

>    if (IF_STMT_CONSTEXPR_P (if_stmt)
>        && !type_dependent_expression_p (cond)
>        && require_constant_expression (cond)
> @@ -1134,16 +1137,11 @@ finish_if_stmt_cond (tree orig_cond, tre
>  	 converted to bool.  */
>        && TYPE_MAIN_VARIANT (TREE_TYPE (cond)) == boolean_type_node)
>      {
> -      maybe_warn_for_constant_evaluated (cond, /*constexpr_if=*/true);
>        cond = instantiate_non_dependent_expr (cond);
>        cond = cxx_constant_value (cond);
>      }
> -  else
> -    {
> -      maybe_warn_for_constant_evaluated (cond, /*constexpr_if=*/false);
> -      if (processing_template_decl)
> -	cond = orig_cond;
> -    }
> +  else if (processing_template_decl)
> +    cond = orig_cond;
>    finish_cond (&IF_COND (if_stmt), cond);
>    add_stmt (if_stmt);
>    THEN_CLAUSE (if_stmt) = push_stmt_list ();
> --- gcc/testsuite/g++.dg/cpp2a/is-constant-evaluated15.C.jj	2024-04-04 12:23:36.706962932 +0200
> +++ gcc/testsuite/g++.dg/cpp2a/is-constant-evaluated15.C	2024-04-04 12:22:29.915882859 +0200
> @@ -0,0 +1,28 @@
> +// PR c++/114580
> +// { dg-do compile { target c++17 } }
> +// { dg-options "-Wtautological-compare" }
> +
> +namespace std {
> +  constexpr inline bool
> +  is_constant_evaluated () noexcept
> +  {
> +#if __cpp_if_consteval >= 202106L
> +    if consteval { return true; } else { return false; }
> +#else
> +    return __builtin_is_constant_evaluated ();
> +#endif
> +  }
> +}
> +
> +template <typename T>
> +void foo ()
> +{
> +  if constexpr ((T) std::is_constant_evaluated ())	// { dg-warning "'std::is_constant_evaluated' always evaluates to true in 'if constexpr'" }
> +    ;							// { dg-bogus "'std::is_constant_evaluated' always evaluates to false in a non-'constexpr' function" }
> +}
> +
> +void
> +bar ()
> +{
> +  foo <bool> ();
> +}
> 
> 	Jakub
> 

Marek

Re: [PATCH] c++: Fix up maybe_warn_for_constant_evaluated calls [PR114580]

[Jason Merrill]

On 4/5/24 14:47, Marek Polacek wrote:
> On Fri, Apr 05, 2024 at 09:40:48AM +0200, Jakub Jelinek wrote:
>> Hi!
>>
>> When looking at maybe_warn_for_constant_evaluated for the trivial
>> infinite loops patch, I've noticed that it can emit weird diagnostics
>> for if constexpr in templates, first warn that std::is_constant_evaluted()
>> always evaluates to false (because the function template is not constexpr)
>> and then during instantiation warn that std::is_constant_evaluted()
>> always evaluates to true (because it is used in if constexpr condition).
>> Now, only the latter is actually true, even when the if constexpr
>> is in a non-constexpr function, it will still always evaluate to true.
>>
>> So, the following patch fixes it to call maybe_warn_for_constant_evaluated
>> always with IF_STMT_CONSTEXPR_P (if_stmt) as the second argument rather than
>> true if it is if constexpr with non-dependent condition etc.
>>
>> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
>>
>> 2024-04-05  Jakub Jelinek  <jakub@redhat.com>
>>
>> 	PR c++/114580
>> 	* semantics.cc (finish_if_stmt_cond): Call
>> 	maybe_warn_for_constant_evaluated with IF_STMT_CONSTEXPR_P (if_stmt)
>> 	as the second argument, rather than true/false depending on if
>> 	it is if constexpr with non-dependent constant expression with
>> 	bool type.
>>
>> 	* g++.dg/cpp2a/is-constant-evaluated15.C: New test.
>>
>> --- gcc/cp/semantics.cc.jj	2024-04-03 09:58:33.407772541 +0200
>> +++ gcc/cp/semantics.cc	2024-04-04 12:11:36.203886572 +0200
>> @@ -1126,6 +1126,9 @@ tree
>>   finish_if_stmt_cond (tree orig_cond, tree if_stmt)
>>   {
>>     tree cond = maybe_convert_cond (orig_cond);
>> +  maybe_warn_for_constant_evaluated (cond,
>> +				     /*constexpr_if=*/
>> +				     IF_STMT_CONSTEXPR_P (if_stmt));
> 
> I don't think we need the comment anymore since it's clear what the
> argument does, and then the whole call can fit on a single line.
> 
> But either way, the patch looks good, thanks.

Agreed, OK with that change.

Jason