[Bug tree-optimization/105329] [12/13 Regression] Bogus restrict warning when assigning 1-char string literal to std::string since r12-3347-g8af8abfbbace49e6

["rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105329

--- Comment #12 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #11)
> (In reply to Richard Biener from comment #10)
> > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104336#c1 contains a workaround
> > for libstdc++
> 
> Maybe
> 
>       _GLIBCXX20_CONSTEXPR
>       basic_string&
>       assign(const _CharT* __s)
>       {
>         __glibcxx_requires_string(__s);
>         return _M_replace(size_type(0), this->size(), __s,
>                           traits_type::length(__s));
>       }
> 
> isn't the most efficient way to assign a string (constant) looking at
> the complexity of _M_replace and its overlap checks (that we all expect
> to be optimized away but which are not).  In fact the _M_disjunct check
> isn't optimized because we cannot tell if the std::string storage
> overlaps the constant pool ...

Maybe

diff --git a/libstdc++-v3/include/bits/basic_string.h
b/libstdc++-v3/include/bits/basic_string.h
index c3fbc53953c..e83f3e8afa4 100644
--- a/libstdc++-v3/include/bits/basic_string.h
+++ b/libstdc++-v3/include/bits/basic_string.h
@@ -407,7 +407,8 @@ _GLIBCXX_BEGIN_NAMESPACE_CXX11
       bool
       _M_disjunct(const _CharT* __s) const _GLIBCXX_NOEXCEPT
       {
-       return (less<const _CharT*>()(__s, _M_data())
+       return (__builtin_constant_p (__s)
+               || less<const _CharT*>()(__s, _M_data())
                || less<const _CharT*>()(_M_data() + this->size(), __s));
       }

but oddly we folds the __builtin_constant_p to false early
but we treat __builtin_constant_p ("foo") as true.  That's inconsistent in
fold_builtin_constant_p:

  /* If this expression has side effects, show we don't know it to be a
     constant.  Likewise if it's a pointer or aggregate type since in
     those case we only want literals, since those are only optimized
     when generating RTL, not later.
     And finally, if we are compiling an initializer, not code, we
     need to return a definite result now; there's not going to be any
     more optimization done.  */
  if (TREE_SIDE_EFFECTS (arg)
      || AGGREGATE_TYPE_P (TREE_TYPE (arg))
      || POINTER_TYPE_P (TREE_TYPE (arg))
      || cfun == 0
      || folding_initializer
      || force_folding_builtin_constant_p)
    return integer_zero_node;

fixing that (removing POINTER_TYPE_P) makes the workaround work (and
the generated code simplifies greatly).