[PATCH] c++, libstdc++, v3: Implement C++26 P2641R4 - Checking if a union alternative is active

[Jason Merrill]

On 9/1/26 4:34 AM, Jakub Jelinek wrote:
> On Thu, Aug 27, 2026 at 11:51:05AM -0700, Jason Merrill wrote:
>>> Shall we remove the hash_map record on deallocation instead or store
>>> something magic other than void_node?
>>
>> Removing it sounds elegant, if there isn't a diagnostic quality reason to go
>> the other way.
> 
> As I wrote yesterday, remove instead of values.put (t, void_node); when
> going out of storage caused too many diagnostic regressions, so this patch
> goes for two magic values instead of one.
> 
>>> So do you want instead 2 magic values, one for zero initialization
>>> implicit stuff and one for uninitialized implicit stuff?
>>> Or it can be a separate flag or flags from the val variable.
>>> Say bool is_zero_init = false, is_unitialized = false;
>>
>> Either sounds fine, but I lean toward 2 magic values.
> 
> Used 2 magic values, but as everything here is contained in the same
> function, named constants with those magic values and used comparisons
> against those constants.
> 
>>>>> +	if (TREE_CODE (val) != CONSTRUCTOR)
>>>>> +	  return boolean_false_node;
>>>>
>>>> This seems like an assertion rather than a return false?
>>>
>>> Wonder about error_mark_node or something similar, but perhaps I could make
>>> it into assert, yes (not done in the following patch yet).
> 
> Used gcc_assert now.
> 
>> It seems a bit confusing to have overloads with subtly different meanings
>> (modifying or not).  We might call the new one get_raw_value_ptr and not
>> have special handling of void_node, so it's simpler to distinguish between
>> out-of-lifetime and no object?
> 
> Added get_raw_value_ptr.
> 
>>> +	  else if (DECL_P (obj) && ctx->global->is_outside_lifetime (obj))
>>> +	    {
>>> +	      error_at (loc, "%qs on %qE outside its lifetime is not a "
>>> +			"constant expression",
>>> +			"__builtin_is_within_lifetime", obj);
>>> +	      inform (DECL_SOURCE_LOCATION (obj), "declared here");
>>> +	    }
>>
>> I'd expect this to return false once we resolve the distinction between end
>> of lifetime and end of storage (no object).
> 
> It doesn't, the patch now uses this for both
>    if (!valp || *valp == void_node)
> and for the !valp case ctx->global->is_outside_lifetime (obj) will be
> false, but for *valp == void_node it will be true.
> While the void_list_node case (used for CLOBBER_OBJECT_END rather than
> CLOBBER_STORAGE_END or going out of scope in other ways) doesn't make it
> to this and instead returns boolean_false_node (it shouldn't appear in
> DECL_INITIALIZER and so check_mutable doesn't need to be handled for that
> case either).
> 
> +  void destroy_value (tree t, bool past_storage_end = true)
>     {
>       if (TREE_CODE (t) == VAR_DECL
>   	|| TREE_CODE (t) == PARM_DECL
>   	|| TREE_CODE (t) == RESULT_DECL)
> -      values.put (t, void_node);
> +      values.put (t, past_storage_end ? void_node : void_list_node);

This seems backwards to me, since void_node is also used for 
out-of-lifetime subobjects (which are still within storage duration).

Jason