[Bug c++/125555] [16/17 Regression] error: mangling of ... conflicts with a previous mangle [import std]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125555

--- Comment #8 from Patrick Palka <ppalka at gcc dot gnu.org> ---
Ah but presumably we want to treat instantiations of a non-template friends vs
template friends consistently, which the TI_TEMPLATE (ti) != entry->tmpl check
won't do.  For non-template friend instantiations the check will always be
false, and we'll never record any (full) specializations of non-template
(hidden) friends.

I think this would consistently set is_friend=false on full specializations of
template and non-template friends:

--- a/gcc/cp/module.cc                                                          
+++ b/gcc/cp/module.cc                                                          
@@ -15468,9 +15468,10 @@ depset::hash::add_specializations (bool decl_p)        
                }                                                               
              else if (is_friend)                                               
                {                                                               
-                 if (TI_TEMPLATE (ti) != entry->tmpl                           
-                     || !template_args_equal (TI_ARGS (ti), entry->tmpl))      
+                 if (TREE_CODE (spec) == TEMPLATE_DECL)                        
                    goto template_friend;                                       
+                 else                                                          
+                   is_friend = false;                                          
                }                                                               
            }                                                                   
          else                                                                  

but it regresses a modified a non-template friend version of friend-13.C:

diff --git a/gcc/testsuite/g++.dg/modules/friend-13.C
b/gcc/testsuite/g++.dg/modules/friend-13.C                                      
index 8930eca7ded5..7fe24bafb218 100644                                         
--- a/gcc/testsuite/g++.dg/modules/friend-13.C                                  
+++ b/gcc/testsuite/g++.dg/modules/friend-13.C                                  
@@ -4,7 +4,7 @@                                                                 

 module;                                                                        
 template <typename T> struct tuple {                                           
-  template <typename U> friend void f(tuple, U);                               
+  friend void f(tuple, int) { }                                                
 };                                                                             
 template <typename T> tuple<T> make_unique();                                  
 export module M;                                                               

gcc/testsuite/g++.dg/modules/friend-13.C:10:8: internal compiler error: in
import_entity_index, at cp/module.cc:4337


While perhaps not fully correct/consistent, the TI_TEMPLATE (ti) != entry->tmpl
check seems strictly better than the current buggy code which ignores all
friend specializations, so perhaps we should go with it for now?