[libcxx-commits] [libcxx] [ASan][libc++] std::basic_string annotations (PR #72677)

[Louis Dionne via libcxx-commits]

================
@@ -892,41 +906,56 @@ public:
 #endif
       : __r_(__value_init_tag(), __a) {}
 
-  _LIBCPP_CONSTEXPR_SINCE_CXX20 basic_string(const basic_string& __str)
+  _LIBCPP_CONSTEXPR_SINCE_CXX20 _LIBCPP_STRING_INTERNAL_MEMORY_ACCESS basic_string(const basic_string& __str)
       : __r_(__default_init_tag(), __alloc_traits::select_on_container_copy_construction(__str.__alloc())) {
     if (!__str.__is_long())
+    {
       __r_.first() = __str.__r_.first();
+      __annotate_new(__get_short_size());
+    }
     else
       __init_copy_ctor_external(std::__to_address(__str.__get_long_pointer()), __str.__get_long_size());
   }
 
-  _LIBCPP_CONSTEXPR_SINCE_CXX20 basic_string(const basic_string& __str, const allocator_type& __a)
+  _LIBCPP_CONSTEXPR_SINCE_CXX20 _LIBCPP_STRING_INTERNAL_MEMORY_ACCESS basic_string(const basic_string& __str, const allocator_type& __a)
       : __r_(__default_init_tag(), __a) {
     if (!__str.__is_long())
+    {
       __r_.first() = __str.__r_.first();
+      __annotate_new(__get_short_size());
+    }
     else
       __init_copy_ctor_external(std::__to_address(__str.__get_long_pointer()), __str.__get_long_size());
   }
 
 #ifndef _LIBCPP_CXX03_LANG
-  _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 basic_string(basic_string&& __str)
+  _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20
+  basic_string(basic_string&& __str)
 #  if _LIBCPP_STD_VER <= 14
       _NOEXCEPT_(is_nothrow_move_constructible<allocator_type>::value)
 #  else
       _NOEXCEPT
 #  endif
-      : __r_(std::move(__str.__r_)) {
+      // Turning off ASan instrumentation for variable initialization with _LIBCPP_STRING_INTERNAL_MEMORY_ACCESS
+      // is inconsistent and that initialization may be annotated.
+      // Therefore, to copy __str memory, we have to unpoison it first (if object is poisoned and not external buffer,
+      // so only the short string case).
----------------
ldionne wrote:

```suggestion
      // Turning off ASan instrumentation for variable initialization with _LIBCPP_STRING_INTERNAL_MEMORY_ACCESS
      // does not work consistently for the initialization of __r_, so we instead unpoison __str's memory manually first.
      // __str's memory needs to be unpoisoned in the case where it's a short string.
```

https://github.com/llvm/llvm-project/pull/72677